In the current landscape of network automation operations and maintenance assessment methods, several issues have been identified, including poor noise resistance, low testing accuracy, instability, and extended testing durations, all of which impact the overall effectiveness of the testing process. To overcome these challenges, we introduce an innovative evaluation and testing approach. This method leverages the VRNN algorithm for detecting anomalies in network server data and incorporates a quantifiable security assessment model for AI-driven operations and maintenance testing. Our experimental results validate the effectiveness of this approach by demonstrating robust noise resistance and achieving a testing accuracy of over 92%. The testing process is characterized by its stability and is efficiently completed within a short timeframe. Remarkably, it can conduct 3000 data tests in as little as 60 seconds, thus underscoring the efficacy of our proposed methodology.