基于ISSA-BP神经网络的光伏阵列故障诊断方法  

Fault Diagnosis Method for Photovoltaic Array Based on ISSA-BP Neural Network

作　　者：文力 谭功全 毛国斌 王旭东 庞宏杰 WEN Li;TAN Gongquan;MAO Guobin;WANG Xudong;PANG Hongjie(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川宜宾644000

出　　处：《四川轻化工大学学报(自然科学版)》2025年第1期57-68,共12页Journal of Sichuan University of Science & Engineering(Natural Science Edition)

基　　金：人工智能四川省重点实验室科研项目(2019RYJ08)。

摘　　要：针对反向传播神经网络(BPNN)在光伏阵列故障诊断中存在收敛速度慢、易陷入局部最优解、故障诊断准确率低等问题,提出一种改进麻雀搜索算法(ISSA-BP)优化BP神经网络的权值和阈值。首先,使用Cubic混沌映射,提高种群初始位置的空间覆盖率;然后,在发现者中引入惯性权重,加快收敛速度,并增强局部搜索的能力;最后,通过动态调整预警者的数量来维持多样性,从而强化全局搜索的能力。利用MATLAB/Simulink仿真模型,获取光伏阵列在正常状态和故障状态下的短路电流、开路电压、最大功率点电流和最大功率点电压,共4个特征参数,并将得到的特征参数分别输入到6种故障诊断模型。通过与传统的BP、GA-BP、PSO-BP、SSA-BP、SOA-SVM模型进行对比验证,实验结果表明,ISSA-BP模型不仅能够快速跳出局部最优解,加快收敛速度,且故障诊断准确率能达到97.5%。In response to the issues of slow convergence speed,susceptibility to local optima,and low accuracy in fault diagnosis of photovoltaic arrays using back propagation neural network(BPNN),an improved sparrow search algorithm(ISSA-BP)for optimizing the weights and thresholds of the BP neural network has been proposed.Firstly,the Cubic chaotic mapping is employed to enhance the spatial coverage of the initial population positions.Subsequently,an inertia weight is introduced among the discoverers to accelerate convergence speed and strengthen local search capabilities.Finally,the diversity is maintained by dynamically adjusting the number of scouts,enhancing global search capabilities.The MATLAB/Simulink simulation model is utilized to obtain four feature parameters,namely short-circuit current,open-circuit voltage,maximum power point current,and maximum power point voltage,under normal and fault conditions in a photovoltaic array,which is input into six fault diagnosis models.Comparative verification with traditional BP,GA-BP,PSO-BP,SSA-BP,and SOA-SVM models is conducted.The experimental results demonstrate that the ISSA-BP model not only rapidly escapes local optima and accelerates convergence speed but also achieves a fault diagnosis accuracy of 97.5%.

关 键 词：光伏阵列 故障诊断 反向传播神经网络 故障特征提取 改进麻雀搜索算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TM615[自动化与计算机技术—控制科学与工程]