Research of the Direct Normal Irradiance Prediction for Solar Thermal Power Stations based on SVMD-PSO-LSTM model

The inherent randomness and variability of solar energy can cause instability in the output power of solar thermal power generation systems that rely on it as an energy source. To better leverage the advantages of solar thermal power stations in practical operations, a prediction model for Direct Normal Irradiance (DNI) in solar thermal power stations is proposed, taking a solar thermal power station in Qinghai Province as the research object. Firstly, the DNI data is decomposed and reconstructed using the Successive Variational Mode Decomposition (SVMD) to obtain multiple feature vectors. Then, the PSO algorithm is employed to fine-tune the parameters of the LSTM network, thus enhancing the predictive precision for each individual attribute. Finally, the predicted results of each component of IMF are integrated and accumulated, resulting in the successful acquisition of the entire predicted value of the DNI time series. In comparison with LSTM, VMD-LSTM and PSOLSTM model, the SVMD-PSO-LSTM model demonstrates superior prediction efficiency, reduced MAE and RMSE, and enhanced $\mathbf{R}^{\mathbf{2}}$, thereby substantiating its elevated precision and robust generalisation capability.