Research On Vulnerability Mining Techniques For Industrial Control Protocols Using Generative Adversarial Networks And Probabilistic Mutation Models

With the rapid development of global industrial informatization,the interconnection of industrial control systems with traditional Internet systems has led to a series of industrial control security vulnerability problems.In the vulnerability mining techniques for the industrial control network protocols,the fuzzy testing has been widely concerned by research scholars because of its low cost and strong vulnerability mining ability.However,generation-based fuzzy testing requires manual extraction of protocol specifications and has low test case pass rate.The construction of mutation strategies based on fuzzy testing of mutations requires expert empirical knowledge in relevant fields,and the abnormality detection power of mutation use cases is weak.These problems greatly restrict the application and development of fuzzy testing techniques.In order to solve the problem that generation-based fuzzy testing relies on manual analysis of protocol specifications and low pass rate of test cases,this paper first proposes a GRUSACNNFuzzer,an automatic fuzzy testing model for industrial control protocols based on generative adversarial networks.The model enables the convolutional operations in the generative adversarial network model to take into account the interrelationships between different protocol fields and extract the global features of the protocols through the introduction of a self-attentive layer.In addition,the GRU-SACNN joint network is introduced into the discriminator,which can fully learn the time-step dimension and spatial structure dimension of the protocol,and then automatically generate test cases that are more compliant with the protocol specification,resulting in a higher passing rate of the test cases.The results of the comparison experiments show that the GRU-SACNNFuzzer has an 8.3% improvement in test case pass rate and finds a higher number of protocol vulnerabilities than the best performing fuzzer CGFuzzer.In order to solve the problems of complicated mutation strategy construction and weak anomaly detection ability of mutation-based fuzzy testing,this paper proposes a CNNLSTMMutator,a probabilistic mutation-based fuzzy testing model.Unlike existing mutation fuzzy testing models based on genetic algorithms or reinforcement learning,this model uses the best mutation probability for different protocol fields to mutate,which is simpler and more efficient in terms of mutation strategy.The model uses a CNN-LSTM network structure to deeply extract local structural and temporal features of the protocol and pass them over long distances,so as to learn the protocol specification and generate the most appropriate probabilistic mutation vector for guiding the protocol mutation.Experimental results show that the CNN-LSTMMutator has a shorter model execution time and higher anomaly detection efficiency than the GAFuzzer,a genetic algorithm-based fuzzer,and the mutation strategy is more efficient.In addition,CNN-LSTMMutator has an 8.1% improvement in test case pass rate and finds more anomalies compared to CGFuzzer.Finally,according to the characteristics of industrial control system,this thesis designs a set of industrial control fuzzy testing vulnerability mining system,Fuzzy System.The test results indicate that Fuzzy System can achieve the purpose of protocol vulnerability detection in industrial control environment through the two fuzzy testing models proposed in this thesis,and has good testing performance.