Photovoltaic Output Power Prediction Model Based On K-GCN-LSTM Neural Network

The high proportion of distributed photovoltaic integration into the distribution grid can assist the power grid peak shaving while simultaneously reducing transmission losses. However, accurately predicting photovoltaic output power is challenging due to its volatility and randomness. To address the issue of not considering the spatial characteristics among photovoltaic arrays in general photovoltaic output power prediction, this paper introduces a predictive model based on Kendall coefficient, Graph Convolutional Neural Networks and Long Short-Term Memory Neural Network(K-GCN-LSTM). First, the model sets the neighbor matrix of GCN with Kendall coefficients between the output powers of 32 photovoltaic arrays and extracts the spatial correlation between the output powers of PV arrays using GCN. Then, the model uses Long Short-Term Memory Neural Network to learn the temporal correlations of photovoltaic output power curves. Finally, the proposed prediction model is compared with the traditional prediction models, which confirms that the proposed model is more accurate in the prediction results and has a positive impact on the safe operation of the power system to cope with the high proportion of PV integrated into the distribution network.