Research On Photovoltaic Power Prediction Based On Machine Learning

With the increasingly prominent global environmental issues,the continuous growth of energy demand and the gradual maturity of photovoltaic power generation technology,the photovoltaic power generation industry has developed rapidly in recent years.However,photovoltaic power generation has strong randomness and volatility,as the proportion of photovoltaic power generation installed capacity continues to increase,the uncertainty of its output power has brought a series of scheduling operation problems.The accurate prediction of the photovoltaic power generation is helpful to reduce the impacts of uncertainty,it can provide decision-making basis for the power dispatching department to formulate reasonable dispatching plan,reduce the adverse impact of grid connected photovoltaic units,and improve the level of photovoltaic power consumption.This paper firstly studies the basic principles of photovoltaic power generation and the impacts of several major meteorological factors on photovoltaic power generation,and introduces data preprocessing methods,including data normalization and abnormal value processing.Next,the principles of BP neural network(BP),extreme learning machine(ELM),long short-term memory neural network(LSTM),and support vector machine(SVM)are analyzed in detail,according to different applications,a photovoltaic power prediction model based on particle swarm optimization(PSO)optimized extreme learning machine(ELM)and a photovoltaic power prediction model based on long short-term memory neural network(LSTM)were established.The specific contents are as follows:Based on the historical power and NWP data of photovoltaic power plants,the Pearson correlation coefficient is used for correlation analysis,and the irradiance,temperature,and humidity in the NWP data are determined as the input of the prediction model,a photovoltaic power prediction model is constructed by establishing the mapping relationship between historical NWP data and historical photovoltaic power.Aiming at the problem of unstable model prediction performance caused by the random generation of initial weights and biases during extreme learning machine training,a prediction model based on particle swarm optimization optimized extreme learning machine(PSO-ELM)is constructed.In order to verify the feasibility of the established PSO-ELM model,three comparison models of BP,ELM and SVM were set up,and the prediction results of each model under three different weather conditions were analyzed to show the prediction accuracy of the PSO-ELM model is highest.Considering that photovoltaic power data is a kind of time series data,it is not only related to meteorological factors,but also has a strong autocorrelation with historical power,in the short-term power prediction,introducing historical power as model input will improve the prediction accuracy of the model.Utilizing the excellent performance of long short-term memory neural network(LSTM)in dealing with time series prediction problems,a power prediction model based on LSTM is built.Through power autocorrelation analysis,it is determined that the historical power data2 hours before the prediction point is used as the model’s input to construct the prediction model.Two prediction schemes are set up according to the needs of different prediction occasions.The first scheme uses NWP and historical power data as the input of the model,which is suitable for photovoltaic power plants with meteorological observatories that can obtain NWP data.The second scheme uses historical power data as the input of the model,which is suitable for photovoltaic power plants that cannot obtain NWP data.In order to verify the feasibility of the LSTM model established in this paper,two comparative models of SVM and PSO-ELM are established,and the prediction results of the two schemes under three different weathers are compared,the results show that the LSTM model has a higher prediction accuracy.