pg&e ac cycling ancillary service pilot - smarten · paging company via dial-up modem ......
TRANSCRIPT
PG&E AC Cycling Ancillary Service Pilot
Design & Results of Operations
Bashar KellowSr. Program Manager
Emerging DR Markets and TechnologiesApril 29, 2010
2
Overview
• Ancillary Services (AS) Market Overview
• Pilot Objectives
• Test Plan
• Design Description
• Data Analysis & Results:
- Load Impact
- Response Time/Signal Latency
- Customer Impact/Satisfaction
3
Participation in Spinning/Non-Spinning Reserve Market
• Historically, generators provided spinning reserve.
• For CAISO, a load can act as a resource in non-spinning market.
• Real-time telemetry is required for CAISO operations.
• Can Air Conditioning (AC) cycling programs participate in this market?
- AC load and load impacts must be
predictable and accommodate for
weather uncertainties.
- Telemetry requirement needs to be met.
- Time response requirements must be met.
4
Pilot Objective
• Analyze the time response of the AC resource.
• Establish a relationship between load reduction and environmental conditions.
• Design a telemetry, database management and monitoring system that provide a near real-time visibility of operations.
• Develop a load prediction model based on the gathered data.
• Study the feasibility of such a resource in the AS market.
NOTE - In this pilot:
- Did not bid this resource in the CAISO market.
- CAISO did not dispatch any load.
- Telemetry measurement was on the AC unit, not premise.
5
Approach to the Pilot
• Simulate a non-spinning reserve operation.
• Observe the load impacts for this operation.
• Execute the test under varying conditions: time of day, day of week and ambient temperatures.
• Measure and display feeder and AC load impacts in real time.
• Execute a post season customer survey to evaluate their experience and acceptance.
6
Vacaville
Antioch
Fresno 1
Fresno 2
Sample Design
• Four feeders
• 500 subscribers in each feeder (total of 2000).
• Control device distribution as follows:
- 400 DLC’s (Direct Load Control) per feeder
- 100 PCT’s (Programmable Communicating Thermostat) per feeder
• Feeder selection criteria:
- Mostly residential.
- Large population of Smart AC customers.
• Random distribution of measuring devices:
- 100 devices per feeder
• 30-40 telemeters per feeder
• 70 loggers per feeder
8
Testing Protocol
• Notch testing:
- 15 minutes shed.
- Loads interrupt immediately on receipt of signal.
- Completely off for 15 minutes.
- Random return of control over 2 minutes.
• Testing schedule:
- Two tests daily between 12:00 and 19:00.
- Weekdays only.
- Live event monitoring over 45 minutes.
- Operated within SmartAC program tariff.
- Executed 72 events over 2 months.
• Real-time Load measurement:
- Per feeder and cumulative for all feeders.
- Aggregated monitored AC load per minute.
• Extrapolated AC load per minute.
- Percentage of monitored units active.
- Baseline measurements.
10
Real-Time Telemetry Load Measurement Technology (Energy ICT)
• Instrumentation
- Dent PowerScout 3 (WIMeter)
- Two current transformers with voltage reference for power measurements
• Data logging and communication
- EICT WebRTU-Z2
- 500 ms modbus sampling
- kW, kWh & AMP per phase
- Store data at one minute intervals
- GPRS communications
- One minute data transmission frequency during event
- One hour data transmission frequency for non-event periods
• Central Software
- EICT EIServer MDM software running in ASP to house telemetry and SCADA data.
12
Load Impact Dependencies
• Significant load impact observed in real time on all feeders on certain days.
• Load impact vary by time of day, temperature and climate
- The higher the temperature the higher the impact.
- The later in the afternoon the higher the impact.
- The hotter the climate zone the higher the impact.
• Higher device response rate during events with higher number of active AC units.
13
“A Day with Significant Load Reduction”
• Over half the AC units were active “ON”.
• 80% response rate.
• Clearly observed impact on the feeder level (SCADA) and at the AC sample level.
• Response time within two (2) minutes.
• Cumulative 3.3 MW load reduction across the four feeders (Temp 99.5 Fº).
99.5 Fº
14
Load Impact - Temperature and Time Dependencies
• 35% active units.
• 0.7 kW average load per unit.
• 66% response rate.
94 Fº at 3pm
98 Fº at 6pm
• 55% active units.
• 1.5 kW average load per unit.
• 90% response rate.
15
Load Impact Performance Challenges
Several performance challenges were encountered during testing:
- Communication network coverage – Intermittent performance by control devices.
- Multiple events scheduled at the same time –Communication system congestion.
- Splintering some customers from the main SmartAC group –Incorrect group assignment for some devices.
16
• Aggregated data used to develop a forecasting model based on temperature.
• The regression model was applied to the 5 minute prior to the event and the 15 minutes of the event.
• The post-event period was excluded.
• The prediction model supports a 24 hour lead time with a 24 hour forecasting window.
Load Impact Predictability
>100 Fº
95-100 Fº
17
Load Impact Results
• AC load and load impact are weather dependant.
• Average AC load can range from zero to over 4 kW per unit.
• The relationship between AC load/load impact and temperature in non-linear:
- On 91 Fº days load impact is more than double the impact on an 86 Fº days .
- On 100 Fº days load impact is 3.5 times the impact on an 86 Fº days.
• In spite of all the variations due to dependencies, AC load and load impact is highly predictable.
• Generally, more load reduction is available under extreme summer weather events.
18
Signal Propagation
Yukon begins the
scheduled event
Yukon sends
event to comm.
queue
Establish
connection with
paging company
via dial-up
modem
Control message
is broadcasted to
control devices
Control device
receives the
message and
shuts AC
compressor
(resulting in a
load drop)
NOTE: This pilot did not measure the time delay from CAISO to PG&E.
20
Signal Latency Causes
• Signal traverses through multiple stages.
• Existing system utilizes low-speed internet connectivity (dial-up).
• This dial-up connection contributed to the majority of the signal latency.
• Message congestion on the paging network was the second major contributor.
Yukon begins the scheduled event
Yukon sends event to comm. queue
Establish connection with paging
company via dial-up modem
Control message is broadcasted to
control devices
Control device receives the message
and shuts AC compressor (resulting in
a load drop)
21
Signal Latency Results
• 60 seconds median response time.
• 92% response rate at 120 seconds.
• 69.4 seconds average time.
• 60.6 – 78.3 secondsaverage response time at 95% confidence interval.
0
20
40
60
80
100
120
140
0% 20% 40% 60% 80% 100%
Percentage of responding devices
Tim
e t
o A
C c
om
pre
ss
or
sh
uto
ff d
uri
ng
ev
en
ts (
se
co
nd
s)
AVERAGE
These values are well within the 10 minute time requirement defined by
CAISO.
22
• Post season survey was conducted.• 814 participants: 314 pilot customers & 500 general
SmartAC customers.• 72 pilot events were executed, one (1) general SmartAC
event was executed.• Random sample selection divided into strata based on
feeder and control device.• SmartAC sample was chosen from regions/cities near
pilot participants, not on the same feeder.• 55% Survey response rate (454),
180 were pilot participants.
Customer Impact/Satisfaction
23
Customer Survey Results
• No statistically significant differences between pilot and SmartAC participants’ satisfaction.
• No statistically significant differences to the number of noticed events between the two groups.
• Average thermostat set point for pilot group is 71.91 Fº vs. 73.76 Fº for SmartAC group.
• 17% of both groups reported that PG&E controlled their AC during the season.
• General SmartAC group reported on average 3.23 events –1 event.
• Pilot group reported on average 2.79 events – 72 events.
24
Conclusion
• Substantial load impact in some tests.
• Major dependency on ambient temperature and the time of day.
- On 80 Fº days, impact was about 0.2 kW.
- On 95 Fº days, impact was between 0.54 kW and 1.4 kW.
- On 100 Fº days, impact was between 0.62 kW and 1.44 kW.
• High predictability in the relationship between load impact and weather.
• Non-linear relationship between load impact and ambient temperature.
• Load impact ranged from almost zero (on 75 Fº days or less) to upwards of 3.3 MW (on 99 Fº days).
• On Average, load reduction occurred within 70.1 seconds.
• About half of the times, load reductions were present within 60 seconds.
• On 95% of the test occasions, load reductions was reached within 126 seconds.
• Virtually all of the load reduction obtainable from the pilot participants occurred within the first 3 minutes of operations.
• Generally, pilot participants were satisfied with the program.