Prediction model of photovoltaic power generation based on improved granular computing and neural networks

The application potential of solar energy generation is immense, yet its volatility poses significant challenges to power supply. With the aim of enhancing the accuracy of photovoltaic power generation prediction to better address its variability and stochastic nature, this study proposes a method based on enhanced granular computing and neural networks. The model utilizes wavelet transform to analyze the original time series in the frequency domain and employs fast Fourier transform to extract major periodic components. These components are then used to construct granularity matrices for training a backpropagation neural network (BPNN), aiming to achieve precise prediction of photovoltaic power generation. The research findings demonstrate that the model achieves an average error rate of around 17% on the test dataset, exhibiting outstanding performance compared to other classical time series prediction models. This study provides an effective method and reference for photovoltaic power generation prediction, contributing significantly to the field.