Elevator fault classification based on multi-grained cascade forest with variable importance measure
Yijin Ji, Haoxiang Sun, Xu Zhou · International Journal of Automation and Control · 2025
Current deep learning based elevator fault classification models are mostly elevator-specific due to the lack of a general fault dataset. The performance is also significantly limited if the training data is insufficient or the hyper-parameter is not well tuned. In this work, a new elevator fault dataset is firstly proposed for different elevators based on massive fault recordings from 96,333 elevator emergency disposal service platform and the basic parameters of elevators. Variable importance is then evaluated using random forest with mean decrease accuracy (MDA) for better feature understanding. Using the variables with high importance as input features, a multi-grained cascade forest model is finally proposed for more accurate and faster elevator fault classification. Experiment results validate the superior performance of the proposed model, including an easy training on relatively less training data, a higher classification accuracy than traditional models, and a higher running efficiency than not using variable importance measure.