Adaptive Deep Learning Algorithm for Vulnerability Detection and Repair in Big Data Audit System

To solve the problems of low detection accuracy and delayed repair response in complex situations such as massive data streams, frequent changes, and diverse attack methods faced by big data auditing, this project plans to adopt adaptive deep learning (DL) technology to dynamically optimize the detection and repair process, improve the security performance of the system, and meet the rapidly changing data environment. Firstly, this project plans to use a combination of multi-layer convolutional neural networks (CNN) and long short-term memory networks (LSTM) for in-depth research. Secondly, a learning method based on adaptive learning is proposed, which enables dynamic adjustment of the established model according to changes in the environment in different application scenarios. On this basis, a maintenance decision optimization method based on deep reinforcement learning is proposed. The accuracy of this algorithm remains at a level of $0.928-0.933$ as the complexity of the scene increases. The response time for the correction will be slightly extended, from 1.9 ms to 2.3 ms, due to the increased complexity of the scenario. Through testing of the BD system, the effectiveness of this method has been verified, providing a new perspective for the security issues of the BD system.