This integrated study aims to provide a novel performance evaluation algorithm for double-layer arch-truss girder composite bridges based on an improved PCA-K-means-feature analysis method, and to apply it to the performance evaluation of cable-stayed systems in the maintenance phase based on safety monitoring. This algorithm mainly includes monitoring data collection, pre-processing of heterogeneous data from multiple sources, determination of key factors, improvement of K-means cluster analysis, determination of target thresholds, and performance evaluation based on eigenvalue analysis. By collecting bridge performance monitoring data, cleaning these data, and then using the cubic spline interpolation method to preprocess multi-source heterogeneous data. Determine key factors based on principal component analysis and classify performance parameter data into three categories using an improved K-means cluster method. Then, based on the finite element calculation results, the target threshold for the mechanical performance state of the measurement point corresponding to the measurement point position is determined. The performance state of the bridge is evaluated by comparing the extracted feature values such as mean and variance with the target threshold. Validate the method through examples and provide suggestions for practical applications and future research directions. Research has shown that the improved K-means cluster method can improve the accuracy and reliability of clustering analysis. Based on the improved PCA-K means cluster feature analysis method, the performance status of bridge structures can be evaluated.