IGRA-LSTM Photovoltaic Power Generation Prediction Model Considering Weather Factors and Similar Days

Photovoltaic power generation has developed rapidly because of its clean and renewable characteristics of solar energy[1]. However, during large-scale grid integration, the impact of fluctuations, intermittency, and randomness in photovoltaic output power will become increasingly significant.This paper proposes a newly short-term energy power prediction model that combines data preprocessing with Long Short-Term Memory(LSTM).Firstly, this model uses physical exclusion and quartering to identify and eliminate collected abnormal data, which improves the accuracy of data usage.Secondly, based on the 8 main features related to power generation, the Pearson algorithm is used to calculate their correlation, and clustering analysis is used to classify the weather data into 4 categories.Thirdly, an improved IGRA algorithm is proposed to compare predicted days with historical days, extract the 8 most similar feature terms and historical photovoltaic power generation on the aforementioned days. By combining the historical data of the correlation and similarity days of 8 major feature items with a time recurrent neural network, future photovoltaic power can be predicted. Finally, based on historical data from a photovoltaic power plant in Liaoning, a hybrid IGRA-LSTM neural network model was used for prediction.