A Novel Deep Learning Approach for Short Term Photovoltaic Power Forecasting Based on GRU-CNN Model

Abstract

The integration of photovoltaic power brings the key to clean energy. However, the increasing proportion of photovoltaic (PV) energy in power systems due to the random and intermittent nature of solar energy resources is causing difficulties for system operators to dispatch PV power stations. To reduce the negative influence of the use of PV power, it is great significant to predict PV power accurately. In this paper, we propose a high-precision hybrid neural network model that employs Gated Recurrent Units (GRU) and Convolution Neural Network (CNN) to build a GRU-CNN model to forecast PV system output power. The proposed framework has two major phases. Firstly, the sample data is divided into training set and test set. For this, the temporal characteristics of the data set are extracted using a GRU model and the spatial characteristics are obtained using the CNN model. Secondly, the final predicted PV power is obtained through the output layer. The forecasting accuracy of GRU-CNN is determined by the mean absolute error (MAE), mean square error (MSE), determination coefficient (R2) and root mean square error (RMSE) values. The findings of the comparison experiments show that the GRU-CNN model has better accuracy than some deep learning methods, including, GRU, CNN and long-short term memory model (LSTM).