Probability Prediction of Short-Term Wind Power Based on Quantile Regression Forest and Variable Bandwidth

It is important to predict the probability interval of wind power with uncertainty, which is usually applied to optimize the capacity of spinning reserve in electric power system. In order to quantify the confidence interval of the wind power prediction, a kind of probability prediction method called the quantile regression forests model (QRF) has been proposed to construct the non-linear mapping relations between multi-inputs and double-outputs, in which the clustering effect of different forecasting results is estimated by nonlinear regression analysis to obtain the lower-upper bounds of confidence interval under different confidence levels. Instead of the fixed bandwidth, an index of variable bandwidth based on the instantaneous mean value and standard deviation is proposed, in order to acquire the higher coverage rate and the narrower bandwidth. Case study suggests that it has a better application effect on probability interval prediction of wind power by the proposed method in the paper, compared with the BP neural network & Bayesian network & Support Vector Machines model.