Photovoltaic power prediction based on multi-scale photovoltaic power fluctuation characteristics and multi-channel LSTM prediction models

The global shortage of non-renewable energy sources has catalyzed the vigorous development of photovoltaic (PV) energy. Accurate prediction of PV power output is essential for ensuring the safety and stability of integrating small-scale PV systems into the power grid. Therefore, this paper proposes a hybrid multi-station parallel PV power prediction method (MCFC-MAOA-MCLSTM-Attention) based on multi-scale historical PV power fluctuation feature extraction. First, for the problem of variable weather types due to the existence of strong fluctuations in meteorological factors, a weather classification algorithm based on multi-scale fluctuation characteristics (MCFC) is proposed, and combined with the similar day algorithm to select the classified meteorological data secondly and improve the correlation between the data. Subsequently, this paper proposes a multi-channel structured long and short-term neural network modeling method (MCLSTM) to further extract the spatio-temporal correlation of different PV sites in the region and realize the integrated prediction based on geographic location and time series. To address the challenges associated with calibrating model parameters, which significantly impact the prediction accuracy, the modified Archimedean optimization approach (MAOA) was employed to optimize these parameters. The experimental results demonstrate that the model is both highly reliable and generalizable for predicting photovoltaic power data.