Research On Short-term Photovoltaic Output Power Forecasting Based On The Similar Day Clustering And Multi-model Fusion

Solar energy is bound to replace traditional fossil energy in the future and is widely used in energy supply in various fields.The high-performance photovoltaic output power prediction model plays a vital role in improving the economy of grid dispatching,maintaining the safety of system operation,and maintaining the balance of supply and demand in the power market.In order to further improve the accuracy and robustness of the short-term prediction of photovoltaic output power,and to ensure that the prediction model can show good prediction performance in different seasons and climatic conditions,this paper proposes a method based on similar day clustering and multi-model fusion Short-term photovoltaic output power prediction modelFirst,in order to avoid the influence of abnormal points in the historical sample data set on the prediction performance of the proposed model,anomalous data identification and data reconstruction models were established to realize the effective identification of abnormal data.By analyzing and mining the internal relationship between the output characteristics of the photovoltaic power generation system and the output power and the influencing factors,comprehensively comparing the effectiveness of several types of abnormal data identification algorithms,an abnormal data identification method that integrates the quartile method and the K proximity algorithm is proposed.,And select the K proximity algorithm to reconstruct the data of the suspected abnormal sample points.After verification,the proposed model can better realize the identification and reconstruction of abnormal samples.Secondly,considering the impact of different climate environments on the comprehensive performance of the prediction model,a similar day clustering model based on multiple evaluation indicators was established.At the same time,considering the three indicators of Person coefficient,Spearman coefficient and XGBoost feature importance score,the input features of the clustering model in different seasons are screened;the performance of the three types of single-layer clustering models is compared from the perspectives of clustering accuracy and computing efficiency.The fuzzy C-means algorithm(FCM)is selected as the basis of the similar day clustering model,and the original data set is divided into different weather types.After that,considering the influence of the future solar radiation intensity on the prediction results of photovoltaic output power,and further improving the comprehensive performance of the daily solar irradiance prediction model,a bi-layer collaborative prediction model based on multi-dimensional feature analysis is proposed.Through the establishment of a two-layer collaborative prediction framework that includes the base layer and the promotion layer,numerical weather prediction(NWP)is selected as the input of the base layer,and LightGBM is used to construct the reference prediction model in the feature learning mode;the reference prediction value and processing The latter target training set is used as the sample input,and the Multiple Hidden Layer Extreme Learning Machine(MH-ELM)combined with the improved AdaBoost algorithm is introduced into the promotion layer;the actual solar radiation of a photovoltaic power station in the central region of my country is selected The data is simulated by a numerical example,which verifies the rationality and effectiveness of the model.Finally,based on the results of the above two types of models,an adaptive combined forecasting model considering similar day clustering is proposed.For the divided training sets of samples,WNN is selected as the base learner for the improved AdaBoost framework,and the wavelet factors and inter-layer weights of WNN are optimized by the improved hybridizing grey wolf optimization(IHGWO)algorithm;the same applies to China The actual output power data of a photovoltaic power station in the central region is simulated by a calculation example,and the comprehensive comparison with other models is carried out.The experimental results further confirm that in different seasons and weather types,even if compared with the IHGWO-WNN model with the same excellent comprehensive performance,the proposed The average prediction error of the model is only 63.7%of the former;it is verified that the model has strong adaptability and robustness while effectively improving the accuracy of photovoltaic short-term output power prediction.