In order to accurately achieve elevator fault diagnosis, an elevator fault diagnosis based on AO-VMD-BF and multi model fusion was proposed. Firstly, the variational mode decomposition(VMD) optimized by the aquila optimizer(AO) algorithm was used to decompose the signal into multiple modal components, and the pearson correlation coefficient was used to remove false components. To address the problem of noise in the remaining signal, the butterworth filter(BF) was used for secondary denoising, the denoised vibration signal could be obtained by reconstructing the filtered modal sub quantum sequence. Then extract time-domain, frequency-domain, and entropy features to form a multi domain feature vector set. Finally, a Stacking ensemble learning elevator fault diagnosis model was established based on convolutional neural network(CNN), random forest(RF), support vector machine(SVM), and adaptive boosting(AdaBoost) models, with extreme gradient boosting(XGBoost) as the meta classifier. The analysis of experimental results shows that the proposed method can effectively extract fault features from elevator car vibration signals, and accurately and effectively diagnose elevator faults.