how does relative humidity affect electricity demand? · 2016. 3. 12. · 1 . outline...

23
How Does Relative Humidity Affect Electricity Demand? Ying Chen, Tao Hong PhD 6/23/15 at Charlotte 1

Upload: others

Post on 06-Feb-2021

3 views

Category:

Documents


0 download

TRANSCRIPT

  • How Does Relative Humidity Affect Electricity Demand?

    Ying Chen, Tao Hong PhD

    6/23/15

    at Charlotte

    1

  • Outline

    • Introduction and motivation • Data • Methodology • Results • Conclusion

    2

  • Introduction

    Load-temperature scatter plot (Hong, 2010)

    Winter Summer

    3

  • Motivation

    • To find a benchmark model that including both temperature and relative humidity information for load forecasting.

    • Requirements of benchmarking process (Hong, 2010): • Simple • Creditable • Widely applicable • Interpretable • Reproducible

    4

  • • 2009-2011 hourly load data from North Carolina Electric Membership Corporation

    • 2009-2011 hourly weather data from 27 weather stations in North Carolina

    Data

    5

  • Time series of load (2009-2011) Time series of relative humidity (2009-2011)

    Data (cont.)

    6

  • Data (cont.)

    Load-relative humidity scatter plots for 12 months Load-relative humidity scatter plots for

    summer (6, 7, 8, and 9)

    7

  • Load-relative humidity scatter plots in summer by hour

    Data (cont.)

    8

  • Methodology

    Choose a benchmark model

    Build new models by adding relative humidity terms to the benchmark model

    Feed the data to all the models, and compare the Mean Absolute Percentage Errors (MAPEs) of validation data

    9

  • • Cross validation is used to calculate the MAPEs.

    • The average of MAPEs from the three cases is used for model comparison.

    Case number Training data Validation data

    Case 1 2009, 2010 2011

    Case 2 2009, 2011 2010

    Case 3 2010, 2011 2009

    Methodology (cont.)

    10

  • • Three benchmark models are used.

    • Trend: A linear trend variable. • Tt: Current hour temperature. • Tt-k: Temperature of the previous k-th hour. • Ta: Average temperature of the past 24 hours. • Month, Weekday, Hour are class variables (Hong, Pinson, & Fan, 2014).

    Main Effects Cross Effects

    B1

    Trend, Month, Weekday, Hour,

    Tt , Tt2, Tt

    3

    Tt*Month, Tt2*Month, Tt3*Month, Tt*Hour, Tt

    2*Hour, Tt3*Hour,

    Weekday*Hour

    B2

    B1+

    Ta, Ta2, Ta

    3

    B1+

    Ta*Month, Ta2*Month, Ta3*Month, Ta*Hour, Ta

    2*Hour, Ta3*Hour

    B3

    B2+

    Tt-1, Tt-12, Tt-1

    3

    B2+

    Tt-1*Month, Tt-12*Month, Tt-1

    3*Month,Tt-1*Hour, Tt-12*Hour, Tt-1

    3*Hour

    Methodology (cont.)

    11

  • • RH: Current hour relative humidity. • Summer(S): dummy variable. June, July, August and

    September is defined as summer. • RHS: RH*S. • RHS2: RH*RH*S.

    • Candidates to be added:

    Main Effects: RHS RHS2

    Cross effects: Tt*RHS Tt2*RHS Tt*RHS

    2 Tt2*RHS2 ...

    RHS*Hour RHS2*Hour

    Methodology (cont.)

    12

  • Results: B1

    B1 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B1M1 x 5.21%

    B1M2 x x 5.20%

    B1M3 x x x 5.08%

    B1M4 x x x x 5.03%

    B1M5 x x x x x 4.91%

    13

  • B2 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B2M1 x 4.10%

    B2M2 x x 4.10%

    B2M3 x x x 4.03%

    B2M4 x x x x 3.98%

    B2M5 x x x x x 3.96%

    B2M6 x x x x x x 3.96%

    B2M7 x x x x x x x 3.84%

    Results: B2

    14

  • B2 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B2M1 x 4.10%

    B2M2 x x 4.10%

    B2M3 x x x 4.03%

    B2M4 x x x x 3.98%

    B2M5 x x x x x 3.96%

    B2M6 x x x x x x 3.96%

    B2M7 x x x x x x x 3.84%

    B2M8 x x x x x x 3.84%

    Results: B2

    15

  • B2 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B2M1 x 4.10%

    B2M2 x x 4.10%

    B2M3 x x x 4.03%

    B2M4 x x x x 3.98%

    B2M5 x x x x x 3.96%

    B2M6 x x x x x x 3.96%

    B2M7 x x x x x x x 3.84%

    B2M8 x x x x x x 3.84%

    B2M9 x x x x x 3.85%

    Results: B2

    16

  • B3 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 Tt-1*RHS Tt-1

    2*RHS Tt-1*RHS

    2 Tt-1

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B3M1 x 3.87%

    B3M2 x x 3.87%

    B3M3 x x x 3.82%

    B3M4 x x x x 3.76%

    B3M5 x x x x x 3.74%

    B3M6 x x x x x x 3.74%

    B3M7 x x x x x x x 3.74%

    B3M8 x x x x x x x x 3.74%

    B3M9 x x x x x x x x x 3.68%

    Results: B3

    17

  • B3 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 Tt-1*RHS Tt-1

    2*RHS Tt-1*RHS

    2 Tt-1

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B3M1 x 3.87%

    B3M2 x x 3.87%

    B3M3 x x x 3.82%

    B3M4 x x x x 3.76%

    B3M5 x x x x x 3.74%

    B3M6 x x x x x x 3.74%

    B3M7 x x x x x x x 3.74%

    B3M8 x x x x x x x x 3.74%

    B3M9 x x x x x x x x x 3.68%

    B3M10 x x x x x x 3.68%

    Results: B3

    18

  • B3 RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2 Ta*RHS Ta

    2*RHS Ta*RHS

    2 Ta

    2*RHS2 Tt-1*RHS Tt-1

    2*RHS Tt-1*RHS

    2 Tt-1

    2*RHS2 RHS*Hour RHS2*Hour MAPE

    B3M1 x 3.87%

    B3M2 x x 3.87%

    B3M3 x x x 3.82%

    B3M4 x x x x 3.76%

    B3M5 x x x x x 3.74%

    B3M6 x x x x x x 3.74%

    B3M7 x x x x x x x 3.74%

    B3M8 x x x x x x x x 3.74%

    B3M9 x x x x x x x x x 3.68%

    B3M10 x x x x x x 3.68%

    B3M11 x x x x x 3.70%

    Results: B3

    19

  • Results: summary

    • The proposed benchmark model including both temperature and relative humidity effects are:

    + RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2

    RHS*Hour RHS2*Hour

    Effects to temperature benchmark models

    3%

    4%

    5%

    6%

    B1 B2 B3

    5.21%

    4.10% 3.87%

    4.91%

    3.85% 3.70% MA

    PE

    Benchmark model Plus RH

    20

  • Conclusion • Scatter plot shows there is relationship between load and relative

    humidity, especially during summer months (6-9), and the influence of relative humidity has hourly difference.

    • Adding relative humidity terms to three different benchmark models shows that they can reduce the models’ MAPEs around 0.2-0.3%.

    • The proposed benchmark model including both temperature and relative humidity effects are:

    + RHS RHS2

    Tt*RHS Tt

    2*RHS Tt*RHS

    2 Tt

    2*RHS2

    RHS*Hour RHS2*Hour

    effects to temperature benchmark models

    21

  • Q&A

    22

  • Reference

    • Hong, T. (2010). Short term electric load forecasting (Doctoral dissertation). Retrieved from ProQuest Dissertations and Theses. (Accession Order No. AAT 3442639).

    • Hong, T., Pinson, P., & Fan, S. (2014). Global Energy Forecasting Competition 2012. International Journal of Forecasting, 30(2), 357-363.

    23