Research And Implementation Of Intranet User’s Behavior Audit Model Based On Genetic Neural Network

With the rapid development of Internet and expansion of its structure, thesecurity concern of information becomes increasingly severe. How to guaranteeinformation security has aroused great public concern in terms of politics, economicsand all other walks of society. As the relatively mature research outcomes of firewalland IDS system provide well defense against external attacks, cyber-attack hastransformed from external attack to internal attack. Because internal attack hasattributes of generality and imperceptibility, it can easily elude the supervision offirewall and IDS, and therefore becomes much more difficult to be prevented andcauses more harm. One important factor leads to the insecurity of internet is internaluser behavior, while such kind of behavior is usually regarded as safe. So it is ofabsolute significance to audit the internal user’s behavior so as to protect the safetyof intranet.This thesis summarized some characteristics of internal user’s behavior andrepresentation of abnormal behavior. Combined with the local accurate research ofBP algorithm and global searching capacity of genetic algorithm, the author proposesan intranet behavior audit model based on genetic neural network. Different fromtraditional security audit model, this model adopts actual user behavior data insteadof traditional data from internet. Meanwhile, it conducts appropriate improvement oncrossover operator of genetic algorithm, abandons traditional algorithm which adoptsfixed crossover rate and introduces the concept of individual similarity. By virtue ofjudging the similarity of two individuals, whether crossover operations happen or notcan be identified. In this way, the positive model of father individual can be passeddown to next generation and the convergence rate of algorithm will be enhanced.This model is mainly constituted by distributed data collection module, datapre-process module, audit analysis module based on genetic neural network, data storage module, response module and audit data query and scan module.This thesis applies improved initial weight values of neural network optimizedby genetic algorithm, and then utilizes optimized initial weight values to train theneural network, finally searches the optimal network structure. After genetic neuralnetwork learned the model of user behavior, it can conduct analysis of behavior auditand make response to abnormal user behavior. Through algorithm comparison andexperiment results analysis, this model is proved to be effectively utilized in the auditof intranet user’s behavior.