准最小二乘的神经网络在光伏MPPT中的应用  

Quasi-least Square Based Neural Network for MPPT of Photovoltaic Power System

作　　者：张海洲[1] 张正江[1] 胡桂廷 陈倩[1] 闫正兵[1] 郑崇伟[1] ZHANG Hai-zhou;ZHANG Zheng-jiang;HU Guiting;CHEN Qian;YAN Zheng-bing;ZHENG Chong-wei(National-Local Joint Engineering Laboratory of Electrical Digital Design Technology,Wenzhou University,Wenzhnu 325035,China)

机构地区：[1]温州大学电气数字化设计技术国家地方联合工程实验室

出　　处：《控制工程》2018年第12期2257-2262,共6页Control Engineering of China

基　　金：国家自然科学基金项目(61703309);浙江省科技计划项目(2015C31157;2014C31074;2014C31093);浙江省大学生科技创新活动计划暨新苗人才计划(2017R426019)

摘　　要：传统的神经网络以最小二乘(LS)为学习函数,对训练数据的准确性有较高要求。考虑存在测量误差的训练数据对传统神经网络的影响,提出了一种基于准最小二乘的神经网络(QLS-NN)并应用于光伏发电系统的最大功率点跟踪(MPPT)上。根据光伏电池的内部结构和伏安特性建立其数学模型。根据模型所反映的规律,将温度和照度作为输入变量,最大功率与对应的电压作为输出变量,构建了用于MPPT的QLS-NN。神经网络训练后对最大功率点进行预测与跟踪。仿真结果表明QLS-NN具有较高的鲁棒性,可显著提高了光伏发电系统MPPT的精度。The traditional neural network takes the least square(LS) as the learning function, which requires the high accuracy in the training data. This paper considers the influence of the training data with measurement errors on the traditional neural network, and then proposes quasi-least squares based neural networks(QLS-NN) for applications in maximum power point tracking(MPPT) of photovoltaic power systems. Considering the internal structure and volt ampere characteristics of photovoltaic cells, the mathematical model is established. According to the law of the PV model, the temperature and illumination are considered as the input variables, the maximum power and the corresponding voltage are used as the output variables, and the QLS-NN for MPPT is constructed. Prediction and tracking of maximum power points can be achieved after training the neural network. Simulation results show that the QLS-NN has high robustness, and the accuracy of MPPT of the photovoltaic power system can be improved significantly.

关 键 词：光伏发电系统 最大功率点跟踪 测量误差 准最小二乘 神经网络 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]