Research On The Application Of Adversarial Environment Algorithm Based On Deep Reinforcement Learning In Network Security

With the rapid development of Internet technology,the Internet has penetrated into all corners of society and is inseparable from everyone’s daily life.In this context,the Internet carries an increasingly high economic value and more and more user data.The reliable operation of the network is of great significance for both economic development and social stability.However,at the same time,network security incidents occur globally,and network security is still facing huge threats and challenges.In the past,traditional machine learning algorithms and deep learning algorithms have been used in a variety of ways in the area of network security,especially in intrusion detection systems.But with the further complexity of the network environment,the traditional intrusion detection technology has gradually failed to adapt to new attack scenarios,the research on new intrusion detection technology is urgent.Deep reinforcement learning is a new technology,which is a fusion of deep learning and reinforcement learning,and it has been gradually applied to the field of network security.This study focuses on the application of adversarial environment algorithm based on deep reinforcement learning in network security,and propose three anomaly traffic detection algorithms based on deep reinforcement learning adversarial environment,including:first,an anomaly detection optimization algorithm based on deep reinforcement learning adversarial environment is proposed.After the training of the traditional online learning anomaly detection algorithm is completed,the algorithm uses the adversarial environment agent based on deep reinforcement learning to conduct adversarial training with anomaly detection algorithm,so as to realize the correction and optimization of the anomaly detection algorithm.Then,a sample pool-based adversarial environment anomaly detection algorithm model is proposed,which constructs a more general single-stage adversarial training model.And the sample pool mechanism is proposed to accelerate the training of the adversarial model,solving the problem of high training time cost of the adversarial model.Finally,an anomaly detection algorithm using adversarial environment-based sampling probability space model is proposed,which is called AEPSM-MLP.It innovates a sampling probability space model to solve the problem of collapse in the adversarial environment,and realizes a more intelligent,robust and superior anomaly detection algorithm model.In addition,this study uses the NSL-KDD dataset and performs preprocessing operations such as one-hot coding and normalization on the dataset.The proposed algorithm model is trained and tested on the original training set and test set divided by the dataset,and compared with other algorithm models.The experimental results show that the research outcomes of this thesis have superior performance in various evaluation indicators such as accuracy and time performance,which strongly confirms effectiveness of the research results and the superior performance for anomaly traffic detection of the algorithm proposed in this thesis.