CM-AW-TCF: A cross-modal adaptive wavelet transformer-TCN fusion fault prediction model for special equipment in vertical transportation

Elevators are critical special equipment in vertical transportation. To address the limited modeling of operating-cycle dependencies and the degradation of multi-source information fusion caused by the multi-scale nonstationarity, multi-source heterogeneity, and strong coupling characteristics of elevator power system data, a cross-modal adaptive wavelet Transformer–TCN fusion model (CM-AW-TCF) is proposed for fault prediction. First, to address insufficient shallow signal representation, an adaptively weighted wavelet dictionary is used to highlight sensitive scales and enhance the representational capability of multi-scale energy features. Second, to address the limitation of collaborative representation in deep modeling, a dual-stream Transformer–TCN architecture is constructed to capture global correlations and local variations, respectively, thereby improving deep collaborative feature modeling. Finally, to reduce key information loss during modal interaction, a bidirectional cross-modal attention mechanism with a gating strategy is introduced to enable efficient fusion of modal information and improve fault prediction accuracy. Experimental results show that the proposed method achieves a coefficient of determination of 0.977, a root mean square error of 0.046, and a mean absolute error of 0.035, outperforming the comparative models across all evaluation metrics and effectively improving fault prediction accuracy.