Assessing climate sensitivity of hourly electricity demand in Japan - PowerPoint PPT Presentation

The 25th International Workshop Assessing climate sensitivity of hourly electricity demand in Japan Yuki HIRUTA, Lu GAO and Shuichi ASHINA November 18-19, 2019 National Institute for Environmental Studies Center for Social and Environmental Systems Research Yuki Hiruta Research Associate

■Motivations International Workshop Global warming • Highly fluctuating electricity demand • Decrease the efficiency of electricity supply system • Increase the fossil fuel consumption • Warming urban environment in summer • Multiple problems such as heat strokes Urbanization It is important to clarify the relationship between weather conditions and the hourly electricity demand

■Overview Target Areas Jurisdiction of 10 electric power companies in Japan • (EPC) Method over view Build regression models for each EPC Predictor Explained variable Hourly Multiple variables weather conditions Related to electricity consumption • daily cycle of human activities • Simulation Represent the Temperature response functions Based on simulation by the constructed models Understanding the relationship between weather conditions and the hourly electricity demand

■Temperature response function (TRFs) *Also called Energy signature U-shaped V-shaped Electricity consumption Electricity consumption BPT: Base Demand Base Demand Balance point temperature Comfort Zone BPT(Balanced point temperature) Temperature( � ) Temperature( � ) Application: Important! TRFs → Basic Units Affects assumptions of • BPT → Reference temperature for HDD/CDD other models! • BPT , Slope → Electricity demand projection 4

■Data and Variables Data Sauce Names （Abbreviations） Units Data Description Organization for Cross-regional Explained Hourly electricity demand in each EPC Historical electricity demand （ EC ） Coordination of MWh Variable jurisdictional area. Transmission Operators Japan, Hourly averaged air temperature in FY2016 Temperature （TEMP） ℃ and FY2017 Predictors Hourly averaged relative humidity in FY2016 Humidity （ HUM ） % and FY2017 Hourly averaged total radiation in FY2016 Solar radiation （SUN） MJ/m � Japan Historical and FY2017 Meteorological weather Average wind speed in ten minutes before Agency Wind speed （WIND） m/s each hour in FY2016 and FY2017 Hourly total rainfall amount in FY2016 and Rainfall amount （RAIN） mm FY2017 Hourly total snow depth in FY2016 and Snow depth （SNOW） cm FY2017 Discomfort Index. Derived from Discomfort Index （DI） ­ Thermal Historical DI � 0.81 � TEMP � � 0.01 � HUM � � 0.99 � TEMP � � 14.3 � � 46.3 Weather Data index Wind Chill Index. Wind Chill （WCI） ­ Above 10.45 � 10 WIND � . � � WIND WCI � 33 � TEMP Weekends, holidays, the New Year, and the Calendars in Holidays and weekends dummy FY2016 and Obon Festival were set as 1, and all other ­ （HDD） FY2017 days were set as 0 in FY2016 and FY2017 Human Working people The percentage of people who are working at % activity （WORK%） the hour. NHK Broadcasting Awake people The percentage of people awake in their Culture Research % （WAKE%） homes at the hour. Institute. Sleeping people The percentage of people who are sleeping at % （SLEEP%） the hour. 5

■Human activity data % Human activity in each hour 100 90 80 70 60 50 40 30 20 10 0 Hour 1 6 12 18 24 (Data source: NHK Broadcasting Culture Research Institute, 2015) 6

■Model construction MARS ： multivariate adaptive regression splines (Friedman, 1991) Captures the complexity of the potential model by applying a locally • linear models. Selects important variables during the model building process. • Showed excellent prediction performance in short-term power • consumption modeling (Sigauke & Chikobvu, 2010; Al-Musaylh, Deo, Adamowski, & Li, 2018) Intuitive understanding for MARS performance • Can model nonlinearities in Flexible and Highly generalizable high-dimensional data: flexible modeling is expected • Highly generalizable 7

■Constructed models Performed well in all power company models. 0.870(Kyushu ) 〜 0.953(Okinawa) Hokkaido Tohoku Tokyo Chubu Hokuriku Kansai Chugoku Shikoku Kyushu Okinawa R � 0.922 0.909 0.935 0.902 0.890 0.933 0.908 0.903 0.873 0.953 generalized R � 0.921 0.907 0.934 0.900 0.887 0.932 0.907 0.901 0.870 0.953 Number of terms 35 39 29 37 39 25 29 32 35 30 Number of predictors adopted 9 9 8 10 9 7 7 7 8 9 Number of input predictors 11 11 11 11 11 11 11 11 11 11 TEMP 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 HUM - - - - - - - - - - SUN 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 WIND - - - 〇 〇 - - - 〇 〇 RAIN 〇 - 〇 〇 - - - - - 〇 SNOW 〇 〇 - - 〇 - - - - - Adopted predictors DI 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 WCI - 〇 - 〇 - - - - - 〇 HDD 〇 〇 〇 〇 〇 〇 〇 〇 〇 - WORK% 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 WAKE% 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 SLEEP% 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 8

■Model performance x-axis：Actual （ MWh ）y-axis：Prediction（ MWh ） High-quality models in terms of 〇 ：In-sample result ＋ ：Out-of-sample result both fitting and generalization ：OLS regression line for in-sample result ：OLS regression line for out-of-sample result � ：The coefficient of determination of OLS (in-sample) � �� � ：The coefficient of determination of OLS (out-of-sample) � ��� Hokkaido Tohoku Tokyo Chubu Hokuriku 5,500 25,000 14,000 5,000 5,000 50,000 4,500 12,000 20,000 4,000 40,000 4,000 10,000 15,000 3,500 3,000 30,000 8,000 3,000 � � 0.922 � � 0.909 � � 0.935 � � 0.902 � � 0.890 � �� � �� � �� � �� � �� 10,000 2,500 � � � � � � ��� � 0.909 � ��� � 0.888 � ��� � 0.934 � ��� � 0.882 � ��� � 0.895 20,000 2,000 6,000 2,000 3,000 4,000 5,000 20,000 30,000 40,000 50,000 10,000 15,000 20,000 25,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 8,000 10,000 12,000 14,000 Kansai Chugoku Shikoku Kyushu Okinawa 11,000 16,000 1,400 5,000 25,000 10,000 14,000 9,000 1,2,00 12,000 4,000 20,000 8,000 10,000 1,0,00 7,000 3,000 15,000 8,000 � � 0.933 � � 0.9 08 � � 0.9 03 � � 0.873 800 � � 0.95 3 6,000 � �� � �� � �� � �� � �� 6,000 5,000 � � � � 0. 896 � � � ��� � 0.922 � ��� � 0.917 � ��� � ��� � 0.896 � ��� � 0.960 600 2,000 10,000 10,000 15,000 20,000 25,000 5,000 6,000 7,000 8,000 9,000 10,000 11,000 2,000 3,000 4,000 5,000 6,000 8,000 10,000 12,000 14,000 16,000 600 800 1,0,00 1,2,00 1,400 9 Observed values vs estimated values

Simulating Temperature Response Functions 10

■Hourly simulation Settings for the simulation Weather predictors x-axis：Temperature（ � ） Regular sequences of the value TEMP y-axis：Power consumption（ MWh ） Weather predictors Average values of each time period at each location in weekdays. HUM, SUN, WIND, RAIN, SNOW The thermal indicators 〇 Observation Calculated from the weather predictors . DI, WCI Hourly simulation Holiday dummy Weekdays: 0 HDD Hour Human activity predictors Values of each time period in weekdays. WORK%, WAKE%, SLEEP% Hokkaido Tohoku Chubu Hokuriku Tokyo 55,000 25,000 14,000 5,000 5,000 50,000 4,500 4,500 45,000 12,000 20,000 4,000 40,000 4,000 3,500 10,000 35,000 3,500 15,000 3,000 30,000 3,000 8,000 2,500 25,000 10,000 2,500 20,000 2,000 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 Kansai Chugoku Shikoku Kyushu Okinawa 1,400 25,000 10,000 5,000 14,000 4,500 9,000 1,2,00 12,000 4,000 20,000 8,000 1,0,00 10,000 3,500 7,000 3,000 15,000 8,000 800 6,000 2,500 5,000 6,000 600 10,000 2,000 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 � 10 0 10 20 30 11