发电技术 ›› 2025, Vol. 46 ›› Issue (1): 72-82.DOI: 10.12096/j.2096-4528.pgt.23117

• 新能源 • 上一篇    下一篇

基于改进灰狼算法优化极限学习机的光伏阵列故障诊断方法研究

杨琛, 牛锋杰, 韩茂林, 周宁, 周定璇   

  1. 上海海洋大学工程学院,上海市 浦东新区 201306
  • 收稿日期:2023-12-06 修回日期:2024-03-01 出版日期:2025-02-28 发布日期:2025-02-27
  • 作者简介:杨琛(1978),女,博士,副教授,研究方向为物联网大数据和智能控制技术、物联网技术、渔光互补技术,cyang@shou.edu.cn
    牛锋杰(1999),男,硕士研究生,研究方向为光伏系统故障诊断,m210811396@st.shou.edu.cn
    韩茂林(1999),男,硕士研究生,研究方向为光伏系统故障诊断,maolinH0812@163.com;
    周宁(2000),男,硕士研究生,研究方向为石化设备故障诊断,ningz281231@163.com
    周定璇(1997),男,硕士研究生,研究方向为基于边缘智能的故障诊断,594280963@qq.com
  • 基金资助:
    上海市崇明区科委2023年度可持续发展科技创新行动计划项目(CKST2023-01)

Research on Fault Diagnosis Method of Photovoltaic Arrays Based on Improved Grey Wolf Algorithm Optimized Extreme Learning Machine

Chen YANG, Fengjie NIU, Maolin HAN, Ning ZHOU, Dingxuan ZHOU   

  1. College of Engineering Science and Technology, Shanghai Ocean University, Pudong New District, Shanghai 201306, China
  • Received:2023-12-06 Revised:2024-03-01 Published:2025-02-28 Online:2025-02-27
  • Supported by:
    Chongming District Science and Technology Commission 2023 Sustainable Development Science and Technology Innovation Action Plan Project(CKST2023-01)

摘要:

目的 光伏阵列在复杂室外工作条件下,发生的故障类型多样且程度不同,为了判断光伏阵列的工作状态,提出一种基于改进灰狼算法优化极限学习机(improved grey wolf optimized extreme learning machine,IGWO-ELM)的故障诊断方法。 方法 首先,针对9种故障仿真输出特性进行分析,建立了由短路电流、开路电压、最大功率点电流、最大功率点电压、填充因子组成的5维故障特征向量。其次,针对灰狼算法初始位置分布不均匀、全局搜索和局部开发过程不均衡的缺点,引入Circle映射和非线性收敛因子,提出一种改进的灰狼优化算法,优化极限学习机的输入层权重和隐含层节点偏置,以提高算法性能。最后,搭建仿真模型和实验平台并获取故障数据,基于K折交叉验证对数据集进行划分,代入IGWO-ELM模型进行正确率验证,并与其他算法模型进行对比。 结果 IGWO-ELM模型对光伏阵列不同故障具有较高的识别率,对仿真数据和实验数据的分类正确率分别达到98.32%和95.48%。 结论 基于IGWO-ELM的故障诊断方法识别率高,迭代次数少,收敛速度快,可有效判断光伏阵列的工作状态。

关键词: 太阳能发电, 光伏阵列, 故障诊断, 改进灰狼优化(IGWO)算法, 极限学习机(ELM), K折交叉验证, 特征提取, 仿真

Abstract:

Objectives Photovoltaic arrays operating under complex outdoor conditions encounter various fault types with varying degrees of severity. To accurately assess the working status of photovoltaic arrays, a fault diagnosis method based on an improved grey wolf optimized extreme learning machine (IGWO-ELM) is proposed. Methods Firstly, nine fault simulation output characteristics are analyzed, and a five-dimensional fault feature vector is established, consisting of short-circuit current, open-circuit voltage, maximum power point current, maximum power point voltage, and fill factor. Secondly, to address the limitations of the grey wolf algorithm, such as uneven distribution of initial position and imbalance between global search and local exploitation, Circle mapping and nonlinear convergence factors are incorporated. An improved grey wolf optimization algorithm is then proposed, which optimizes the input layer weights and hidden layer node biases of the extreme learning machine to improve performance. Finally, simulation models and experimental platforms are developed to collect fault data, which are divided using K-fold cross validation. The data are input into the IGWO-ELM model for accuracy verification and compared with other algorithms. Results The IGWO-ELM model demonstrates high recognition rates for various fault types in photovoltaic arrays, achieving classification accuracy of 98.32% and 95.48% for simulation and experimental data, respectively. Conclusions The fault diagnosis method based on IGWO-ELM offers high accuracy, requires fewer iterations, and achieves fast convergence speed, effectively judging the working state of photovoltaic arrays.

Key words: solar power generation, photovoltaic arrays, fault diagnosis, improved grey wolf optimization (IGWO) algorithm, extreme learning machine (ELM), K-fold cross validation, feature extraction, simulation

中图分类号: