基于混合深度学习的变电站巡检机器人目标识别算法研究  

Research on Target Recognition Algorithm for Substation Inspection Robot Based on Hybrid Deep Learning

作　　者：戴瑞成 董翔[1] 赵璧[1] 张潇 葛东阳 秦彬 杨金龙 DAI Ruicheng;DONG Xiang;ZHAO Bi;ZHANG Xiao;GE Dongyang;QIN Bin;YANG Jinlong(State Grid Beijing Electric Power Company Maintenance Branch,Beijing 100073,China;College of Electrical and New Energy,China Three Gorges University,Yichang 443000,China)

机构地区：[1]国网北京市电力公司检修分公司,北京100073 [2]三峡大学电气与新能源学院,湖北宜昌443000

出　　处：《智慧电力》2025年第3期117-123,共7页Smart Power

基　　金：国家自然科学基金资助项目(52107108)。

摘　　要：为提升变电站智能巡检中图像识别的准确性与效率,提出一种基于改进萤火虫算法的混合深度学习方法,用于变电站巡检机器人的目标识别。首先,提出了一种改进的萤火虫算法,通过动态调整随机参数和光强衰减系数,实现卷积神经网络超参数的全局优化,显著提升了模型的收敛速度和精确性;然后,将基于改进萤火虫算法优化的卷积神经网络与支持向量机相结合,利用卷积神经网络对巡检图像进行高层次特征提取,再将提取的特征输入到支持向量机分类器中完成图像分类,实现了对巡检图像的高效、精确识别。仿真结果表明,所提方法在各类故障检测任务中具有良好的识别效果。To improve the accuracy and efficiency of image recognition in substation intelligent inspection,the paper proposes a hybrid deep learning method based on an improved firefly algorithm for target recognition in substation inspection robots.Firstly,an improved firefly algorithm is introduced,which dynamically adjusts random parameters and light intensity attenuation coefficients to achieve the global optimization of convolutional neural network(CNN) hyperparameters,significantly enhancing the convergence speed and accuracy of the model.Secondly,the optimized CNN based on the improved firefly algorithm is combined with a support vector machine(SVM).The CNN is used for the high-level feature extraction of inspection images,and the extracted features are fed into the SVM classifier to complete image classification,enabling the efficient and accurate recognition of the inspection images.Simulation results shows that the proposed method achieves excellent recognition performance in various fault detection tasks.

关 键 词：卷积神经网络 巡检机器人 识别方法 萤火虫算法 

分 类 号：TM732[电气工程—电力系统及自动化] TP391[自动化与计算机技术—计算机应用技术]