ARIMA model, RBF neural network model, Distributed new energy, Power peak prediction, High-frequency data, Low-frequency data, Renewable energy sources, TJ807-830, Energy industries. Energy policy. Fuel trade, HD9502-9502.5

Abstract Distributed new energy generation is influenced by various factors, resulting in high volatility and uncertainty. To enhance control, promote renewable energy consumption, and balance power supply and demand, an ARIMA–RBF-based method is proposed for predicting peak power of distributed new energy connected to the grid in different periods. Historical power data are selected using a similar-day method and segmented into different periods using the Fisher optimal division method. Wavelet transform is applied to divide the power time-series data into high-frequency (fluctuation information) and low-frequency (trend term) components. These components serve as input for ARIMA and RBF neural network models, respectively, for power peak prediction. By integrating the prediction results of both models, the final peak power predictions are obtained. Experimental results demonstrate that the method effectively selects similar days, reducing computational requirements, and achieves high prediction accuracy with minimal deviations. By dividing the power data into four periods, the prediction of power peaks for distributed new energy is improved. This study presents an innovative approach to managing and integrating renewable energy sources into the grid.