Abstract: The traditional method of abnormal noise recognition has high requirements for test equipment and can be easily affected by the experience difference of experimenters. In view of this, a method of vehicle interior abnormal noise recognition is proposed based on multi-feature extraction and SVM optimized by gray wolf optimization. This method took 6 kinds of common abnormal noises obtained in the test as the research object, and extracted short-term energy, Mel frequency cepstral coefficient optimized by wavelet transform (DWT-MFCC) and its first-order difference to form mixed characteristic parameters. Gray wolf optimization was applied to the parameter optimization of SVM to establish a noise recognition model and perform recognition classification. Result shows that the 25 dimensional mixed features extracted by this method can effectively convey abnormal noise information, and this method can better identify abnormal noise because of its obvious advantages in convergence speed and effect.