A Short-term net load hybrid forecasting method based on VW-KA and QR-CNN-GRU

Abstract

The penetration of distributed photovoltaics on the user side increases the fluctuation of the net load and makes it harder to forecast. To improve forecasting precision, a short-term net load hybrid forecasting model is proposed, which considers weather classification and neural network. First, a weather conditional factor is constructed to describe weather conditions. A k-means algorithm based on the volatility of weather conditional factor (VW-KA) is put forward for weather classification. Besides, due to the temporal differences in the impact of weather factors on the net load, the maximal information coefficient is used to choose the input features by segmented selection. Then, a hybrid neural network consisting of quantile regression and convolutional neural network-gated recurrent unit (QR-CNN-GRU) utilizes the results of the sample selection stage as inputs to perform hybrid prediction, and generated the probability density curves by kernel density estimation. The superiority of the model is validated by comparing the obtained results with other models.