关键词: 油料储运, 工控系统安全, 半实物仿真系统, 工控系统安全数据集, 迁移学习数据集

Abstract: With the rapid development of a new generation of information technology such as artificial intelligence, big data, and the Internet of Things, the wave of industrial Internet has swept the world, and the security problems of industrial control systems have become more and more prominent. In particular, in the industrial control system in the field of oil storage and transportation, the focus of solving safety problems lies in business safety. At present, the main factor restricting the anomaly detection algorithm at the service level is the lack of data sets of business anomaly data of the industrial control system. In this paper, a semiphysical simulation system based on a real oil storage and transportation industrial control system business is studied, and the system is attacked against the underlying business, so that a negative sample is obtained, and together with the normal data, a business security data set of the oil storage and transportation industrial control system (hereinafter referred to as the oil storage and transportation data set) is formed. The oil storage and transportation dataset was compared with the Mississippi dataset and the Singapore Water Plant dataset, and the transfer learning experiment was carried out using the above three datasets. The comparison results showed that the oil storage and transportation dataset contained more types of attacks than the other two datasets, and negative samples accounted for the highest proportion of the three datasets. The results of the transfer learning experiment show that the accuracy rate obtained by transferring the dataset of this paper to the dataset of the Singapore water plant dataset is higher than that of transferring from the Singapore water plant dataset to this dataset. Maybe because the attacks of the data set of this paper are relatively comprehensive and more attack samples are “seen” when transferring to the Singapore water plant dataset, a better accuracy rate is obtained; The same transfer learning algorithm has a high transfer accuracy for Singapore dataset and Mississippi dataset, but from the process flow analysis of industrial control system, the two datasets have no similarities, and there may be over learning phenomenon; The accuracy of transfer learning between this dataset and Mississippi dataset is low. In fact, the two datasets are completely different industrial control processes, which comply with the objective law.Key words oil storage and transportation; industrial control system security; hardware in the loop simulation system; security dataset of industrial control system; transfer learning dataset

Key words: oil storage and transportation, industrial control system security, hardware in the loop simulation system, security dataset of industrial control system, transfer learning dataset