Short-term photovoltaic power prediction based on improved EMA variable weight combination

Accurate and consistent regional photovoltaic power prediction is very important for efficient scheduling and utilization of renewable energy, and photovoltaic power prediction plays a key role in all aspects of system energy planning and management. In order to improve the prediction accuracy of photovoltaic power, a short-term photovoltaic power forecasting model based on improved exponentially weighted moving average (EMA) variable weight combination is proposed in this paper. First of all, this paper considers the correlation between the influencing factors and the generation power in the photovoltaic power generation process, uses the Spearman rank correlation coefficient analysis method to obtain the correlation coefficient of each factor, and takes the factors with high correlation as the input of the characteristic variables of the prediction system. Secondly, the random forest method is used to preprocess the missing and outliers in the data set. Then, the preliminary prediction of photovoltaic power is obtained by using convolutional neural network combined with long-term memory network (CNN-LSTM), extreme gradient lifting (XGBoost) and support vector regression (SVR). Then, by improving EMA’s weight dynamic allocation method and combining multiple forecasting algorithm, a short-term photovoltaic power forecasting model with variable weight and multiple combination is constructed. Finally, a photovoltaic electric field in Yunnan is taken as a case study. The analysis results show that the combined multi-algorithm prediction model is better than the single prediction algorithm, and the evaluation index of the prediction results is better than the single algorithm, which can make a more accurate and effective prediction of photovoltaic power.