PV Power Forecasting Based on VMD-RIME-LSTM

Addressing the challenges posed by the strong nonlinearity and numerous influencing factors of PV power data, this study proposes a novel combined forecasting method that integrates the Variational Modal Decomposition (VMD) and the frost-ice optimization algorithm (RIME) with the Long- and Short-Term Memory neural network (LSTM). Initially, the historical PV power data is decomposed using VMD to break down the complex, non-linear time series into simpler components. Subsequently, the parameters of the LSTM neural network are optimized using the RIME algorithm, which enhances the network's ability to model and predict the decomposed data effectively. Each decomposed time series component is then input into the LSTM neural network individually. The predicted values of each component are subsequently aggregated to obtain the final predicted time series values. The experiment utilizes actual PV power data from a 250MW PV power plant located in a region in the northern hemisphere. By comparing and analyzing the performance of the standalone LSTM model and the VMD-LSTM model, it is demonstrated that the proposed VMD-RIME-LSTM model significantly enhances forecasting accuracy. The results indicate that this combined approach effectively addresses the inherent complexities of PV power forecasting and showcases substantial potential for practical application in real-world scenarios.