Research on Distributed Photovoltaic Data Prediction Based on Deep Learning Method

Abstract

Aiming at the problem of distributed photovoltaic output fluctuation caused by weather state change and the technical bottleneck of the existing prediction model lacking refined meteorological information support, this paper constructs a long short-term memory (LSTM) network prediction model that integrates the characteristics of meteorological historical data. Firstly, based on the   method meteorological information fusion technology, the historical meteorological data set is divided into four basic categories: sunny, cloudy, cloudy, rain and snow according to typical weather characteristics, and special attention is paid to the weather transition process between various categories to form a fusion data set covering the complete weather evolution scene. Then, based on the data set training, a distributed photovoltaic short-term power prediction model is constructed. Finally, by inputting the weather type parameters of the day to be predicted, the LSTM network power prediction results are output and the validity of the model is verified. The experimental data show that compared with the traditional BP neural network model, the average prediction error of the fusion model proposed in this paper is significantly reduced, which can provide more accurate decision-making basis for distributed energy access planning and optimal scheduling and resource allocation of power system.