Short-Term Forecasting and Uncertainty Analysis for Wind Power Generation Based on Hybrid Deep Learning and Cloud Model

Abstract

Due to the fluctuating and intermittent characteristics of wind energy, it leads to uncertainty in forecasting. In order to improve the wind power forecasting (WPF) accuracy, the paper proposes a CNN-BiLSTM model based on the multi-convolution and multi-pooling(MCP) method for the short-term forecasting model of wind power and photovoltaic power generation, and performs multi-scale forecasting and analysis of the output power in a wind farm. The result analysis verified the forecasting accuracy of CNN-BiLSTM model at 4 hours, 24 hours and 72 hours is higher than those of LSTM, BP neural network, BP-PSO hybrid model and wavelet neural network. The uncertainties in WPF caused by different forecasting models at different time scales are qualitatively described by the expectation, entropy, and hyper-entropy of cloud model. The quantification of the uncertainties in WPF are measured by the confidence intervals based on the non-parametric kernel density estimation (NPKDE). The results show that the proposed method can improve the predict accuracy on the uncertainties in WPF at different confidence levels.