Short term photovoltaic power prediction based on GJO-RNN-LSTM model

To address the issue of low prediction accuracy of Ridgelet Neural Network (RNN) when data fluctuates greatly, a combined neural network model combining RNN and Long Short Term Memory (LSTM) is proposed to improve prediction accuracy. Then, in response to the overfitting problem in the RNN-LSTM model, the Dropout Method was used to optimize the model structure, and the Golden Jackal Optimization (GJO) algorithm was used to optimize the hyperparameters of the RNN-LSTM. In response to the problem that the GJO algorithm may fall into local optima, improvements have been made to the GJO algorithm to further enhance its convergence accuracy. Finally, simulation analysis was conducted using actual data, and the power prediction accuracy MAE was found to be 0.85kW, RMSE was 1.10KW, and MAPE was 0.29%, confirming that it can significantly improve prediction accuracy.