Short-term Forecasting Approach Based on bidirectional long short-term memory and convolutional neural network for Regional Photovoltaic Power Plants

Abstract

Accurate photovoltaic (PV) generation output prediction is one of the effective ways to ensure the safe operation of power grid, develop reasonable dispatching plan and improve the efficiency of clean energy. With the large-scale operation of PV power plants in recent years, forecasting regional PV output becomes more significant. We proposed a short-term forecasting approach based on bidirectional long short-term memory and convolutional neural network (BiLSTM-CNN) for regional PV power plants. First, the k-means algorithm is used to divide power plants with similar generation characteristics into the same output subregion. Second, a representative power plant in each subregion is selected based on three correlation coefficients. Then, we develop a regional prediction model based on BiLSTM-CNN method. This model takes historical operation and meteorological data of the representative power plant as input, and takes the total subregional power generation as output. Finally, this short-term forecasting approach is tested using real data from PV power plants in Chuxiong and Dali region, Yunnan province, China. The comparison of numerical results shows this proposed method can effectively improve the short-term prediction accuracy of regional PV generation output.