The uncertainty of wind output how it declines during the day.

Transcription

1 Conference: "The role of market design for the achievement of the 20% Renewables Target", Brussels, The uncertainty of wind output how it declines during the day. State of the art and research priorities in wind power forecasting George Kariniotakis Head of Renewable Energies & SmartGrids Team (CEP)

2 Outline 1. Introduction 2. General principles of wind power forecasting 3. Typical accuracy of state of the art models 4. R&D projects & priorities 2

3 Introduction GW GW GW (!!) Challenges g Reliable large-scale integration Source: EWEA Economic and secure management of a power system Competitiveness of wind energy in a liberalised electricity market 3

4 Introduction Wind power forecasting for the next minutes to days is recognised today as a necessary tool for 'smooth' wind integration However, forecasting wind power is a complex problem. Some of the reasons Power/Pnominal [p.u.] Power [kw] Time [x 1 hour] Wind is highly variable by nature (example: wind production of a wind farm during a month) P[i] steep slope in the usual wind speed range v[i] cut-off effect Wind speed [m/s] The wind turbine characteristic curve introduces important non-linearities. 4

5 The principle p 3. Production [MW] TERRAIN 2. WEATHER FORECASTS 6 1. SCADA PREDICTION MODEL t0 +1h +6h +12h +18h +24h +30h +36h. +48h 5

6 Types of forecast errors Level error Phase error M1-M5: wind power predictions by statistical and physicam models 6

7 Factors affecting accuracy Prediction horizon Time of the year (climatic conditions) Terrain complexity Types of inputs/types of models Spatial smoothing effect Level of predicted power and speed. 7

8 Factors affecting accuracy: Prediction horizon Average error increases with horizon Case Study: wind farm in Ireland Error [% of Pn] Mean Abs. E Norm. M Comparison of 3 statistical (M1-M3) and 2 physical (M4-M5) models 8

9 Factors affecting accuracy: Time of the year Error varies throughout the year Case Study: wind farm in Ireland [%] NMAE Persistence F-NN model Oct. Nov. Dec. Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Total Month of the year and Total 5 0 9

10 Factors affecting accuracy : Terrain complexity Error increases with terrain complexity 10 Average NMAE (% nomina l power) TUN (offshore) KLI WUS FLAT TERRAIN SOT GOL COMPLEX TERRAIN ALA HIGHLY COMPLEX TERRAIN Terrain Roughness Index RIX (%) 10

11 Factors affecting accuracy: Spatial smoothing effect Example : Prediction of 2200 MW in Jutland area in DK SCADA measurements only from 23 WFs (~200 MW). Differenced measurements of the total production. Prediction of 23 WFs output Then upscale to the total area. NMAE (% of Pn) Singe WFs perform. Upscaling Model 1 Upscaling model Look-ahead time (h) 11

12 Typical performance MEASURE : Normalised Mean Absolute Error (NMAE) Case Study: wind farm in Ireland Comparison of 3 statistical (M1-M3) and 2 physical (M4-M5) models TYPICAL VALUES : 4% -18% for a single wind farm in complex terrain. Norm. Mean Abs. Error r [% of Pn] Much lower error for regional/national forecasting due to spatial smoothing effect for horizons >4-6 hours. ~5-10% for hours. 12

13 Outline 1. Introduction 2. General principles of wind power forecasting 3. Typical accuracy of state of the art models 4. R&D projects & priorities 13

14 Context Research in Wind Power Forecasting Considerable research carried out in the last 20 years HIGHLIGHT PROJECTS : ANEMOS : FP5, Meteorology Wind power forecasting technology

15 Context Research in Wind Power Forecasting Considerable research carried out in the last 20 years HIGHLIGHT PROJECTS : ANEMOS : FP5, ANEMOS.plus : FP6, Meteorology ANEMOS.plus Wind power forecasting technology Operational decision making 15

16 Context Research in Wind Power Forecasting Considerable research carried out in the last 20 years HIGHLIGHT PROJECTS : ANEMOS : FP5, SafeWind : FP7, ANEMOS.plus : FP6, Meteorology SafeWind Wind power forecasting technology Operational decision making 16

17 R&D priorities* Better estimation of uncertainty - Probabilistic forecasting - Ensembles Prediction of "extreme events" (i.e. ramps) Research in meteorology oriented to wind power forecasting Methods based on geographically distributed measurements Methods for alarming and warning Prediction risk indices Scenarios Reliable prediction systems Probability density (*) priorities in Anemos.plus and SafeWind projects 17

18 R&D priorities* Decision making based on forecasts and uncertainty estimation Advanced tools are demonstrated in Anemos.plus for : reserves estimation, power system scheduling, coordination of wind with storage, congestion management, strategic bidding in markets Probability density (*) priorities in Anemos.plus and SafeWind projects 18

19 Thank you for your attention 19