Abstract: When using a fixed kernel function, only a single and fixed linear or simple nonlinear mapping of fault characteristics can be performed, which leads to poor accuracy in fault detection of the control circuit of the feedwater pump group in power plants. To this end, research is being conducted on fault detection of control circuits for power plant feedwater pump units based on improved SVM. Firstly, data is collected and denoised through mean filtering. Time domain, frequency domain, and time-frequency features are extracted and fused together to fuse multi-source feature data of the power plant feedwater pump control circuit. Then, a mixed kernel function is used to combine the linear kernel with the RBF kernel, and an improved PSO algorithm is used to automatically search for the optimal parameter combination, thereby improving the SVM algorithm to optimize the classifier parameters. Finally, three classifiers including improved SVM, random forest, and extreme gradient boosting were integrated, and the results were fused using an improved weighted voting method to achieve fault detection in the control circuit of the power plant feedwater pump group. The experiment shows that the research method makes the predicted classification results highly consistent with the actual classification, and the median accuracy of various fault detection methods is significantly higher than that of the comparison method, which improves the accuracy of fault detection in the control circuit of the power plant feedwater pump group.