Use probabilistic forecasting to model uncertainties in electricity markets-A wind power example

Uncertainties in electricity markets increases with the variations of load, generation conditions, fuel prices, and renewable energy resources. Probabilistic forecasting can quantify uncertainties and assist decision making in electricity markets. This chapter uses wind power generation as an example to demonstrate probabilistic forecasting of renewable generation in power grids with high penetration of renewable energy resources. This work first develops an enhanced k-nearest neighbor (KNN) algorithm to find similar weather condition days in historical dataset. Then a novel kernel density estimator (KDE) is developed and applied to derive the probability density of wind power generation from k-nearest neighbors. Logarithmic transformation is utilized to reduce the skewness of wind power density, and the boundary kernel method is used to eliminate the density leak at the bounds of wind power density. In addition, an outlier detection tool is employed to remove the data of wind power output being zero in a successive 48h period due to wind turbine maintenance. An advantage of this approach is that it could provide both point forecasts and probabilistic forecasts for wind power generation. Evaluation results on the public dataset from Global Energy Forecasting Competition 2014 (GEFCom 2014) demonstrate the high accuracy and reliability of the proposed technology.