Research On Photovoltaic Power Generation Power Prediction Based On PWVRSM Algorithm

Photovoltaic power generation is significantly affected by meteorological and environmental factors,with periodic,intermittent and random fluctuations.Large scale photovoltaic grid connection will have a huge impact on the security and stability of power grid.Therefore,accurate and real-time photovoltaic power prediction is conducive to estimate the fluctuation range of power generation and update the control instructions in the process of grid connection,reduce the occurrence of residual current,islanding effect,harmonic pollution and other problems affecting power quality,and ensure the safe,stable and efficient operation of the grid after photovoltaic integration.In order to explore the role of generation data quality and fusion model in shortterm photovoltaic power prediction,and improve the prediction performance of the model,this paper constructs a short-term photovoltaic power prediction model based on feature selection and PWVRSM algorithm.The following is the specific work of this paper:Firstly,the advantages and disadvantages of existing research on photovoltaic active power output prediction are analyzed;secondly,based on historical photovoltaic power generation data and historical meteorological data,the internal mechanism of various meteorological factors affecting photovoltaic power output is deeply explored and analyzed;the analysis shows that season type,irradiance,temperature and humidity have an important impact on the fluctuation characteristics of power generation.At the same time,we construct two features,namely dis2 peak and first-order difference of wind speed,to capture the periodicity and randomness of the data,and use them as the input features of the prediction model.Secondly,a photovoltaic power prediction model based on Stacking framework is built.Aiming at the blindness and variability of primary learners selection in the process of Stacking framework fusion,the process of learners selection is added to select the learners that can be used for fusion.Considering the differences of each algorithm,linear regression algorithm,random forest algorithm,support vector regression algorithm and XGBoost algorithm are selected to build photovoltaic power prediction models,and the models are evaluated and selected according to the prediction results of the models;Light GBM algorithm is used to fuse the best three of the four models(random forest model,XGBoost model and support vector regression model)to build the Stacking integrated model,so as to improve the prediction accuracy of the model.Thirdly,an improved Stacking model with precision weighting and new vector representation(PWVRSM)is proposed.In this paper,a three-layer algorithm structure is adopted.Considering that the different prediction models generated by the same base learner show different prediction accuracy under cross validation,different prediction results are weighted.In the first layer,the original training set is trained;in the second and third layers,a new vector representation is used to increase the training data,so that the size of the input vector of the individual regressors in the second and third layers will not increase with the increase of the number of regressors;in the third layer,the individual regressor is used to learn the learning results of the previous layer again to reduce the noise.In order to intelligently select individual regressors in PWVRSM algorithm,sparrow search algorithm is used to optimize their combination.The accuracy of using Stacking ensemble learning model to predict the output of photovoltaic power generation is higher than that of single prediction model,and the fitting effect is better.The improved 2-layer and 3-layer Stacking algorithm further improves the prediction performance compared with the classical Stacking algorithm,and the 3-layer model performs better.The MSE of the 3-layer model is 14.93% lower than that of the 2-layer model,and the R-squared is 1.26% higher than that of the 2-layer model.At the same time,the model optimized by sparrow search algorithm has better prediction effect,which indicates that the intelligent algorithm can be used to adjust parameters in the optimization of PWVRSM algorithm.