Hybrid Neural Network based Power Generation Prediction of Photovoltaic Farms

The large-scale penetration of intermittent photovoltaic power generation has brought direct challenges to the safe and reliable operation of the power systems. Accurate prediction of photovoltaic power generation can reduce the negative impact of generation fluctuations and support optimal power system planning and energy dispatch. This paper presented a hybrid neural network based photovoltaic power generation prediction solution. Firstly, the data preprocessing is carried out for prediction, e.g., wavelet denoising for power data and polynomial feature engineering for weather factors. Then, the long short-term memory network (LSTM) and convolutional neural network (CNN) models are adopted and the SE channel attention mechanism is integrated into the convolutional neural network. The proposed solution can achieve complementary advantages and presents the divergence-fusion structure CNN-SE-LSTM hybrid model. The proposed solution is assessed through simulation experiments and the numerical result confirms the effectiveness of the feature engineering and the superiority of the hybrid model in photovoltaic power prediction.