Fault Diagnosis Method Of Bagging-IGWO-SVM For Photovoltaic System

Photovoltaic(PV)power generation technology,as a significant field in the realm of renewable energy,has reached a high level of maturity.However,it also poses greater demands on the maintenance of PV equipment.The PV module,being a crucial component of the PV power generation system,requires safe and stable operation for effective equipment maintenance.This thesis proposes a combined algorithm aimed at enhancing the model’s generalization ability and applies it to the field of PV array fault diagnosis.The objective is to address the issue of significant variance between the training set and the test set in the PV equipment fault dataset,which often leads to overfitting.This research primarily focuses on the following contributions and innovations:Firstly,to tackle the problem of incomplete inclusion of feature distributions in small datasets,which commonly results in model overfitting,a combination of the bagging algorithm and support vector machine(SVM)is introduced.The bagging algorithm’s data sampling method is modified to improve the independence of its sub-models,reduce model variance,and enhance the confidence of the integrated model through Platt calibration on the SVM’s output.Additionally,the dimensionality of the dataset is reduced using the Linear Discriminant Analysis(LDA)method to achieve a model that balances accuracy and generalization.Platt calibration is evaluated using randomly generated data,and the performance of the classification algorithm is verified using a publicly available dataset.Results indicate that the improved bagging support vector machine ensemble model surpasses both the pre-modified bagging support vector machine model and the single support vector machine model.Secondly,to address the challenge of determining the parameters of the combined model,an improved gray wolf algorithm is proposed to optimize the combination method of hyperparameters for each sub-model of the bagging support vector machine integrated model.This optimization aims to reduce sub-model bias and enhance the combined model’s performance.The improved gray wolf algorithm incorporates the Latin Hypercube Sampling(LHS)population initialization method,nonlinear convergence factor,and dynamic position updating strategy to enhance the algorithm’s global optimization capabilities.The effectiveness of the algorithm is verified through experiments on benchmark functions,which demonstrate improved performance compared to the original gray wolf algorithm.Additionally,experiments are conducted on the classification algorithm optimized by the improved gray wolf algorithm using a randomly generated classification dataset,confirming the optimization algorithm’s performance improvement effect on the combined model.Finally,considering the output characteristics of PV arrays,a PV array simulation model is developed to analyze the impact of solar irradiance and temperature changes on array output characteristics.Furthermore,simulations of various PV array faults are performed to analyze and compare the resulting output characteristics,enabling the identification of essential features for PV fault diagnosis.The proposed combined algorithm is validated using a self-developed PV microgrid system,where experiments are designed and real data is collected under different fault operating conditions.By training and classifying the fault data using the proposed combined model,the classification results are compared with eight other algorithms,including logistic regression,support vector machine,K-nearest neighbor,and decision tree.The combined model achieves the highest F1 value,accuracy rate,and recall rate index in the comparison algorithm,thereby demonstrating superior fault diagnosis effectiveness.