In the modern interconnected world, intelligent networks and computing technologies are increasingly being incorporated in industrial systems. However, this adoption of advanced technology has resulted in increased cyber threats to cyber-physical systems. Existing intrusion detection systems are continually challenged by constantly evolving cyber threats. Machine learning algorithms have been applied for intrusion detection. In these techniques, a classification model is trained by learning cyber behavior patterns. However, these models typically require considerable high-quality datasets. Limited attack samples are available because of the unpredictability and constant evolution of cyber threats. To address these problems, we propose a novel federated Execution & Evaluation dual network framework (EEFED), which allows multiple federal participants to personalize their local detection models undermining the original purpose of Federated Learning. Thus, a general global detection model was developed for collaboratively improving the performance of a single local model against cyberattacks. The proposed personalized update algorithm and the optimizing backtracking parameters replacement policy effectively reduced the negative influence of federated learning in imbalanced and non-i.i.d distribution of data. The proposed method improved model stability. Furthermore, extensive experiments conducted on a network dataset in various cyber scenarios revealed that the proposed method outperformed single model and state-of-the-art methods.
Intrusion detection is essential for ensuring the security of industrial control systems. However, conventional intrusion detection approaches are unable to cope with the complexity and ever-changing nature of industrial intrusion attacks. In this study, we propose an industrial control intrusion detection approach based on a combined deep learning model for communication processes that use the Modbus protocol. Initially, the network packets are classified as carrying information and noncarrying information based on key fields according to the communication protocol used. Next, a template comparison approach is employed to detect the network packets that do not carry any information. Furthermore, an approach based on a GoogLeNet-long short-term memory model is used to detect the network packets that do carry information. This approach involves network packet sequence construction, feature extraction, and time-series level detection. Subsequently, the detected intrusions are classified into multiple categories through a Softmax classifier. A gas pipeline dataset of the Modbus protocol is used to evaluate the proposed approach and compare it with existing strategies. The accuracy, false-positive rate, and miss rate are 97.56%, 2.42%, and 2.51%, respectively, thus confirming that the proposed approach is suitable for intrusion detection in industrial control systems.
Intrusion detection is only the initial part of the security system for an industrial control system. Because of the criticality of the industrial control system, professionals still make the most important security decisions. Therefore, a simple intrusion alarm has a very limited role in the security system, and intrusion detection models based on deep learning struggle to provide more information because of the lack of explanation. This limits the application of deep learning methods to industrial control network intrusion detection. We analyzed the deep neural network (DNN) model and the interpretable classification model from the perspective of information, and clarified the correlation between the calculation process of the DNN model and the classification process. By comparing the normal samples with the abnormal samples, the abnormalities that occur during the calculation of the DNN model compared to the normal samples could be found. Based on this, a layer-wise relevance propagation method was designed to map the abnormalities in the calculation process to the abnormalities of attributes. At the same time, considering that the data set may already contain some useful information, we designed filtering rules for a kind of data set that can be obtained at a low cost, so that the calculation result is presented in a more accurate manner, which should help professionals lock and address intrusion threats more quickly.
Vulnerability mining technology is used for protecting the security of industrial control systems and their network protocols. Traditionally, vulnerability mining methods have the shortcomings of poor vulnerability mining ability and low reception rate. In this study, a test case generation model for vulnerability mining of the Modbus TCP based on an anti-sample algorithm is proposed. Firstly, a recurrent neural network is trained to learn the semantics of the protocol data unit. The softmax function is used to express the probability distribution of data values. Next, the random variable threshold and the maximum probability are compared in the algorithm to determine whether to replace the current data value with the minimum probability data value. Finally, the Modbus application protocol (MBAP) header is completed according to the protocol specification. Experiments using the anti-sample fuzzer show that it not only improves the reception rate of test cases and the ability to exploit vulnerabilities, but also detects vulnerabilities of industrial control protocols more quickly.
7 December 2021 in Journal of Artificial Intelligence and Technology
Wireless sensor networks have become increasingly popular due to the rapid growth of the Internet of Things. As open wireless transmission mediums are easy to attack, security is one of the primary design concerns for wireless sensor networks. Current solutions consider routing and data encryption as two isolated issues, providing incomplete security. Therefore, in this paper we divide the WSN communication process into a data path selection phase and a data encryption phase. By proposing an improved transmission method based on ant colony optimization and threshold proxy re-encryption for wireless sensor networks,named as ACOTPRE, it resists internal and external attacks and ensures safe and efficient data transmission. In the data path selection stage, the ant colony optimization algorithm is used for network routing. The improvement of the pheromone concentration is proposed. In order to resist attacks from external attackers, proxy re-encryption is extended to WSN in the data encryption stage. The threshold secret sharing algorithm is introduced to generate a set of re-encryption key fragments composed of random numbers at the source node. We confirm the performance of our model via simulation studies.