Short-term power load probability density forecasting based on quantile regression neural network and triangle kernel function

Abstract

Highly accurate short-term power load forecasting (STLF) is fundamental to the success of reducing the risk when making power system planning and operational decisions. For quantifying uncertainty associated with power load and obtaining more information of future load, a probability density forecasting method based on quantile regression neural network using triangle kernel function (QRNNT) is proposed. The nonlinear structure of neural network is applied to transform the quantile regression model for constructing probabilistic forecasting method. Moreover, the triangle kernel function and direct plug-in bandwidth selection method are employed to perform kernel density estimation. To verify the efficiency, the proposed method is used for Canada's and China's load forecasting. The complete probability density curves are obtained to indicate the QRNNT method is capable of forecasting high quality prediction interval (PIs) with higher coverage probability. Numerical results also confirm favorable performance of proposed method in comparison with the several existing forecasting methods.