基于CNN和LSTM结合的电网网络攻击检测  

Power Grid Network Attack Detection Based on the Combination of CNN and LSTM

作　　者：吕首琦 海涛 郑茂兴 LÜShouqi;HAI Tao;ZHENG Maoxing(China Energy Engineering Group Shaanxi Electric Power Design Institute Co.,Ltd.,Xi’an Shaanxi 710054,China;School of Computer and Information,Qiannan Normal University for Nationalities,Duyun Guizhou 558000,China;School of Computer Sciences,Baoji University of Arts and Sciences,Baoji Shaanxi 721007,China)

机构地区：[1]中能工程集团陕西电力设计院有限公司,陕西西安710054 [2]黔南民族师范学院计算机与信息学院,贵州都匀558000 [3]宝鸡文理学院计算机科学学院,陕西宝鸡721007

出　　处：《电子器件》2023年第3期824-830,共7页Chinese Journal of Electron Devices

摘　　要：为应对电力系统网络攻击检测面临的挑战,开发了一种基于深度学习的模型。该模型采用增强自适应弹性网络进行电力数据的特征提取,以增强数据的灵敏性并提高模型的训练和分类能力。此外,采用归一化和粒子群优化-K均值(PSO-K均值)噪声数据处理技术,以提高模型对噪声数据的适应性并缓解过拟合问题。采用基于CNN和LSTM的多层集成学习模型对噪声数据进行训练,从而提高分类器的准确性。在验证阶段,与K最近邻(KNN)、随机森林(RF)、支持向量机(SVM)、卷积神经网络(CNN)和其他模型相比,多层集成分类器表现出更优异的性能。值得注意的是,最佳分类器的准确率达到了88.91%。该模型的有效性对于指导电力系统的稳定性和安全管理具有重要意义。To address the challenges facing power system network attack detection, a deep learning-based model has been developed. The model employs an enhanced adaptive elastic network for the feature extraction of electric power data to enhance the sensitivity of the data and improve the model's training and classification capability. Additionally, normalization and particle swarm optimization-K-means(PSO-K-means)noise data processing techniques are employed to improve the model's adaptability to noise data and alleviate the over-fitting issue. A multi-layer ensemble learning model based on CNN and LSTM is utilized to train the noisy data, thereby increasing the classifier's accuracy. In comparison to K-nearest neighbor(KNN),random forest(RF),support vector machine(SVM),convolutional neural network(CNN),and other models during the verification phase, the multi-layer integrated classifier demonstrates superior performance. Notably, the accuracy of the best classifier reaches 88.91%. The model's effectiveness is significant for guiding power system stability and safety management.

关 键 词：电力系统 深度学习 攻击检测 特征提取 均值 集成学习 粒子群优化 卷积神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP309.1[自动化与计算机技术—控制科学与工程] TP391.4