★ 国网上海市电力公司信息通信公司 刘逸逸

摘要：为实现电力工程失稳状态的远程、精准校核，本文提出了一种基于改进卷积神经网络的电力工程数字化校核技术。该技术架构分为校核层、站控层、间隔层：在间隔层，通过多组振动传感器采集电力工程的电气设备振动信号，并采用经验模态分解算法对原始信号去噪；在站控层网关支持下，将去噪后的信号发送至校核层；校核层通过基于改进卷积神经网络的电力工程电气设备异常校核方法，完成去噪后电气设备振动信号的特征提取与融合，并启动Softmax分类器识别设备异常工况，以此判断电力工程是否存在失稳情况，实现电力工程数字化智能校核。实验结果表明，本文所提技术可远程、精准完成电力工程失稳风险校核，校核性能可满足实际应用要求。

关键词：卷积神经网络；电力工程；电气设备；数字化；校核技术；特征提取

Abstract: To remotely and accurately assess instability in power engineering systems, this paper proposes a digital verification technology based on an improved convolutional neural network. The technical architecture is divided into a verification layer, a station control layer, and an bay layer. The bay layer collects vibration signals of electrical equipment in power engineering using multiple sets of vibration sensors, and then employs a signal denoising method based on empirical mode decomposition to filter out noise information mixed into the signals. With the support of the gateway in the station control layer, the denoised signals are transmitted to the verification layer. At the verification layer, an electrical-equipment anomaly verification method based on the improved convolutional neural network is used to extract and fuse features from the denoised vibration signals. A Softmax classifier is then used to identify abnormal operating conditions and determine whether the power engineering system is unstable, thereby completing digital intelligent verification. Experimental results show that the proposed technology can remotely and accurately verify whether instability exists in power engineering, and its digital verification capability meets application requirements.

Key words: Convolutional neural network; Power engineering; Electrical equipment; Digital verification; Verification technology; Feature extraction

在线预览：基于改进卷积神经网络的电力工程数字化校核技术研究.pdf

摘自《自动化博览》2026年6月刊