A New Improved Combined Model Algorithm for the Application of Photovoltaic Power Prediction

A new algorithm was proposed in this paper. Based on the principle that the single model prediction results with small errors accounted for a large proportion of the combined results, the BP neural network and the Autoregressive Moving Average (ARMA) Model were improved and combined to form a new combined model algorithm (NEW ARMA-BP) to predict PV output to get more accurate predictions. Taking the output power collected by a 30 MW photovoltaic power plant as an input sample, based on the ARMA algorithm and BP neural network, these two prediction models were built in sequence under the Matlab environment to predict the short-term PV output. Using the “normal test chart of error” in probability theory, we judged that the prediction error level based on the BP neural network was closer to the normal distribution, indicating that the prediction was more accurate. Based on the prediction errors of the ARMA algorithm and BP neural network algorithm, new methods were used to calculate the weights, and the prediction results were recombined based on the weights to obtain new prediction values. Finally, based on the three kinds of error assessment indicators and Pearson coefficient, the error levels of NEW ARMA-BP algorithm and uni-prediction model were compared to verify the applicability of the combined forecast in the field of PV output prediction.