Ultra-Short-Term Photovoltaic Power Prediction Based on Improved Kmeans Algorithm and VMD-SVR-LSTM Model

As the proportion of photovoltaic connected to the power grid increases year by year, the safe and stable operation of the power grid is facing new challenges. Thus accurate prediction of photovoltaic power generation is of great significance to the economic dispatch of the power grid and photovoltaic consumption. In this paper, historical data are clustered based on Kmeans++ algorithm and then decomposed by variational mode decomposition algorithm, which reduces the volatility of data and fully excavates the characteristics of data used for prediction. After the combined prediction model based on support vector regression and long short-term memory network is established, the model with better prediction effect is selected for the prediction of each component obtained by decomposition, which improves the prediction accuracy of single modal component, and then improves the final prediction accuracy after reconstruction. Simulation results show that the proposed hybrid model can give full play to the advantages of each algorithm, and prediction effect is better than traditional single prediction model.