云边协同框架下结合深度学习与随机森林的电力设备识别

Power Equipment Identification Combined with Deep Learning and Random Forest under the Cloud-side Collaboration Framework

针对变电站巡检系统中电力设备识别准确率不高和识别时间过长的问题,提出了一种部署在云边协同框架下结合深度学习和随机森林的图像识别方法。该方法使用初始标记图像在云端完成随机森林分类器和改进卷积神经网络的训练;将完成训练的分类器下发至部署于变电站处的边缘计算设备以完成电力设备的本地识别;将完成识别的图像异步上传至云端以迭代更新优化分类器和改进卷积神经网络。通过不同算法和不同运行环境的对比测试检验了所提方法的性能。结果表明,所提方法的电力设备平均识别精度为86.6%,优于基于传统卷积神经网络和Softmax分类器的图像识别算法,所提方法运行于云边协同框架的平均运行时间比运行于单机运行环境的缩短约33.7%。

Aiming at the problems of low recognition accuracy and long recognition time of power equipment in the substation inspection system,an image recognition method that combines deep learning and random forest deployed under the cloud-side collaboration framework is proposed.This method uses the initial labeled image to complete the training of the random forest classifier and the improved convolutional neural network in the cloud,and delivers the trained classifier to the edge computing device deployed at the substation to complete the local identification of the power device.The recognized images are asynchronously uploaded to the cloud to iteratively update and optimize the classifier and improve the convolutional neural network.The performance of the proposed method is verified through comparative tests of different algorithms and different operating environments.The results show that the average recognition accuracy of the proposed method for power equipment is 86.6%,which is better than image recognition algorithms based on traditional convolutional neural networks and Softmax classifiers.The average running time of the proposed method running on a cloud-side collaborative framework is longer than running on a stand-alone machine.The operating environment has been shortened by approximately 33.7%.