Hebei University of Water Resources and Electric E

Hebei University of Water Resources and Electric E ›› 2025, Vol. 35 ›› Issue (3): 22-29.DOI: 10.16046/j.cnki.issn2096-5680.2025.03.004

• Artificial Intelligence and Robotics • Previous Articles     Next Articles

Identification of Maintenance Tools and Instruments for Hydropower Plants Based on YOLOv5 Algorithm

CHEN Tiehua1, WU Guangxin1, XU Ming1, WANG Yiyao2, YANG Zhiwen1   

  1. 1. Scool of Energy and Power Engineering, Changchun Institute of Technology, 130103, Changchun, Jilin, China;
    2. Liaoning Qingyuan Pumped Storage Co., Ltd., 113300, Fushun, Liaoning, China
  • Received:2024-09-22 Revised:2024-10-23 Online:2025-09-30 Published:2025-10-09

基于YOLOv5算法的水电厂检修工器具识别系统研究

陈铁华1, 吴广新1, 许明1, 王艺瑶2, 杨智文1   

  1. 1.长春工程学院能源与动力工程学院,吉林省长春市朝阳区宽平大路395号 130103;
    2.辽宁清原抽水蓄能有限公司,辽宁省抚顺市清原满族自治县清原镇浑河南路22号 113300
  • 通讯作者: 吴广新(1996-),男,吉林长春人,在读硕士,主要研究方向:远程监测及自动控制,智慧电厂。E-mail:wuguangxin0719@163.com
  • 作者简介:陈铁华(1972-),女,吉林省吉林市人,教授,研究方向:远程监测及自动控制,智慧电厂。E-mail:0211024@ccit.edu.com
  • 基金资助:
    吉林省科技厅重点研发项目(20230203154SF);国网吉林省电力有限公司松原供电公司项目(SGJLSYO0KJS2200906)

PDF

Abstract: With the introduction of the concept of smart power plants, the traditional management mode of hydropower plants is gradually transitioning towards a smart management system. Based on the current management status of maintenance tools in hydropower plants, and combined with the rapid development of deep convolutional neural network technology in the field of image recognition in recent years, especially with the increasingly mature object detection technology, a tool recognition system based on YOLOv5 neural network model is proposed, aiming to achieve accurate recognition of maintenance tools in hydropower plants. The system utilizes calibration tool data. And the collected data is divided into training set, validation set, and testing set, which are used for neural network training, model validation, and performance evaluation, respectively. The training results show that the system displays excellent recognition ability with precision (P), recall (R), and mean average values (mAP) of 90.3%, 72%, and 83.4%, increased by 2.1%, 1.6% and 1.3%, respectively compared to the original model. Further testing is conducted on the trained network using the validation set, and the test results show that the system can accurately identify the types of tools, automatically generating visualized anchor frames during the recognition process. And it accurately displays the coordinates and confidence of anchor frames. This feature not only improves the efficiency of access and retrieval work, but also enhances the accuracy and convenience of tool management. The system has demonstrated high recognition accuracy and stability, providing strong technical support for the intelligent management of hydropower plants.

Key words: hydroelectric plant, Yolov5, algorithm model, tool recognition

摘要： 随着智慧电厂概念的提出,水电厂的传统管理模式正逐步向智慧管理体系转型。针对当前水电厂检修工器具的管理现状,结合近年来图像识别领域深度卷积神经网络技术的飞速发展,特别是目标检测技术的日益成熟,提出了一种基于YOLOv5神经网络模型的工器具识别系统,旨在实现水电厂检修工器具的准确识别。该系统利用标定工器具数据,收集的数据被分为训练集、验证集和测试集,分别用于神经网络的训练、模型验证和性能评估。通过Mosaic数据增强技术和锚框优化改进,训练结果显示,该系统在精确率(P)、召回率(R)、平均精度值(mAP)分别达到90.3%、72%和83.4%,相较于原模型分别提升了2.1%、1.6%和1.3%,展现了出色的识别能力。进一步通过验证集对训练后的网络进行测试,测试结果表明,该系统能够精准识别工器具的种类,在识别过程中自动生成可视化锚框,并能准确显示锚框角点坐标信息和置信度,这一特性不仅提高了领存取工作的效率,还增强了工器具管理的准确性和便捷性。系统展现了较高的识别精度和稳定性,为水电厂的智能化管理提供了有力的技术支持。

关键词: 水电厂, YOLOv5, 算法模型, 工器具识别

CLC Number: