摘要
由于光伏阵列常年处于恶劣的环境中,光伏组件时常发生故障。用深度信念网络(deep belief network,DBN)模型进行光伏组件故障诊断时,由于权重和偏置初始化的随机性,导致模型在训练和学习的过程中易陷入局部最优且收敛速度缓慢,因此提出麻雀搜索算法(sparrow search algorithm,SSA)优化深度信念网络权重和偏置的故障诊断方法。首先,通过SSA算法对DBN网络的可见层权值进行编码;其次,采用适应度函数对动量参数进行优化,以减少训练过程中的误差;最后,不断更新种群的速度和位置,以寻求个体最优和全局最优。实验分别与传统DBN网络和深度卷积神经网络(DCNN)的诊断准确率及重构误差两个方面进行了对比分析,结果证明该优化DBN网络增强了网络的泛化能力,提高了光伏故障诊断的识别精度。
Because the photovoltaic array is in a bad environment all year round,photovoltaic modules often fail.When deep belief network(DBN)model is used for photovoltaic module fault diagnosis,due to the randomness of traditional DBN network weight and bias initialization,the model is easy to fall into local optimum and slow convergence speed in the process of training and learning.Therefore,sparrow search algorithm(SSA)was proposed to optimize the weight and bias of deep belief network.Firstly,the visible layer weight of DBN network was encoded by SSA algorithm;secondly,the fitness function was used to optimize the momentum parameters to reduce the error in the training process;finally,the velocity and position of the population were updated to seek the individual optimal and global optimal.The experimental results were compared with the traditional DBN network and deep convolutional neural network(DCNN)in terms of diagnostic accuracy and reconstruction error.The results show that the optimized DBN network enhances the generalization ability of the network and improves the recognition accuracy of photovoltaic fault diagnosis.
作者
姜萍
郭欢欢
代金超
JIANG Ping;GUO Huanhuan;DAI Jinchao(College of Electronic Information Engineering,Hebei University,Baoding Hebei 071000,China)
出处
《电源技术》
CAS
北大核心
2022年第8期925-929,共5页
Chinese Journal of Power Sources
基金
河北省自然科学基金重点项目(A2020201021)。
