Research On Short-term And Ultra-short-term Power Generation Prediction Methods For Photovoltaic Power Plants

With the rapid economic development,energy demand is increasing.Photovoltaic power generation has the advantages of green environmental protection and renewable energy.The new energy utilization method of photovoltaic power generation worldwide has shown steady growth;however,the power output of photovoltaic power generation systems is affected by many factors and shows a high degree of volatility.Especially when large-scale photovoltaics are connected to the grid,the safe operation and power quality of the power system will be greatly affected.Therefore,photovoltaic forecasting technology is of great significance to grid planning and stable operation.In this paper,after in-depth analysis of photovoltaic power generation principles and power influencing factors,a hybrid nuclear SVR short-term photovoltaic power prediction method based on feature hierarchical clustering and a VMD-DESN-MSGP model are proposed for the two cases with and without meteorological data.Ultra-short-term photovoltaic power prediction method.For photovoltaic power stations with meteorological observation stations,this paper proposes a hybrid nuclear SVR photovoltaic power prediction method based on feature hierarchical clustering.First,the idea of hierarchical partitioning is used,and historical data is divided by two clusters using different feature factors.Go to different weather categories,and then construct a hybrid kernel function SVR prediction model,strengthen the generalization ability of the prediction model to deal with different weather conditions,and use uncertain knowledge particle swarm optimization to optimize the model parameters.Optimization,and finally use KNN to classify the forecast day into different weather types,and use the forecast model corresponding to the weather condition to predict the power of the forecast day to obtain the photovoltaic power forecast power.The simulation results show that the effect of this method in different weather types is superior to the traditional clustering and single kernel function SVR prediction model,even in complex weather conditions,it can still maintain a high prediction accuracy.For photovoltaic power stations lacking meteorological observation stations,the inability to obtain photovoltaic meteorological data is a major problem.In the absence of meteorological data,this paper proposes a combination of variational mode decomposition(VMD),deep echo state network(DESN)and Sparse Gaussian Mixture Process Expert Model(MSGP)ultra short-term photovoltaic power prediction method.First,the VMD is used to decompose the photovoltaic power time series into different modes,which reduces the non-stationarity of the data;then,in order to improve the prediction ability of the model at ultra-short time series,a DESN prediction model is established for each mode,and each mode is Summing and reconstructing the prediction results;then in order to further improve the prediction accuracy of the model,analyze the characteristics of the error,and use MSGP to re-predict the prediction error to compensate the prediction result;finally,the predicted value of the error and the predicted value of the original power Superimposed as the final prediction result.The simulation results show that this method has a better effect on photovoltaic power time series prediction than traditional prediction models,and effectively improves the accuracy of ultra-short-term photovoltaic power time series prediction.