Research On Distributed Photovoltaic Power Prediction Based On Machine Learning

In recent years,our country has vigorously promoted solar energy technology to cope with the increasingly serious energy and environmental crisis.Among them,photovoltaic power generation technology has become the main direction of current research.However,since the output of photovoltaic power generation is greatly affected by meteorological factors and has strong instability,the connection of a high proportion of distributed photovoltaics will have a greater impact on the power quality,causing huge impacts and challenges to the power system.Accurate prediction of photovoltaic power generation output can not only improve the operating efficiency of photovoltaic power plants,but also help power dispatching departments to adjust the operation mode to ensure the safe,stable and economical operation of the power system after a high proportion of distributed photovoltaics are connected.Therefore,the effective and accurate prediction of distributed photovoltaic power is of great significance to the safe and stable operation of power grids and high-quality power supply.This paper mainly studies the problem of distributed photovoltaic power prediction from three aspects:input dimension selection,machine learning model prediction,and BP neural network model prediction optimized by intelligent algorithm.The specific work is as follows:(1)The meteorological factors and distributed photovoltaic power panel data were constructed,and the variance inflation factor method was used for collinearity analysis to deal with the multi-collinearity problem between meteorological factors.Two estimation methods,SCC and FGLS,were used for the interpreted regression analysis on panel data,and dominance analysis method was used to compare the relative importance of meteorological factors on distributed photovoltaic power.The importance of each influencing factor was sorted and analyzed,which provided a reference for the selection of input dimensions for distributed photovoltaic power prediction.(2)The data set was divided into four weather types:sunny,cloudy,overcast and rainy,and a similar day discrimination model based on grey correlation degree and cosine similarity was constructed,and the similar days of each type of forecast days were selected as the training set of the machine learning models.In different weather conditions,the prediction performance of a total of seven machine learning models,including BP neural network,random forest,XGBoost,K-nearest neighbor configured with mean regression,K-nearest neighbor configured with distance weighted regression,support vector machine configured with linear kernel function,and support vector machine configured with radial basis kernel function was compared and analyzed.The results showed that BP neural network model had the best prediction effect in sunny days,cloudy days and overcast days,and the support vector machine model configured with radial basis kernel function had the best prediction effect in rainy days.(3)The initial weights and thresholds of the BP neural network(BPNN)were optimized by the Genetic Algorithm(GA)and the Sparrow Search Algorithm(SSA)respectively.After constructing GA-BPNN and SSA-BPNN models,the prediction performance of BPNN,GA-BPNN and SSA-BPNN models was compared and analyzed.The results showed that the prediction accuracy of SSA-BPNN model in each weather type was higher than that of BPNN and GA-BPNN model.