Electricity demand forecasting and the problem of embedded generation

Transcription

1 Electricity demand forecasting and the problem of embedded generation Place your chosen image here. The four corners must just cover the arrow tips. For covers, the three pictures should be the same size and in a straight line. John Young 6 th March 2013

2 Operating the system Electricity National Control Centre 2

3 Operating the system Generation Demand Frequency Hz Normal operating frequency Upper statutory limit Generators tripping Lower statutory limit Demand disconnection starts Demand disconnection 49.1 complete

4 Demand profile shapes Shape of demand curves in terms of turning points and points of inflections - remains fairly constant from day to day Exact position of turning points, both in vertical (Demand) and horizontal (Time) directions varies, at least partially because of weather and non-weather variables Shape evolves slowly over time, with some abrupt discontinuities 4

5 GB National Demand A Typical Daily Profile: January Winter Peak ~ 56,000 MW Winter Minimum ~ 32,000 MW 5

6 GB National Demand A Typical Daily Profile: January Winter Peak ~ 56,000 MW Winter Minimum ~ 32,000 MW 6

7 GB National Demand A Typical Daily Profile: February Winter Peak ~ 56,000 MW 7

8 GB National Demand A Typical Daily Profile: March Winter Peak ~ 56,000 MW 8

9 GB National Demand A Typical Daily Profile: April Winter Peak ~ 56,000 MW 9

10 GB National Demand A Typical Daily Profile: May Winter Peak ~ 56,000 MW 10

11 GB National Demand A Typical Daily Profile: June Winter Peak ~ 56,000 MW 11

12 GB National Demand A Typical Daily Profile: June Winter Peak ~ 56,000 MW Summer Maximum ~ 40,000 MW Summer Minimum ~ 20,000 MW 12

13 Forecasting electricity demand Typical demand profile shape 2 distinct shapes: GMT and BST 13

14 What Else Affects Demand? Time of Day Bank Holidays School Holidays Day of Week 14

15 Day of week impact Demand curve for weekday in GMT Friday Demand curve for Saturday in GMT Demand curve for Sunday in GMT Monday 15

16 What Else Affects Demand? Time of Day Day of Week Bank Holidays School Holidays Weather Special Events TV 16

17 Demand Effect (MW) Temperature COLD High Demand HOT Quite High Demand MILD Low Demand Temperature 17

18 Demand Effect (MW) Illumination DULL High Demand BRIGHT Low Demand Radiation 18

19 Demand (MW) The Impact of Weather Cooling Power of the Wind Low Wind: Low Demand Strong Wind: High Demand Wind Speed 19

20 The Impact of Weather Rain 20

21 The Impact of Weather Some Numbers Temperature (1 C fall in cold conditions) MW Cloud cover (clear sky to thick cloud) + 1,500 MW Precipitation (no rain to heavy rain) + 1,000 MW Temperature (1 C rise in hot conditions) MW Cooling power (10 mph rise in cold conditions) + 1,000 MW 21

22 Weather Variables 4-Hourly Average Temperature [TO] Effective temperature [TE] Wind Speed [WS] Cooling Power of the Wind [CP] a function of Wind Speed and Temperature (TO) Effective illumination of the Sky [EI] - A derived quantity calculated from radiation levels and measurements of cloud type and cover 22

23 Non-weather variables Day of week Year Effect indicator variable for different years: mostly owing to different economic conditions Time of year seasonality Time of Sunrise and Sunset School Holidays - % of schools on holiday Annual Holidays indicator variable from common August holiday weeks Bank Holidays excluded from data set for purposes of modelling, then deal with on an ad hoc basis 23

24 GB National Demand Cardinal Points DP 2B 1B 24

25 Standard Linear Regression Conventional Models Model Inputs: Weather Component Day of Week Component Basic Demand Historic Demands Historic Weather Heathrow, Glasgow, Manchester, Bristol, Leeds, Bimingham Additional Effects School Holidays, Day of Week, Time of Year 25

26 Modelling Construct different models for each of the Cardinal Points (CPs) Construct different models for GMT and BST Construct models of two different types for each CP: Standard linear regression models (Conventional Models) Time series models with linear regression (Trend models) Depending on the CP we construct 7 day models, 5 day models, Saturday models and Sunday models On any day of the week there are at least two (and up to four) models that we forecast with 26

27 2B Demand (12:30) 27

28 2B Model On Day Of Week Effect Actual Demand Vs Fitted Values 28

29 Including A Seasonal Effect Actual Demand Vs Fitted Values 29

30 Including A Weather Effect Actual Demand Vs Fitted Values 30

31 Including A School Holiday Effect Actual Demand Vs Fitted Values 31

32 Time Series with Linear Regression Trend Models Yesterday Today 1F 1A 04 1B 1F 1A 04 1B 2A 2B 3B 3C/DP 4B 32

33 Including A School Holiday Effect Actual Demand Vs Fitted Values 33

34 Including A Trend Component Actual Demand Vs Fitted Values 34

35 The Model Symbolically L2B ~ L04 DWK SCH EI CP Day of Week Effect Weather Effect Trend term Error term 35

36 Modelling CP Demand Construct forecast models using variables that make sense Use best model possible with variables that reduce the residual error significantly Weather Component Day of Week Component Track Basic Demand, non-model component of demand Basic Demand 36

37 Basic Demand Manually track and forecast basic element of demand 37

38 Profile Matching Check how well CP forecasts match historic days 48 Daily Demand Profile Actual_ Forecast_ Eview s

39 Choosing Basics Forecast basic demand 2B model: Jan-Feb 2013 Aim is to reduce risk of error 39

40 Embedded PV Generation 1,800 1,600 1,400 1,200 1,

41 Embedded PV Generation Daily Demand Profile Actual_ Forecast_ Eview s

42 Embedded Generation Invisible, non-metered Connected directly into distribution networks Effectively reduces demand on the system Not just PV 42

43 The Impact of Embedded Generation Embedded Generation National Demand True GB Demand is higher than National Grid observe Not a new phenomenon, but an increase in more variable technologies means it is a more significant effect Wind Power ~ 2,000 MW Solar Power ~ 1,500 MW 43

44 Virtual Demand: A True National Demand Embedded Generation National Demand Virtual Demand 44

45 Model Using Virtual Demand Weather Component Week Day Component Virtual Demand Basic Demand 45

46 Forecast Virtual Demand; Adjust for Embedded Generation Weather Component Week Day Component National Demand Basic Demand Embedded Generation 46

47 The Forecasting Process Model using Virtual Demand Forecast Virtual Demand; Adjust for Embedded Generation Embedded Generation Weather Component Weather Component National Demand Week Day Component Week Day Component National Demand Basic Demand Basic Demand Embedded Generation 47

48 Forecasting Embedded Wind Generation Existing Forecasting Methods Wind Power Forecasting System - Metered Wind ~ 5,800 MW - Embedded ~ 2,000 MW Metered Wind Farms Embedded Wind Farms Metered Wind Power Forecast National Demand Forecast 48

49 Load Factor Standard Wind Power Curve Wind Farm Decile wind speed forecast applied to a load curve Load curves for each wind generator, optimised using actual metering Forecast Wind Speed / mph 49

50 03-DEC :00 8:00 12:00 17:00 21:00 04-DEC :00 8:00 12:00 17:00 21:00 05-DEC :00 8:00 12:00 17:00 21:00 06-DEC :00 8:00 12:00 17:00 21:00 07-DEC :00 8:00 12:00 17:00 21:00 Wind Power forecast probabilistic view for next 5 days from Mon 3 rd Dec ,000 Wind Power Forecast - Probabilistic View for Next 5 Days 20% confidence 40% confidence 60% confidence 80% confidence Mean Forecast Excluding Cut-out Mean Forecast 4,000 3,000 2,000 1,000 0 Wind cut out forecast 5,000 Wind Cut-out Forecast 20% confidence 40% confidence 60% confidence 80% confidence Mean Cut-out 4,500 4,000 3,500 3,000 2,500 2,000 1,500 1,

51 Metered wind generation forecast Use same process to forecast embedded wind Have information on location and capacity for all embedded wind generators above 2MW 51

52 Forecasting Embedded Generation Simulated National PV Output 1,800 1,600 1,400 1,200 1,

53 Load Factor Forecasting Embedded Generation Forecasting PV 90% 80% 70% 60% 50% 40% 30% 20% Mark Holland - Load Factor Vs Heathrow Radiation National average radiation forecast Generic power curve National capacity % 0% ,000 1,500 2,000 2,500 3, % 80% 70% David Lenaghan - Load Factor Vs Odiham Radiation % % 40% % % 10% % ,000 1,500 2,000 2,500 3, ,000 1,500 2,000 2,500 3,000 National Average RA 53

54 The challenges of PV Embedded Wind Individual locations and capacities Wind speed forecasts for various locations Experience forecasting metered wind Embedded PV Overall capacity for whole country National Average radiation No operational experience yet 54

55 Density 0e+00 2e-04 4e-04 6e-04 Density 0e+00 2e-04 4e-04 6e-04 X3B_Basic X3C_Basic Forecasting Embedded Generation Forecasting PV Index Index density.default(x = X3B_Basic + Output_1500, bw = 300) density.default(x = X3C_Basic + Output_1700, bw = 300) N = 57 Bandwidth = N = 57 Bandwidth =

56 The Problem with Embedded Generation ~ 3,500 MW installed capacity Variable output dependant on weather effects Changing capacity levels Reliant on estimates of output No means of directly testing forecast models Increases the volatility of National Demand 56

57 Including A Trend Component Actual Demand Vs Fitted Values 57

58 Including Embedded Wind Generation Actual Demand Vs Fitted Values 58

59 Including Embedded Solar Generation Actual Demand Vs Fitted Values 59

60 60

61 Thanks for listening 61

62 Demand and Wind Forecasting 62