Renewable Energy Output Prediction Method Based on Recurrent Neural Network of Double Attention Mechanism

Abstract

In recent years, the implementation of the concept of sustainable development, new energy is widely connected to the grid. However, the power generation of wind power, photovoltaic and other renewable energy sources is influenced by external factors and has obvious volatility and intermittency. The time series of electrical energy output in the system is a typical non-stationary signal, which is difficult to predict accurately. In order to improve the prediction accuracy of the short-term renewable energy output prediction model from the data level, a recurrent neural network based on a dual-attention mechanism (DA-RNN) is established in this paper. In the encoder stage, the input features are extracted based on the driving sequence. In the decoder stage, the importance weights of relevant encoder information on time are adaptively optimized when predicting the target sequence. Through the prediction experiments of solar power and wind power, the method used in this paper has improved the prediction accuracy compared with the statistical method and LSTM, and can effectively smooth out the volatility of new energy production.