To address the problem of low fault recognition rate caused by the small and imbalanced fault data of photovoltaic arrays,an improved fault diagnosis method is proposed by combining the improved synthetic minority oversampling technique(SMOTE)and the regularized extreme learning machine(RELM)optimized by an improved gold rush optimization algorithm(IGRO).First,to solve the overfitting problem caused by traditional SMOTE,a hybrid oversampling method is presented by combining adaptive synthetic(ADASYN)and borderline oversampling(Borderline-SMOTE).Second,the Gold Mining Optimization algorithm(GRO)is improved by integrating Circle chaotic mapping,adaptive weights,somersault foraging strategy,and Cauchy Gaussian mutation perturbation strategy to improve the convergence accuracy of GRO.Finally,the IGRO is used to optimize the weights and biases of RELM to construct the IGRO-RELM model.The simulation results show that compared with other models,the IGRO-RELM model has the best fault diagnosis performance with an accuracy of 94.29％.
维普
万方数据