融合自编码降维的改进DNN水利工控网入侵检测算法

Improved DNN Intrusion Control Algorithm for Water Conservancy Industrial Control Network Combined with Self-encoding and Dimensionality Reduction

为解决工控网异常入侵、水利泵站通信网安全防护的问题。论文提出一种基于深度神经网络的水利泵站工控网入侵数据的检测算法。首先针对泵站工控网内的数据进行预处理,通过自编码算法对数据进行特征提取、降维处理;利用深度神经网络模型结合受限玻尔兹曼机对各类数据进行训练,采用Adadelta算法进行网络模型的参数优化,并由Softmax分类器对工控网数据进行是否合法判别。实验数据集由底层设备实地采集到的水利泵站工控网内流动数据导入到本地数据库。实验结果表明:该方法的准确率对比深度神经网络未改进前的算法提高了3.76%,检测率提高了6.32%,漏报率降低0.5%,从而验证了论文方法的有效性。

To solve the problems of abnormal intrusion in industrial control network and the safety protection of communica⁃tion network in water conservancy pumping station,an intrusion detection algorithm of industrial control network based on depth neural network is proposed.The data in the industrial control network of pumping station are preprocessed and then feature extrac⁃tion and dimensionality reduction of data is processed through self-encoding algorithm.Deep neural network model combined with restricted Boltzmann machine is used to train various types of data,Adadelta algorithm is utilized to optimize network model parame⁃ters,and whether the industrial control network data is legal or not judging by Softmax classifier.The experimental data sets are im⁃ported into the local database from the flow data collected from the industrial control network of the water conservancy pumping sta⁃tion.The experimental results show that the accuracy of this method is improved by 3.76%,the detection rate is increased by 6.32%and the false alarm rate is reduced by 0.5%compared with the algorithm without improvement of deep neural network.