Research on the Integration Method of Software Static Testing Tools Based on Machine Learning

The reliability of autonomous control software is critical to the power grid's operational safety, necessitating reliable means to assure software quality. Static analysis, with its ability to function autonomously, be deployed earlier, and operate without executing code, has been extensively adopted for testing software, becoming a vital tool in ensuring the quality of power grid dispatch automatic control software. Given the plethora of static analysis tools available, each with their unique strengths and varying levels of sensitivity and specificity, addressing their significant rates of false positives and negatives has emerged as a crucial challenge. This study leverages the data from defect detection reports of CppCheck, TscanCode, and Flawfinder, three leading open-source static analysis tools, along with GaussianNB, RF, and KNN — three machine learning classification algorithms. By integrating the analysis from these tools, defects are accurately segregated into true and false positives, with the latter being strategically filtered out. It achieves a substantial increase in the precision of the integrated defect report based on a small reduction in the overall recall rate, providing guidance and reference for the static analysis of grid dispatch automatic control software.