The configuration item values of each component within the autonomous driving system (ADS) are mutually constrained and complexly associated, which leads to problems such as configuration logic faults and memory overflow easily occurring during the configuration process, thus triggering the frequent occurrence of system configuration faults. This paper proposes a dynamic fuzzing method for configuration parameters based on static analysis of source code, through mapping and extracting configuration constraints within the system, realizing dynamic fuzzing for configuration parameters based on source code characterization, and completing the diagnosis of configuration defects and assistive repair functions. The experimental results show that the fuzzing method is effective in detecting and localizing configuration parameter defects under a variety of ADS anomalies, which not only dramatically improves the coverage rate of system configuration defect detection, but also achieves a configuration defect detection rate of 89.9%.