Photovoltaic Power Prediction Method based on Principal Component Analysis and Gated Recurrent Unite Network

Solar energy is a renewable, widely distributed, low-cost clean energy. Photovoltaic (PV) power values are a sequence of values that have strong randomness but can be predicted. Accurate photovoltaic power prediction is an important basis for reasonable power scheduling and an important guarantee for the smooth operation of power systems. This manuscript proposes a PV prediction method based on Principal Component Analysis (PCA) and Gated Recurrent Unite (GRU) networks. We call this method PCA&GRU. PCA&GRU first uses PCA to perform dimensionality reduction on the factor data of photovoltaic output. Through dimensionality reduction, the feature dimension is reduced, and the network prediction time is reduced. Moreover, the prediction network has mined highly correlated photovoltaic output factors. After that, the dimensionality reduced photovoltaic output factor data is input into the GRU networks and the predicted power is obtained. GRU networks have a simpler structure and higher accuracy than these traditional Long Short Term Memory (LSTM) based methods. Residual blocks are added between the GR units in the prediction network, which improves the ability of the GRU to extract short-term features. The simulation results on the Yulara Solar dataset demonstrate the proposed PCA&GRU can accurately predict short-term photovoltaic power.