The multivariate time series data generated during elevator operation are crucial for building an elevator safety monitoring system. To address the issue that existing methods ignore the correlation between series， an elevator multivariate time series anomaly detection method based on Dynamic Spatial-Temporal Encoder （DSTE） was proposed. First， a dynamic graph aggregation module was introduced to compute the global node self-attention through graph convolution and attention mechanism for spatial feature aggregation. Meanwhile， a graph pooling method was employed to dynamically update the graph structure and adjust the degree of correlation between series. Then， a temporal feature aggregation module was constructed by incorporating gated convolution modules into the traditional Temporal Convolutional Network （TCN）， and the temporal feature propagation was calculated utilizing neural ordinary differential equations， thereby realizing precise spatial-temporal feature modeling. Finally， multi-layer convolutions were used for decoding and prediction， so as to identify anomalies. Experimental results on a real-world elevator operation dataset show that compared to the suboptimal model DTAAD （Dual TCN-Attention networks for Anomaly Detection）， DSTE achieves a 0.34% improvement in precision， a 10.85% increase in recall， a 5.39% enhancement in F1-Score， and a 56.78% reduction in False Negative Rate （FNR）， while balancing the FNR and False Positive Rate （FPR）， verifying that the proposed method can identify elevator anomalies effectively and reduce the missed detection of abnormal events.