Analysis of Swedish Wind Power Forecast Quality over Forecast Horizon and Power System Operation Implications

Abstract: Wind power, as one of the fastest growing means of generation, can oer environmentalbenets. However, due to its stochastic nature it is dicult to designaccurate prediction tools, thus the forecast errors are inherently present. Goodunderstanding of the errors that may occur is imperative for greater penetrationof wind power into the system as it can facilitate power system planning andoperation.In this paper, wind power forecasts for dierent price areas in Sweden areanalysed and compared. For this purpose, data starting November 2015 weredownloaded every hour from the Nord Pool spot database and Python code forparsing and analysis was written. Common indicators, such as the root-meansquare(RMSE) and bias error were used to characterise the accuracy of theforecasts. As expected, it was shown that in general the RMSE decreases as theforecast hour approaches the delivery hour. In addition, systematic bias erroraround the day-ahead market closure time was identied and discussed.The paper continues to analyse wind power forecast error distribution withrespect to the forecast time and dierent production levels. Four statistical momentsof distribution function were calculated and compared. It was shown thatfor the forecast horizons between 0 and 36 hours the forecast error distributionfunctions are negatively skewed leptokurtic.The temporal wind power forecast error correlations between dierent horizonsas well as the spatial correlations between dierent price areas are calculatedand discussed. As expected, a stronger correlation was identied betweenneighbouring price areas. In addition, correlation coecients between the forecasterrors and the up and down regulation prices were calculated.Finally, a model was developed to quantify the amount of operating reservelevel needed to compensate the uncertainty in the system due to wind powerand load forecasts. Future scenarios with an increasing wind power penetrationlevel are simulated and the amount of operating reserve level for individualprice areas is calculated. It was shown that with rapidly increasing wind powerpenetration, especially price area SE3 will need to plan for higher operatingreserve levels to successfully cope with the uncertainty in the load and windpower forecasts.