Research And Optimization Of Intrusion Detection Model Based On Bidirectional Generative Adversarial Networks

With the continuous development of the Internet and information technology,people’s production and life are increasingly reliant on networks and digital technology.As a result,network security threats have increased,and network attack methods have become more complex and diverse.This paper takes the AI intrusion detection module in the Attack Risk Identification and Regulatory Audit subsystem of a company’s data security operations management platform as the background,addressing the challenges of high false positive and false negative rates,imbalanced data samples,and the difficulty in identifying unknown attacks in existing intrusion detection systems.Deep learning techniques are applied to intrusion detection,where an optimization is proposed for the classical Bi GAN network,called Add-Bi GAN.Additionally,a method for establishing an intrusion detection feature dataset and optimizing Add-Bi GAN using the feature dataset is presented,and validation is conducted through ablation experiments and comparative experiments.The specific research work in this paper is as follows:Firstly,an optimization is proposed for the classical Bidirectional Generative Adversarial Network,resulting in Add-Bi GAN.This network directly inputs the input,encoding results,or generation results to the discriminator,and utilizes the Add operation in the discriminator to replace the original Concatenate operation for feature combination.Ablation experiments and comparative experiments on the KDDCUP99 benchmark dataset show an F1 score of 96.11%,indicating that the proposed Add-Bi GAN method outperforms mainstream methods.By superimposing semantic information through the Add operation,this network achieves more accurate data classification.Secondly,a method is proposed for establishing an intrusion detection feature dataset and optimizing the Add-Bi GAN using the feature dataset.Taking the company’s application-layer access dataset as a case study,this method is employed to establish the application-layer access feature dataset,which is then used to optimize the proposed Add-Bi GAN.Experimental results demonstrate that the optimized Add-Bi GAN model performs better than the company’s original method on the actual dataset,with an accuracy improvement of 0.8%.Finally,based on the aforementioned research achievements,a software engineering approach is utilized to design an intrusion detection module based on Add-Bi GAN.The module is implemented using Python language,Flask framework,and My SQL database,and integrated into the company’s Attack Risk Identification and Regulatory Audit subsystem.The Bi GAN network optimization method,feature dataset establishment method,optimization method using the feature dataset to improve Add-Bi GAN,and the development process of the intrusion detection module described in this paper are not only applicable to this topic but also contribute to feature dataset establishment and deep learning network optimization in similar application scenarios.