benefits of chronological optimization in capacity planning for electricity markets
TRANSCRIPT
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
1/111
Capturing the benefits of high performance computing for investment
decisions in electricity markets: an emphasis on capacity expansion in
a carbon-constrained and uncertain future
Charles Ikenna Nweke (BEng)
Project submitted in partial fulfilment of the requirements for the degree of
MSc (Energy and Resources)
UCL School of Energy and Resources, Australia
I, Charles Ikenna Nweke confirm that the work presented in this report is my own.
Where information has been derived from other sources, I confirm that this has
b i di d i h
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
2/111
b i di t d i th t
Abstract
As worldwide demand for energy continues to grow, the need for continuous expansion of
global electricity supplythe most usable form of energy - cannot be overemphasized. The
drive to match demand with adequate supply has led to the restructuring of many
electricity regimes around the world. Also the increasing awareness of the threats posed by
greenhouse gas pollution is undoubtedly influencing operating and investment decisions in
different electricity frameworks. For these and other reasons, investment in renewable
energy technologies has resulted in major installations of wind farms across the world in
the last decade.
Meeting the demand growth in a sustainable way requires significant penetration of
renewable energy sources which could be intermittent in nature (e.g wind and solar). In
turn, capacity expansion planning (CEP) to accommodate the different technologies and
perhaps market frameworks requires a shift from orthodox methods used in the past.
Optimization models used in the past to aid investment decisions have always been
constrained by computing performance hence allowing certain level of detail.
The objective of this thesis was to assess the viability of a more thorough method of
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
3/111
being modelled. Nevertheless, this work exposes the frailties of the different methods that
energy planners and analysts need to be wary of knowing that huge commitments will rely
on the outcome of their models.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
4/111
Acknowledgement
This thesis is the outcome of countless hours of dedicated work guided by the trust,
encouragement, ideas, and inspiration of others near and far. I am particularly indebted to
my supervisor Dr. Mohan Kolhe for his input and guidance from the very beginning. Also I
am very grateful to the academic staff at UCL School of Energy and Resources, Australia for
their input in preparing me for this cross-disciplinary research I dared to embark on. Of
course all these wouldnt have been a smooth sail without the impact of the non-academic
staff at UCL SERAus, particularly Ms Maria Stavrinakis, whose unflinching support and
reassurances kept me going in difficult times. I am very grateful to the management of
Energy Exemplar pty and Mr Glenn Drayton (PhD) in particular for supporting this research
and providing the platform necessary to access and utilize the endless data required for
this work. Lastly I would like to acknowledge the backing and reinforcement from my familyand friends who in one way or the other have helped me all through this programme.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
5/111
Table of Contents
Abstract...ii
Acknowledgements.iii
List of
Figures. vi
List of
Tables.....vii
Acronyms &
Abbreviations.. viii
Definition of
Terms.x
1 INTRODUCTION ................................................................................................................. 1
1.1 Problem definition and motivation .................................................................................... 2
1.2 Objective and scope of the Thesis ...................................................................................... 3
1.3 Energy Exemplars Next-generation power market simulator ........................................... 4
1 4 O i d W k Fl f Th i 5
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
6/111
3 INVESTMENT DECISIONS IN ELECTRICITY MARKETS .......................................................... 34
3.1 Regulated Electricity Framework ..................................................................................... 35
3.2 Liberalized Electricity Industry ......................................................................................... 36
3.3 Implications of Capacity Imbalance .................................................................................. 39
3.4 Case Study: The Australian National Electricity Market (NEM) ....................................... 42
4 MODELLING LONG-TERM INVESTMENT DYNAMICS USING PLEXOS ................................. 45
4.1 PLEXOS LT Plan ................................................................................................................. 45
4.1.1 PLEXOS LT Plan formulation ............................................................................................. 46
4.2 Model Overview ............................................................................................................... 51
4.2.1 Bulls eye representation of model variables................................................................... 51
4.3 Model Input ...................................................................................................................... 52
4.4 Scenario Analysis .............................................................................................................. 59
4.5 General Model Hypotheses .............................................................................................. 59
5 SIMULATION AND RESULTS ............................................................................................. 61
5.1 Simulation Parameters and Conditions ............................................................................ 61
5.2 Base Case Simulations ...................................................................................................... 62
5.3 Sensitivity Analysis ........................................................................................................... 67
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
7/111
List of Figures
Figure 1-1: A 3-pointer high-level thesis work flow ............................................................................................... 6
Figure 2-1: Moore's law and the Intel chip evolution in the past four decades [15] ............................................. 9
Figure 2-2: A basic modern processor configuration [22] .................................................................................... 11
Figure 2-3: Capacity expansion planning optimization [37] ................................................................................. 16
Figure 2-4: Relationship between Benders' triplet and Decision triplet [47] ....................................................... 24
Figure 2-5: South Australian demand curve for March 2011 ............................................................................... 29
Figure 2-6: Load Duration Curve representation of Figure 2-5 ............................................................................ 29
Figure 2-7: A 12-block approximation of the LDC using least-squares method ................................................... 30
Figure 2-8: Chronological representation of the load profile using 50 blocks ..................................................... 31
Figure 2-9: 200-Block representation of the demand applied in this research.................................................... 32
Figure 3-1: Regulated electricity framework ........................................................................................................ 35
Figure 3-2: Participant stock and flow diagram in a liberalized electricity framework [56] ................................ 38
Figure 3-3: The Australian wholesale market reserve margin and peak prices after liberalization [3] ............... 40
Figure 3-4: The Spanish wholesale market after liberalization [3] ........................................................ ............... 40Figure 3-5: Optimal mix and level of capacity ...................................................................................................... 41
Figure 3-6: Right capacity with sub-optimal base-load ........................................................................................ 41
Figure 3-7: Excess capacity with optimal base-load ............................................................................................. 41
Figure 3-8: The Australian NEM showing the 5 inter-connected regions [76] ..................................................... 42
Figure 3-9: Shift towards renewable energy sources in the last decade [79] ...................................................... 44
Figure 4-1: Bulls eye representation of the SA model......................................................................................... 52
Fi 5 1 B d l i l ti h i th i t k l d t t l it d i 63
http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392243http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392243http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392243 -
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
8/111
List of Tables
Table 2-1: Comparisons of grid and cloud computing ......................................................................................... 14
Table 2-2: Formats for a typical minimization linear program ............................................................................. 17
Table 4-1: LT Plan key formulation elements ....................................................................................................... 47
Table 4-2: Some common problem variables used in LT Plan .............................................................................. 48
Table 4-3: Description of variables used to define inter-temporal constraints ................................................... 50
Table 4-4: PLEXOS input objects and properties .................................................................................................. 53
Table 5-1: Successful wind penetration levels attained by 2010 compared with simulation results .................. 66Table 5-2: Comparisons of computing demands for LDC and Chrono simulations .............................................. 74
Table 8-1: New entrants build Costs ($/kW) 2010-2030 [12] ............................................................................... xii
Table 8-2: Input parameters for new entrant technologies [12] ......................................................................... xiii
Table 8-3: New entrant gas prices 2010-2030 [85] .............................................................................................. xiv
Table 8-4: Hypothetical improvements in heat rates (GJ/MWh) for new entrant thermal plants [12] ............... xiv
Table 8-5: Fossil fuels and their GHG production rates from combustion [85] ................................................... xvi
Table 8-6: CO2price trajectory ($/tCO
2-e) for financial years ending in 2030 [80] .............................................. xvi
Table 8-7: Notional LRET targets for South Australia .......................................................................................... xvii
Table 8-8:South Australian Winter maximum demand projections [79] ........................................................... xviii
Table 8-9: Summer maximum demand projections [79] ................................................................................... xviii
Table 8-10: Energy projections between 2010 and 2020 [79] .......................................................................... xviii
Table 8-11: 2009-10 base year profile [79] .......................................................................................................... xix
Table 8-12: Demand-side participation (DSP) set at different RRN prices for South Australia [80] .................... xix
http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392277http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392277http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392279http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392279http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392279http://c/Users/Ikenna/Dropbox/Flashdrive/Thesis/Project%20docs/CN_UCL%20Thesis_final2012.docx%23_Toc327392277 -
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
9/111
Acronyms & Abbreviations
AEMO Australian Energy Market Operator
CCGT Combined Cycle Gas Turbine
CCS Carbon Capture and Storage
Chrono Chronological
CEP Capacity Expansion Planning
CF Capacity Factor
CMOS Complementary Metal-Oxide Semiconductor
CPT C02Price Trajectory
DP Dynamic Programming
DSP Demand-side Participation
EGS Enhanced Geothermal System
ESOO El t i it St t t f O t iti
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
10/111
MILP Mixed-integer Linear Programming
MLF Marginal Loss Factor
MSL Minimum Stable Level
NEM National Electricity Market (Australia)
NPV Net Present Value
NTNDP National Transmission Network Development Plan
OCGT Open Cycle Gas Turbine
OR Operations Research
POE Probability of Exceedence
RAM Random-Access Memory
RHS Right Hand Side
RRN Regional Reference Node
SA South Australia
SRMC Short Run Marginal Cost
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
11/111
Definition of Terms
Auxiliary Increment (Aux Incr): This describes the auxiliary loss per megawatt of generation
in a station. It is computed from the auxiliary use which denotes the total of all station
auxiliary loads or in-house energy consumption.
Cache: A relatively small high-speed memory (usually situated within the processor) that
improves computer performance.
Capacity Factor: This percentage measures the utilization of the generation resource over a
period of time. It is defined mathematically as Generation over a period/ (Total Capacity x
time).
Carbon Shadow Price: It is the notional price, which if implemented directly (through say, a
tax) would achieve the same abatement as that targeted by the mitigation measures
actually in place. It is defined as the social cost of emitting a marginal tonne of carbon or
the social benefit of abating a tonne.
Demand-side Participation (DSP): A situation where customers vary their electricity
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
12/111
Integer: Defined as a whole entity in its complete integral state
Integerization: It is a term used to describe enforcement of infrastructure
builds/retirements to remain as integer quantities in capacity expansion planning. It could
also be applied to unit commitment optimizations to ensure unit on/off decisions are
integer-enforced.
Interconnector: A line or group of transmission lines that links the transmission networks in
adjacent regions.
Intermittent: A term used commonly in this thesis to describe generating units which
outputs are not readily predictable. These include solar generators and wind turbine
generators.
Levelized Cost of Electricity: This is the lifetime discounted cost of a generating asset
expressed in cost per unit of power generated.
Marginal Loss Factor (MLF): This is a multiplier used to describe the marginal electrical
energy loss for electricity used or transmitted.
Mi i S bl L l (MSL) h l i (MW) hi h i i
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
13/111
Reliability of Supply: The likelihood of having adequate capacity (generation, DSP, or both)
to meet demand.
Run rate (up/down): This is the rate at which a generating unit ramps up/down generation
level from an offline condition to the MSL and vice-versa (MW/minute). This is defined for
the pre-conditions like hot, warm, and cold states; meaning that a generator may start-up
faster when in hot state (within 48
hours of shutdown).
Short-run Marginal Cost (SRMC): This is the change in total generation costs recorded per
unit change in the quantity of power produced by a plant.
Start-up costs: This accounts for the expenses (fuel, labour, emissions, etc.) involved in
bringing a unit from an offline condition to any output level above the MSL.
Unit Commitment (UC): This refers to a sequence of generator on/off decisions made over
time (usually in the short-term). It seeks an optimal and feasible combination of on/off
decisions for all generating units across a given horizon.
Unserved Energy (USE): This defines the amount of energy that cannot be supplied due to
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
14/111
1 INTRODUCTIONWorld energy demand has grown consistently over the past decades, yet one-fifth of the
worlds population are currently without access to electricity.1Global electricity demand
according to the International Energy Agency (IEA) is expected to increase by more than
200% of the 2007 level in less than half a century [1]. In similar terms energy demand and
CO2emissions could double by 2050 [1]; a path unarguably unsustainable if this demand is
to be met using conventional means alone. Addressing the global growth in electricity
demand while gradually moving towards a low-carbon economy has led to an increasing
reliance on renewable energy sources such as wind and solar energy [2]. For varying
reasons ranging from energy security to environmental concerns, many countries haveencouraged investment in renewable energy technologies which has resulted in major
installations of wind farms especially, across the United States, Europe, Australia and China
within the last decade.
Fundamentally, electricity markets over the world have experienced various forms of
i h l 20 di i l d lib li d f k
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
15/111
tools that will aid in achieving minimal total costs of investment while maintaining
adequate reliability hence avoiding consumer price hikes as much as possible.
Historically, accurate decision-support models for electricity planning have been
constrained by computing power [7, 8] among other issues. Even with the fairly predictable
properties of conventional sources of generation there had been various levels of
compromise between model detail and computing requirements [7]. Following the
increasing share of intermittent generation in various parts of the world, modelling
techniques used in capacity expansion planning (CEP) are being asked to cope with the
complexities of the different technologies that are being integrated into electricity grids.
New methods are expected to maximize computing resources within their physical
constraints to ensure the highest possible accuracies for various purposes.
1.1 Problem definition and motivationElectricity generation accounts for a third of total worldwide fossil fuel use and contributes
about 41% of global energy-related CO2emissions; hence efforts to significantly curb GHG
emission will no doubt be achieved through a transformation of the sector [1]. The overall
objective of minimizing total costs in the midst of uncertainties inherent in long-term
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
16/111
conventional thermal systems [7, 8]. This method reduces the resolution of system
variations such as the intermittency effects of distributed renewables on inter-temporal
impacts of outages and commissioning of thermal plants, leading to a loss of resolution
that may produce sub-optimal solutions to the original detailed problem.
Chronological modelling of the demand on other hand significantly improves the modelling
of intermittency effects on unit-commitment optimization of conventional plants [9] but at
a cost higher computational time and requirements. There are no doubts whatsoever
that the electricity mix is gradually changing (with preference given to cleaner but
unorthodox technologies like wind turbines), and at the same time there have been
significant improvements in computing performance over the years. More so historical
trends where single processor performance upgrade was paramount are tending towards
parallel multi-core systems [10]. It is in this light that this research looks into the increasingcomplexity of accurately modelling the growing percentage of intermittent generation by
effectively utilizing these advancements in computing. Retaining the chronological
information of detailed models in long-term planning is fast becoming indispensable if unit
commitment decisions are to be considered [6, 9] in finding the optimal future technology
mix in line with policies aimed at decarbonizing the world economy.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
17/111
unlocking possible benefits of recent improvements in computing performance and
infrastructures for broad purposes. This thesis focuses on answering important questions
like:
What share of unpredictable renewable injection in a market is permitted afterwhich more robust and apprehensive modelling technics will be needed in capacity
expansion planning?
How do the different levels of renewable integration affect marginal costs ofelectricity in a wholesale market with and without considering intertemporal
effects in the model; and how does it affect investment in other options?
What are the benefits of chronological modelling over the aggregated loadduration curve (LDC) approach given the present developments in computing
technology and what are the likely trade-offs assuming Moores theory holds for
the evolution of processing power in the near future?
Overall this research aims predominantly at retaining chronological information for
production and capacity constraints in a bid to show how decisions could be aided in the
long term which could potentially lead to huge savings following future investments paths.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
18/111
enabling the incorporation and conversion of real-time wind-speed forecasts into wind
generation forecasts are but some of the recent improvements in PLEXOS, allowing market
traders and operators more accurately anticipate swings of up to several thousand
megawatts in wind generation and the corresponding impact on market prices *13+.
The foundation of this research work with Energy Exemplar centred on seeking ways to
improve not only accuracy of decision support tools but also the credibility of such
imperfect but effective techniques. This research work employs the many benefits of the
constantly evolving PLEXOS platform including its use of stochastic optimisation of the
growing uncertainties helped by the efficient utilisation of Mixed integer linear
programming (MILP) techniques [14].With expert support from the Energy Exemplar team
this work effectively probes the status quo in long term planning and the orthodox
contribution of decision-support tools with respect to technology constraints. The questionof whether new methods are well overdue in the planning process makes the subject of
this report and will be investigated over the coming sections, employing high resolution
simulations on a developed South Australian test model.
1.4 Overview and Work Flow of Thesis
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
19/111
In Chapter 3, the two most distinctive electricity frameworks are compared and contrasted
in relation to their explicability using discussed mathematical methods. The impact of
capacity expansion outcomes are discussed in detail, citing arguments for present and
future generations, vis--vis investment capital and payback responsibilities. Finally, a case
study of the investment trends over the period of a decade in restructured markets is
analysed with particular emphasis on the Australian NEM and policies driving the
investment patterns observed.
Initiation Phase:
Execution Phase:
Drafting of thesis body
Update Model/database
Develop and simulateexperiments/scenarios
Closing Phase:
Analyse results ofsimulations
Draw conclusions andmakerecommendations
Finalize thesis report
End
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
20/111
In Chapter 5, results of the CEP simulations carried out on the described model are
compared and contrasted using LDC approximation against Chronological representation of
demand. Sensitivity analysis is performed on some key variables to show the model
appropriateness for this study as well as reflect impact of regulatory decisions on the
framework represented. Finding a compromise between model accurateness and
tractability is further discussed following analysis carried out for solving CEP problems in
chronological fashion. Chapter 6 outlines the conclusions of this research work, including
suggestions for prospective research and further investigation. The Appendix summarizes
important model data and parameters that make up the South Australian Model. Other
hypothetical data included in the model are outlined in the Appendix as well.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
21/111
2 APRACTICALPERSPECTIVEONHIGHPERFORMANCECOMPUTING(HPC)ANDOPERATIONSRESEARCH(OR)
Advancements in computing technologies particularly with the commensurate affordability
of processor components and wares have been enjoyed in virtually every work of life [15].
The past four decades has seen processor frequency grow arguably in par with Moores
predictions [15], however that steep growth seem to be stalling in recent times due to
thermal constraints with the Complementary metal-oxide semiconductor (CMOS)
technology [16]. Nonetheless the seemingly insatiable growth for more computing power
has meant that processor chip manufacturing companies like Intel and AMD are leaning
towards multi-core processing and network-distributed parallel computing, widely viewed
as cloud computing in recent times.
The application of HPC in power systems operation and planning requires significant
computing performance particularly with the need to consider more uncertainties and
variables from the liberalization of markets and the growing share of generation from
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
22/111
necessarily follow the increase in computing performance as figure 2-1 shows. Note a steep
growth between 1995 and 2005 preceded by a steady growth rate in processor speed from
1971.
A more recent study by Stutter [20] on the development pattern followed by chip
compositions (number of transistors per chip), clock speed of processors, power
requirements, and the parallelization of instructions between 1970 and 2010 showed
similar and very interesting results. Whereas the number of transistors continued to
increase in exponential fashion, the other characteristics in contrast have flattened out
right from the early 2000s. These plateaux explain the difficulty chip vendors have endured
in exploiting greater clock speeds viz: excessive heat generation, increasing power
consumption as well as issues with leakage of current [20, 21]. It also means that the
advent of multicore processing has the capability of enabling higher performancecomputing while keeping power requirements fairly constant. The challenge however is
that, applications will have to be redesigned if there are any chances of taking full
advantage of the parallelization benefits multicore processors bring to the fore [20].
This work does not discuss the validity or otherwise of Moores law notwithstanding that it
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
23/111
2.2 Single Processors to Parallel ComputingParallel computing and multicore computers are emerging as the preferred option needed
to deal with the computational workload from power system models [16] given the
physical constraints vendors have had to deal with in improving performance of single
processors. The irony however has been that parallel processing has not really lived up to
the expectations in terms of linear gains in performance over the single core likes. Say for
instance using a dual core chip does not necessarily provide twice the performance of asimilar single chip processor. This is partly due to coordination and handshaking between
the cores and the architecture of the applications running on the parallel processing chip
[20].
Over the past four decades performance throughput has been achieved by CPU designers
(mostly using single core processors) based on clock rate or speed, cache size, and
execution enhancement [20, 21]. In contrast, recent trends have shown an inclination
toward a multi-core or many-coresera which is being helped by improvements in cache
designs and hyper-threading abilities in chips and applications [20, 21].
2.3 Multicore Processing
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
24/111
multi-threaded applications to take advantage of the performance gains. Depending on the
compatibility of the hardware configuration for various purposes, well-written multi-
threaded applications were expected to offer performance boost of up to 40% in 2005 [20].
So far Microsoft and Apple operating systems have led the race among operating
systems compatible with multicore chip processors [22].
High performance growth in computing is not only reliant on drivers like multi-cores and
hyper-threading. Feeding the high-performing cores with the required amounts of data
necessary to prevent latency and bottlenecks requires a proportional expansion of on-chip
memory and/or faster access to data in memory subsystems *21+. The notion that Cache is
kingindeed summarizes the extreme importance of the on-chip cache size in performance
gains by greatly reducing processor latency during data execution and transfer [20]. Figure
2-2 shows a modern generic processor configuration showing the relationship between theon-chip Level 1 (L1) Cache and the off-chip Level 2 (L2) Cache in a core. The challenge has
always been with increasing the sizes of the memory blocks closer to the processor given
the relative difference in their magnitudes. For example, most computer systems have
memory configuration of the range: L1 ~ 32KB, L2 ~ 2MB, Memory (RAM) ~ 4GB and Hard
Disk ~ 500 GB [22].
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
25/111
dedicated caches others being shared by groups of cores or globally by all cores will have to
be done in such a way that does not counteract the expected gains in performances [21].
2.4 Grid ComputingJust like electrical grids where electric current is available for end-use in a pool injected
from centralized and distributed plants, so also does a computational grid provide
computing resources (particularly processing power and data storage systems) for carrying
out complex problems [23]. Grid computing is an unconventional form of HPC2 which
originated in the 1990s after a ground-breaking publication by Ian Foster and Carl
Kesselman [24]. This concept has since been embraced as a cost-effective option for
tackling CPU-intensive problems in organisations and also across geographical locations,
adopting what has become volunteer computing [24].
Grid computing has served to address two major needs: making IT assets remotely
available and cumulating computing power [23]. Accomplishing these needs has been
made possible through the use of applications capable of farming out bits of a program to
thousands of several heterogeneous computers, examples of which include the
SETI@home project [23, 34] and the Globus Toolkit [24, 25, 34]. Grids have been applied to
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
26/111
3% of the total emissions [34]. Hence it is pertinent that the insatiable longing for more
computing performance required to optimize operations and decision-making, in a carbon-
constrained world we find ourselves at the moment, be treated with concern in order not
to incur a rebound effect in the energy efficiency of the supposed solution. This I believe is
where distributed high performance computing (DHPC) has a huge role to play through
more patronage of grid and cloud computing.
2.5 Cloud ComputingThis is similar to grid computing but it goes a step further in its method of resource
provisioning [35]. Cloud computing offers in addition to grid computing, utility models for
processing capacity, business, and software functionalities [27] as value-added services by
vendors to customers. Among Stanoevska-Slabeva et als [34] review of cloud computing
includes a definition that cloud computing
is a large-scale distributed computing paradigm that is driven by economies of scale, in
which a pool of abstracted, virtualized, dynamically-scalable, managed computing power,
storage, platforms, and services are delivered on demand to external customers over the
Internet.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
27/111
future power systems [29] arguably will depend majorly on the architecture of modelling
applications to harness its infrastructural scalability; not diminishing its perceived promise
of reducing carbon footprints of the IT industry [35].
2.6 Grid versus Cloud ComputingQuestionably two of the most common forms of DHPC being used for computationally
complex and data-intensive problems, grid and cloud computing have a handful of similar
characteristics as well as peculiarities between them. The fact that many have argued3that
cloud computing evolved from grid computing means that they easily share similarities like
their ease of scalability. For both systems, CPU, network bandwidth and storage utilization
fluctuates with user demands and are scaled through load balancing of applications running
on separate systems [35]. Another obvious similarity of both DHPC systems are their multi-
tasking and multi-tenancy attributes which allows customers to perform concurrent single
or multiple applications at various instances [35]. Some differences between grid and cloud
computing are summarized in table 2.1 below.
Table 2-1: Comparisons of grid and cloud computing
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
28/111
2.7 Operation Research MethodsOperations research (OR) have applications in all works of human endeavours particularly
as it concerns decision making in economic, engineering, healthcare, military and in
transportation processes [31, 32]. Kaufmann and Faure [32] defines OR as the art of
applying precise reasoning to problems which, for a variety of reasons, cannot be
formulated in the usual precise terms of science. OR seeks to find a best or an optimal
solution out of a number of suitable practical solutions to any given problem [31]. Theroots of OR are dated as far back as during the world war II when scarce resources had to
be optimally distributed, and its application has since evolved following the increasing
intricacy and specialty of organisations from the industrial revolution up to this present
time [31].
The relevance of OR in energy planning can be traced to its numerous optimization tools,
particularly linear programming, which are now highly computer-based due to the
complexity of the tasks involved in real-life problems [31, 33]. Arguably one of the most
renowned science breakthroughs of the 20th
century, linear programming still dominates a
large share of all scientific computations handled by computers [7, 31] and can be argued
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
29/111
Subject to: Physical, technical, and environmental constraints, etc
(2.1b)
Such a problem would generally be tackled using optimization tools like linear
programming or dynamic programming. Depending on the nature, size and complexity of
the problem, further optimization techniques like decomposition methods [7, 36] may be
applied to make the problem computationally tractable and efficient, whereas stochastic
optimization techniques could be employed to enhance the correlation between the model
and its real-life problem through its peculiar modelling of probabilistic elements hence
increasing accuracy [7, 33,].
Capacity expansion planning (CEP) generally refers to determining the optimal location,
capacity, and timing of new generation and/or transmission builds and retirements.
Optimization models typically seek answers to the where, how much and when the
capitally-intensive resources need to be built and/or retired by minimizing expected
expansion costs subject to operational and technical constraints [37]. The total expansion
costs here involves investment costs as well as production costs of meeting short term
demands as shown in figure 2-3. The curves give an illustration of the opposite slopes
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
30/111
2.7.1.1 Linear Programming (LP)Linear programming comprises the planning of activities to achieve an optimal result,
among reasonable options, that most replicates a mathematical model [31]. Typically
aiming to maximize or minimize an objective function, LP utilizes some mathematical
assumptions like proportionality, additivity, divisibility, and certainty or determinism [31,
36] to obtain an optimal solution without violating any of the binding constraints. While the
linear formulation of most practical models satisfy majority of these assumptions, satisfying
the certainty assumption is hardly the case hence the wide application of sensitivity
analysis [31, 7].
A typical linear program is pre-formulated into either standard or canonical formats
depending on the nature of its variables and the constraints [36], after which an
appropriate mathematical algorithm solves the model. An example of these formats for a
minimization linear program from Bazaraa et al [36] is written as shown in table 2-2.
Table 2-2: Formats for a typical minimization linear program
Standard form Canonical form
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
31/111
and/or retirements, fuel costs, other fixed and variable maintenance costs, environmental
externalities, as well as social costs of unserved energy. The constraints would generally
include supply/demand adequacy, capacity reserve requirements (based on forced outage
and maintenance rates of the plants), emission constraints, and other system technical
restrictions like transmission line capacity.
The earliest applications of optimization in the 1950s for capacity planning were modelled
using linear programming to find least cost expansion solutions within constrained
operational boundaries [7, 14]. Recent improvements to the LP theory now permit
automatic recovery from infeasible problems whereby the supported solvers are able to
efficiently relax some constraints in order to repair the infeasibilities [41]. However one key
challenge that linear programming faces in solving CEP problems is the presence of integer
variables associated with investment decisions [37] or unit starts/stops decisions [43] forinstance, as well as non-linear constraints that define power flow equations [37]. In
summary some major shortcomings of linear programming with respect to CEP that make it
less attractive as a stand-alone method include the following [7]:
1. Its tendency to approximate or round up discrete elements like the capacity of
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
32/111
2.7.1.2 Mixed integer linear programming (MILP)This is an enhanced form of linear programming which is able to handle problems wheresome of the decision variables are restricted to integers in the optimal solution [37]. An
MILP generally expands the scope or size of an optimisation problem by the inclusion of
integrality constraints which otherwise would be treated by a linear relaxation (LR)
program. This ensures that some limitations of LP as described in (1), (2), and (5) above are
treated more accurately; of course at exponential computational costs [37, 7]. For instance
binary decision variables (whether to build or not to build, to start or shutdown plants, etc.)
makes CEP a combinatorial optimization problem whose solution space grows
exponentially with the variables [37].
To reduce the computational burden MILP-based models employ heuristics as well as the
branch and bound technique. The branch and bound algorithm works by adopting a
wishful approach that ignores the integrality constraints at the start and hopes that the LP
relaxationsolved would meet the necessary integer requirements [37, 38]. If the relaxation
does not meet the integer requirements the branch and bound method picks one of such
variables and branches, creating sub-problems with the integrality constraint of the
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
33/111
Unlike linear programming, dynamic programming starts by solving a sub-problem for its
optimal solution and gradually expands the problem by finding the following optimal
solution from the preceding one until the problem is solved completely [7, 31].
DP is characterized by the way its problems are handled. Hillier and Lieberman [31]
characterizes DP problems as follows:
They are divisible into stages where decisions are made for every stagesequentially;
Each stage is defined by a finite or infinite states representing the possibleconditions the system could take at such stages;
The policy decision at each stage serves to transform the current state to a stateassociated with the beginning of the next stage with the optimal solution of the
overall problem expected to be solved at the final stage. This way, optimal
solutions for the remaining stages are independent of decisions adopted in
preceding stages.
Possess a recursive relationship that identifies the decision for stage , given thedecision for stage , is available; in which case the solution proceeds in a
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
34/111
account for this, DP algorithm applies some form of heuristic rules to simplify problems by
way of eliminating some feasible states which may affect the accuracy of the decisions
reached [37]. As an example, STRATEGIST energy modelling software reduces the
computational burden on DP problems by specifying a limit to the feasible states that is
treated in single years of a capacity expansion problem [37].
2.7.2 Decomposition MethodsDecomposition, as the name implies, refers to the breaking down of large complex
problems into smaller problems for better tractability and more efficiency while preserving
the focus of the problem [7, 33, 36]. Real-life problems including CEP applications present
linear programs with millions of columns and rows which represent the variables that need
to be reduced into sub-problems of manageable sizes. Decomposition methods (e.g
Dantzig-Wolfe technique [36], Benders method [7, 39, 47], and Langrangian relaxation [36])
are but some common ways of handling such problems. Another method of breaking down
solution complexity include the branch and bound technique [37, 38] among others.
As one of the approaches for solving optimization problems, decomposition is not an
entirely new concept. Ku [7] portrayed decomposition methods as an alternative to linear
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
35/111
1962 (hence the name Benders decomposition) and reviewed by Geoffrion [48] a decade
later creating a method now known and applied across various industries as the
Generalized Benders decomposition [47]. Other forms of Benders decomposition have
been discussed based on their applications across different fields and include the stochastic
Benders decomposition [40], as well as the logic-based Benders decomposition and the
combinatorial Benders decomposition [39]. This thesis however emphasises the
generalized Benders decomposition method which has been key not only in power
systems decision problems but specifically in its algorithm similarity with CEP problem
formulations [47, 49]. Hence it offers a lot more computational efficiency (due to its ability
to exploit parallel computing); its generalized form makes it relevant to different types of
decision problems (such as in power systems operations, maintenance scheduling and in
planning); and its flexibility and scalability makes it ideal for integrating into existing
applications [47].
The generalized Benders decomposition algorithm decomposes large-scale mixed-integer
programming (MIP) problems into a master problem (which could take linear, non-linear,
integer or continuous forms) [47, 48] and a sub-problem (which may be a linear or non-
linear convex program) that serves to feedback linear constraints, known as Benders cuts to
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
36/111
; and is a lower bound of the initial problem (2.3a) and is updated iteratively by the 2ndstage problem.
To check if constraint (2.3c) is satisfied based on the decision , a slack vector isintroduced to test the feasibility of the sub-problem. If any violations occur in the sub-
problem, afeasibility cut5is added to the master problem and it is then re-solved.
2ndStage:
(2.5a) (2.5b)The sub-problem decides on a feasible considering the initial constraint (2.3c) with thegiven
from the 1st stage problem. If the solution is not optimal, an optimality cut6 is
added to the master problem and it is re-solved.
These iterations are repeated as new feasibility and optimality cuts are generated until the
process converges to an optimal solution in a finite number of iterations [39].
The generalized Benders decomposition is well amenable to power system decision
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
37/111
Generalized Benders approach has been proven to suit analyses involving thermal units,
limited energy and storage units, as well as non-dispatchable generators, and load
management services. Its application in generation planning using the Electric Generation
Expansion Analysis System (EGEAS) to adequately estimate incremental costs of meeting
allowed unserved energy reliability targets was acknowledged by the International Atomic
Energy Agency (IAEA) [8].
Figure 2-4: Relationship between Benders' triplet and Decision triplet [47]
The decision triplet is time-decomposedin an optimization problem as they can be treated
separately for different periods of the horizon. Also, each service in the decision triplet can
typically befunctionally-decomposed; hence leaving a lot of positives yet be leveraged with
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
38/111
One application of the two-stage SP in CEP is the scenario-wise decomposition. This form of
Stochastic optimization decomposes all possible paths (or scenarios) of a problem based on
discrete probability functions which they follow [43]. This assumes that the distributions of
the random parameters are represented by discrete finite scenarios* +, withprobabilities * +. The equation (2.7) is a standard representation of a two-stagescenario-wise SIP formulation [44].
Minimize:
(2.7)Subject to:
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
39/111
resultant capacity mix. Hence the problem becomes S times larger than a deterministic
case [43]; however parallelization of the optimization process by most solvers on multi-
cores ensures that such problems are not necessarily Stimes less tractable.
2.8 Dealing with Uncertainties and Imperfections inherent withDecision Support Tools
Decision making in practice is subject to different sources of uncertainty regardless of the
magnitude of the impacts of such decisions. In the same vein, models that try to mimic
reality are only simplifications of the many uncertainties that abound in real-life decision-
making. These uncertainties for instance stem from input data to a model (data
uncertainties), parameters used in simulating the model (parameter uncertainties), and the
approximations of reality that such a model represents (model uncertainties) [50]. Some
ways these uncertainties have been dealt with particularly for long-term decisions include
the use of scenario analysis and sensitivity analysis.
2.8.1 Scenario analysisThe Oxford English Dictionary defines a scenario as a sketch, outline, or description of an
imagined situation or sequence of events which could include a synopsis of future
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
40/111
scenario to reflect a possible future outlook, and then assessing the implications of each
scenario [7, 37]. An example of a scenario analysis using modelling tools is the
development and assessment of the Australian Large-scale renewable energy target (LRET)
in the NEM. A scenario could be constructed to evaluate the effect of meeting the overall
LRET target in 2020 through a fifty percent contribution from investments in Wind
generation in South Australia.
2.8.2
Sensitivity analysis
Unlike scenario analysis that tries to deal with uncertainties by evaluation of hypothetical
future occurrences, sensitivity analysis is more concerned with identifying the relative
effects of the various components of a decision support process. Saltelli et al in [51] defines
Sensitivity analysis as a study of how uncertainty in the output of a model (numerical or
otherwise) can be apportioned to different sources of uncertainty in the model input. It can
be carried out to screen input and output data for a model, the model parameters, as well
as the overall effectiveness of a model for a particular purpose. A method to perform
sensitivity analysis to deal with data uncertainties is explicitly described in [52]. Bertsch et
al [50] focuses on testing the sensitivity of parameters for multi-criteria decision analysis
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
41/111
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
42/111
Figure 2-5: South Australian demand curve for March 2011
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
43/111
figure 2-7. Increasing the number of blocks per LDC as well as the concentration of LDCs in
the planning horizon improves the approximation to the original load profile. For instance
modelling with one LDC per week obviously approximates the load better than with one
LDC per month (using the same number of blocks); in which case commissioning and de-
commissioning of plants can only be made weekly or monthly respectively.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
44/111
Chronological method generally requires more blocks than the LDC approach with
minimum resolutions of two blocks per day to ensure daily peak/off-peak cycles are
captured [41]. PLEXOS approximates the time-dependant load by fitting blocks into the
curves; the more blocks defined the more temporal characteristics are accounted for in the
planning horizon and the more computational resources and time demanded by the model
simulation.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
45/111
Figure 2-9: 200-Block representation of the demand applied in this research
Comparing the slicing of the demand curve in Figure 2-8 and 2-9, one can clearly see that
using more blocks is likely to increase the accuracy of the simulation results given that the
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
46/111
In this research the use of time-dependent demand profiling (chronological demand) over
aggregated load duration curves (LDC) is used to demonstrate how the present computing
systems is able to accommodate more complex and realistic models to aid better decisions.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
47/111
3 INVESTMENTDECISIONSINELECTRICITYMARKETS
Energy policies are being introduced by states and regions around the world for various
reasons including improved reliability and security of supply, environmental concerns, and
to exploit the economic benefits of competitiveness in energy markets. Electricity is
arguably the most useable form of energy, perhaps the reason why its market is very
closely linked with economic growth and social well-being of a people [1, 7, 53]. The past
four decades has recorded an average annual growth of 3.5% in electricity demand globally
[14]. The next four decades on the other hand anticipates that up to 20% of the total world
energy investments (USD 32 trillion dollars) will be committed to the electricity sector [1].
These investments when analysed individually follow market signals like prices andsupply/demand dynamics [54] as one would expect, but most importantly tend to be
influenced by the frameworks under which such markets operate such as in monopoly,
oligopoly or in liberalized markets.
The peculiar nature of investments in energy infrastructures, particularly in electricity
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
48/111
the monopoly and liberalized frameworks are discussed and reviewed. The different
mechanisms currently adopted in these markets and their effects on capacity expansion
and investment decisions are discussed focusing on the Australian (NEM) perspective.
3.1 Regulated Electricity FrameworkThe idea of a regulated framework is very common in a monopolistic system whereby a
vertically integrated utility controls generation, transmission and distribution of electricity
to customers in a region. For a monopolistic electric utility, the obligation to meet demand
at a certain reliability level implies a responsibility of investing in new infrastructure [59].
Figure 3-1 shows a simplified generic representation of an electricity monopoly structure
that illustrates the relationship between a single utility and the end-users or a major
market player with monopolistic influence in a market. In a vertically integrated regulated
monopoly the utility controls all business functions and is obligated to meet entire
customer demand in the region; there is no competition hence the need for government
regulation of the monopoly to prevent abuse of power [60]. A regulated framework could
also have some form of competition particularly at the generation level however
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
49/111
monopolies in practice fall, somewhere between the vertically integrated and the
unbundled monopoly described in [60].
The level of regulation in monopolized markets is seen to be critical in the scientific
representation of these markets as it reduces a great deal of uncertainties in capacity
expansion planning. With limited competition in the market, an optimization exercise
carried out for the monopolies (being regulated) should yield similar results with one
carried out by government regulators (for instance) seeking benefit maximization of allplayers in the market. This is supported by the fact that uncertainties like load demand
growth, discount and interest rates, and even future electricity prices are more harmonized
giving the planning process more of a deterministic outlook. In effect, monopolistic players
are less exposed to investment risk in exchange for lower returns while the customers bear
the consequences of bad investment decisions [59, 57]one way is through an imposition
of cross-subsidized prices on the customers12[57]. This guaranteed return for investors also
hinders technological innovation; a scenario which is unlikely to happen in the liberalized
markets where competition breeds incentives for breakthroughs [59]. The non-existence of
a market means that prices do not accurately reflect supply adequacy. In addition
investments in transmission tends towards eliminating congestion entirely which may not
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
50/111
electricity for customers as well as freedom of negotiation with suppliers of choice, and a
better reflection of demand/supply relationships through the adoption of spot markets.
Whether all of these objectives have been achieved so far in the liberalized frameworks
mentioned above remains to be seen; however distortions from regulations such as the
implementation of price ceilings among others are likely hindrances to these objectives
[72]. Due to the complexity of electricity networks, horizontal unbundling of supply utilities
have more or less been carried out on the generation and retail levels. Ownership and
control of transmission and distribution infrastructure in liberalized frameworks is mainly
that of the regulatory authority or market operator who have as important a role as the
generators in the overall sustainability of unbundled markets.
Unlike in the monopolized framework, decisions to invest and retire capacity are made in a
decentralized profit-oriented move by the competitors in the industry. Market participants
are forced to draw conclusions from price signals and face numerous uncertainties which
mean imperfect foresight of future market conditions [3, 58]. This framework incurs higher
risk profiles to planners who not only have to tend with uncertainties in exogenous
variables like fuel cost and demand but also with short-term and strategic behaviours of
competitors and the regulatory authorities [67]. Future revenue streams are not
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
51/111
reliability levels by the operators [64, 67]. Ironically, market dynamics and economic theory
which are the basis upon which the liberalized frameworks are proposed appear imperfect
due to inherent characteristics of energy infrastructures namely: its investment lumpiness
and the non-linear benefits of economies of scale [58].
Figure 3-2: Participant stock and flow diagram in a liberalized electricity framework [56]
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
52/111
3.3 Implications of Capacity ImbalanceThe objective of an optimal expansion plan is to expand capacity to a (equilibrium) pointsuch that marginal cost exceeds the benefit of retaining the status quo [58]. There have
been a number of literatures comparing the electricity frameworks and their ways of
dealing with demand uncertainty as well as their implications on the reserve margins of
various electricity industries around the world. Ku [7] and Haynes et al [69] discussed the
pros and cons of under- and overcapacity from a CEP perspective; however, Green [68]
discusses the implications of investment in the different capacity mixes (i.e. baseload,
intermediate and peak generation) on market prices. Kilanc and Or [54] have argued based
on experiments carried out, that investments in power markets do not always go in tandem
with demand growth and capacity retirements due to imperfect foresight as well as
investment decision and construction delays among others. Here I synopsize the effects of
capacity inadequacy and imbalance that need to be considered while modelling long term
investment decisions for electricity markets.
Capacity over-design or overcapacity has enjoyed the better share of arguments over time
and have actually occurred more in practice [58, 69]. However the majority notion that
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
53/111
Figure 3-3: The Australian wholesale market reserve margin and peak prices after liberalization [3]
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
54/111
competitive market with respect to the demand profile in a region. Figures 3-5, 3-6, and 3-7
summarize the results of the capacity model discussed in [69].
Figure 3-5: Optimal mix and level of capacity
Total Revenue
Total Cost
Total revenue curve is higher than the
total cost beyond period T* meaning
the base-load generators will operate
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
55/111
3.4 Case Study: The Australian National Electricity Market (NEM)The Australian NEM began operations in 1998 after restructuring of the previousframework leading to a competitive wholesale electricity market. It boasts of the worlds
longest interconnected power system with a distance spanning about 5,000 kilometres,
connecting five regions viz: New South Wales, South Australia, Queensland, Victoria, and
Tasmania. The NEM records electricity transactions of more than $10 billion annually in
meeting demands of over eight million users in the Eastern part of Australia. The Australian
Energy Market Operator (AEMO) is charged with operating the NEM as well as coordinating
the planning of infrastructure for the efficiency, reliability, and security of the system based
on the National Electricity Law and Rules [76].
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
56/111
Marginal pricing is reflected in the incremental increase/decrease in generation or demand
in a spot market where AEMO dispatches generators18 to meet demand at 5-minute
intervals. The (marginal) dispatch prices are averaged every half-hour to determine the
spot prices for each trading interval upon which generators are paid for their supply in each
region. The spot prices though dictated by supply and demand balance (also considering
system and transmission constraints) are capped, as prescribed by the National Electricity
Rules (NER); presently at a maximum of $12,500/MWh and a minimum of negative
$1000/MWh. By way of considering system reliability and transmission constraints into
pricing, the NEM adopts an approximate form of locational pricing by the use of marginal
loss factors (MLF)19calculated at every node based on the generation to demand ratio at
the nodes20
. These locational signals inform market participants about energy variation
within the nodal locations hence guiding the operation of existing generation while
affording new entrants the opportunity of making more informed decisions regarding loss
reduction in the choice of location and technology of plants [77]; this serves the overall
efficiency of the market in a decentralised decision-making pool of participants.
While investment decisions can be seen as decentralized strategic moves by the competing
market participants in the NEM, AEMO retains planning and coordination of the market to
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
57/111
Figure 3-9: Shift towards renewable energy sources in the last decade [79]
AEMO plays a vital role in consolidating the supply adequacy, network planning and future
investment opportunities through its suite of planning documents like the Electricity
Statement of Opportunities (ESOO) [79], National Transmission Network Development Plan
(NTNDP) [80], and a host other documents prepared for each regions like the South
Australian Supply and Demand Outlook (SASDO) [81]. With carbon pricing on the verge of
coming online in Australia from the next financial year [82], and its expected push on
renewable energy investment, more rigorous centralized planning is expected to inform
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
58/111
4 MODELLINGLONG-TERMINVESTMENTDYNAMICSUSINGPLEXOS
Electricity markets exhibit some structural patterns and dynamics that are easily captured
by mathematical models which aid in testing the likely impacts of policy outcomes in a
system. PLEXOS LT plan offers a very flexible and rigorous environment for the formulation
and optimization of Mixed-Integer Linear Programs (MILP or MIP) that effectively mirror
the dynamics in a whole range of market frameworks. Here an electricity model21 is
developed using available data and some hypothesized data that closely mirror the South
Australian wholesale market and would be used to provide answers to questions posed in
the objectives section of this thesis; this is of course in response to the long-term dynamics
of the liberalized framework practiced in the NEM. Policies constraining carbon emissions
in the long term here in Australia are well accounted for in the model including effects of a
likely carbon pricing commencing in 2012.
This chapter discusses the PLEXOS LT Plan components which includes variables,
parameters, and the problem formulation. The model is described in detail, endogenous
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
59/111
only carried out in deterministic fashion as the PLEXOS LT Plan can stochastically find the
single optimal solution in the face of uncertainties in any input like wind generation, load,
fuel prices, etc.
In modelling long-term capacity expansion for electricity markets PLEXOS captures a variety
of possibilities through its many features including expansion and retirement of
infrastructure (such as generators, AC/DC transmission lines, interfaces) in multi-stages,
physical generation and load contracts from participants directly to customers, as well asmodelling mutual exclusivity among projects. A whole range of generating options can be
modelled from conventional thermal plants (with or without CCS) to hydro (run-of-river or
with storages) down to little details like run up rates and ramp rates (depending on cooling
states of generators).
4.1.1 PLEXOS LT Plan formulationThe LT Plan formulation uses Mixed-integer programming for solving capacity expansion
problems using a set of user-defined elements and the necessary problem variables as
shown in table 4-1 and 4-2 respectively. The objective function minimizes the net present
value of build/retire costs, fixed operations and maintenance costs, and the expected
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
60/111
Subject to:
Energy Balance (4.1):
Feasible Energy Dispatch Accounting for Maintenance and Forced Outage Rates (4.2):
( )
Feasible Builds/Retirements (4.3):
Integrality (4.4):
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
61/111
Element Description Unit
Duration of dispatch period Hours Overnight build cost of generator $/KW Maximum number of units of generator allowed to
be built by the end of year
Maximum generating capacity of each unit of
generator MW
Number of installed generating units of generator Value of lost load (energy shortage price) $/MWh
Short-run marginal cost of generator
which is
composed of [Heat Rate] [Fuel Price] + [VO&M
Charge]
$/MWh
Fixed operations and maintenance charge ofgenerator $/KW/year
Average power demand in dispatch period MWSystem peak power demand in year MW
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
62/111
Variable Description
Unserved energy in dispatch period Capacity shortfall in year Excess capacity in year
Constraints (4.1), (4.2), and (4.3) define minimum constraints that typically distinguish
capacity expansion planning from mid- and short-term planning exercises. The integrality
constraint in (4.4) makes the problem more realistic in representing the lumpiness of the
investments, mothballing, or retirement decisions; however it also increases the complexity
of the problem, although with MIP-compliant solvers such problems are tractable. The LT
Plan formulation by default does not require reserve margin constraints in (4.5) (even
though it will be involved in one of the scenarios developed in the model for this thesis)
which means that a trade-off between shortage costs24and the economics of expansion
determines whether infrastructure is built or not; and the resulting reserve margin
may/may not meet reliability standards.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
63/111
GenTurnOff (t):
GenOffk>= GenMDTt for k=tto t+MDT, and all t
Start Cost and Shutdown Cost (4.9):
Objective (Minimize) (4.9.1):
Table 4-3: Description of variables used to define inter-temporal constraints
Variable Description
This is the maximum output capacity of the unit in period This decision variable represents the number of units operating in each
dispatch period,
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
64/111
[ (
)] (4.0.1)
4.2 Model OverviewFor this thesis, an optimization-based model incorporating all the existing thermal and wind
generators in South Australia are modelled. The SA region is treated as a stand-alone
market but with inter-regional transfers between SA and Victoria modelled as a wholesale
market object. Owing to this arrangement electricity is traded between SA and Victoria
(within seasonal interconnector transfer limits) in response to correlations between
endogenously determined prices in SA and exogenous historical Victorian averages defined
as part of the model input. This model serves the purpose of benefit maximization for all
market participants from a market operator point of view assuming a perfectly competitive
framework.
The model excludes transmission limits and congestion modelling in its optimization of
production costs within the SA region however locational losses incurred by all individual
generators are represented by the marginal loss factors as defined in [84] for the
2011/2012 financial year. Model parameters and Input data and based on industry-wide
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
65/111
approximated impact of omitting transmission parameters in this model is cushioned in
build outcomes based on capacity addition limits of certain generation technologies in
different locations; just as the use of MLFs ensures that congestion isnt entirely ignored in
short-term dispatch decisions.
The SA region is modelled as an aggregated case, whereby all generators and the regional
load are connected to a node where the pool price is decided. The settlement method isnt
affected however, as outputs at generator terminals are balanced against the auxiliary
station demands and the marginal losses for the different generators. Correlating marginal
costs of generating in the SA region with hourly prices defined in the Victoria region 26
determines if power is purchased (imported) or sold (exported) depending on the seasonal
interconnector flow limits defined in the model.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
66/111
properties. These properties can be made static if fixed for the whole simulation horizon or
dynamic if more flexibility is to be introduced, particularly in temporal fashion. The
complete list of parameters and input data for every generic object used in this model is
very extensive hence provided in the Appendix. Table 4-4 highlights the generic objects
used in building this specific model as well as brief definitions of their properties and
source of information (where applicable). Intended for use with a long term planning
horizon, the input data are made as dynamic as possible to reflect not only periodic
variations but also season fluctuations and yearly trends for up to twenty years. Data which
reflect such dynamism include generator ratings; build costs, emission prices, wind and
solar profile, Victorian market prices, and fuel prices. Specific times are captured even
further with time slices defined to model peculiar outcomes on Weekdays, Weekends,
Peak, Off-peak, Summer, and Winter, as well as in combination like in Summer Off-peak for
instance.
Table 4-4: PLEXOS input objects and properties
Property Description Source of data
Generator Objects
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
67/111
Property Description Source of data
Max Ramp Up Maximum ramp up rate that applies between the
Min stable level and the rating
AEMO
Run Up Rate Ramp rate that applies while running the unit from
zero to Min stable level.
Hypothetical
Aux Incr Auxiliary energy consumed per unit of generation.
Also important for modelling the high auxiliary
demands of capturing and storing carbon.
ACIL Tasman
[85]
Marginal Loss
Factor
MLFs defined for each generator in relation to
locational supply/demand profiles
AEMO
FO$M Charge Annual fixed operation and maintenance charge Energy Exemplar
Firm Capacity Contribution of each generator to capacity reserves Energy Exemplar
Forced Outage A hypothetical probability of failure for each unit Energy Exemplar
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
68/111
Property Description Source of data
Max Units Built Maximum number of units permitted to be
constructed in aggregate over the planning horizon
reflective of transmission congestion in different
zones
AEMO
Max Units
Retired
Maximum number of units permitted to be retired
from an existing station over the horizon
AEMO
Fuel Objects
Price Price of Fuel ACIL Tasman
[85]
Transport
charge
Additional charge on fuel due to transport
requirements
AEMO
Emission Objects
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
69/111
Property Description Source of data
DSP Bid
Quantity
Bid quantity for demand-side participation AEMO
DSP Bid Price Bid price for demand-side participation AEMO
VoLL Value of lost load used to model the spot price
ceiling
AEMO
Price of Dump
Energy
Price of dump energy per MWh used to mirror the
spot price floor
AEMO
Generator
Settlement
Model
Determines price paid to generators for electricity
supplied
Parameters to
mirror NEM
Load
Settlement
Determines price paid by customers Parameters that
reflect the NEM
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
70/111
Property Description Source of data
Market Object
Price Price of imported/exported energy per MWh based
on historic prices in Victoria
AEMO
Price Scalar Scalar on Market price Hypothetical
Price Incr Increment to dynamic market prices Hypothetical
Max Sales Maximum sales or exports out of the SA region
subject to seasonal line limits
AEMO
Max Purchases Maximum purchases or imports from outside the
SA region
AEMO
Constraint Objects
RHS Used to represent a value that binds the constraint Specific to
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
71/111
Minimum Inertial Constraint (4.6):
Annual minimum contribution of renewable generation based on LRET (4.7):
Where:
is the left-hand side generation coefficient for each of the plants represent the dispatch from base-load and intermediate thermal units (in this
model it comprises coal plants, CCGT, and geothermal units).
represents a penalty price for violating that constraint. Where it is notincluded it means the constraint cannot be violated as in (4.7).
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
72/111
4.4 Scenario AnalysisPLEXOS tool provides a flexible user interface which ensures parametric study is easilycarried out in its models to aid understanding of market dynamics, the effects of
participant strategies and system-wide policies. Focusing on the South Australian region in
the NEM, a limited number of scenarios were developed to demonstrate the sensitivity of
this model and tool, including their potential use for more practical purposes.
The scenarios were constructed to characterize certain changes on capacity expansion in
SA including: long-term demand growth patterns using different probabilities of
exceedence (POE); differing demand-side participation and related economics; carbon price
levels; system reliability dynamics represented by constraints; regional imports and
exports; the LRET targets, capacity obligations of generators and reserve margins;
Technological advancement and learning curves; not omitting annual and aggregate build
limits on certain technologies due to project exclusivity or locational congestions. The
effects of some of these scenarios are shown in the sensitivity analysis discussed in a later
section.
4.5 General Model Hypotheses
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
73/111
and at no extra costs28. Priority isnt given to any projects due to congestion matrices or
proximity to transmission and mutual exclusivity of certain projects are equally not
considered. The absence of intra-regional transmission network means that losses are only
factored in generator settlements using MLFs defined for every financial year by AEMO and
this study assumes the same MLFs are maintained for the entire planning horizon. Inter-
regional transfers between SA and Victoria are modelled in response to an input price data
file dynamically scaled to reflect price increase under the carbon pricing scenarios.
The ratings for existing thermal plants are assumed to be constant for the first ten years
following data drawn from AEMOs forecastseasonal ratings for the plants in SA. However,
the thermal plants are assumed to gradually deteriorate in the following decade. For new
units however, the efficiencies are fixed for the planning span. Technological learning
curves are believed to be adequately incorporated into the model using more efficient heat
rates for new entrants (used in one of the scenarios) and in the annual build cost curves
defined for every technology. Therefore taxes and depreciation were assumed to be
factored into the build costs and annual de-rating of plants for this model.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
74/111
5 SIMULATIONANDRESULTS
In this chapter simulations are carried out on the model presented in the previous chapter
baring a few assumptions as discussed. Having been developed to show a simplified model
of the South Australian electricity framework, simulation results are analysed in terms of
the fundamental causes of the simulated results and relative behaviour rather than exact
figures reported. Most of the model outcomes have been observed in markets such as the
Australian NEM as described at the end of chapter three.
Simulation results are used to illustrate the developments thus far in HPC and my
arguments biased towards proportionate improvements in decision-making pertaining to
investments in electricity markets. This is demonstrated by the use of the relatively less
tractable chronological modelling of load demand in capacity planning compared to
orthodox methods. The long-run response of power markets to policy changes is further
exhibited showing sensitivity analysis of key variables using some developed scenarios
involving reserve margins, carbon pricing, renewable targets, and so on.
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
75/111
defined as input data for different projects. This discount rate also applies perpetually
after the end of the horizon as the optimization considers on-going existence of the market
after the planning horizon ends.
Unit commitment optimization is based on linear relaxation rather than the optimal integer
production decisions to reduce the MIP problem formulation. Stochastic samples of
endogenous variables like wind profile, generator forced outages, and maintenance
outages are modelled in every simulation using Monte Carlo sampling method with the
stochastic results fixed to aid accurate comparison of the Chronological and LDC methods.
The simulation was carried out in PLEXOS 6.202 R11 environment using Xpress-MP solvers.
5.2 Base Case SimulationsThe base case includes all existing scheduled and semi-scheduled generating plants using
ratings reported by AEMO as of July 2011 and includes new entrants expected to be
commissioned by the end of 2011. Committed and proposed investments beyond 2011 are
ignored and subject to economic optimization by the algorithms. Carbon pricing is defined
to come into stream at the beginning of the 2012-13 financial year and retirement of fossil
plants is subject to fixed costs, dispatch economics, reliability levels and other
-
8/13/2019 Benefits of Chronological Optimization in Capacity Planning for Electricity Markets
76/111
Close observation reveals less capacity in figure (5-1b) ending at about 8000 MW compared
to 9000 MW in (5-1a) despite both reserve margins tailing out at 5% at the end of the
planning horizon related to the defined minimum level in the model.