Abstract: ［Objective］ Accurately forecasting the demand for spare parts during critical rail transit vehicle equipment maintenance provides a foundation for scientific spare parts management and improves maintenance cost-efficiency. However, existing forecasting methods lack accuracy and exhibit poor performance. Thus, an in-depth study of spare parts demand forecasting in maintenance is necessary. ［Method］ Based on the characteristics of spare parts data for rail transit vehicle maintenance, monthly time series are constructed for the replacement rate (i.e., the proportion of parts replaced) and maintenance amount. After thoroughly investigating time series forecasting algorithms and comparing their performance, comprehensively considering their accuracy and generalization, a combined method leveraging ARIMA (autoregressive integrated moving average) and SVR (support vector regression) is proposed. ARIMA is used to predict replacement rates, while SVR is applied to forecast maintenance amount. Finally, the spare parts forecast is calculated by combining the prediction results of both. Additionally, an anomaly detection method for replacement rates is proposed by integrating ARIMA′s confidence intervals with the unsupervised clustering algorithm IForest (Isolation Forest). ［Result & Conclusion］ Taking advanced maintenance data of height valve and brake caliper unit two typical products as example, validation and calculation of the proposed forecasting method is conducted. Results show that compared to the existing historical average method, the proposed method significantly improves forecasting accuracy and effectively detects anomalies in both historical and current replacement rates during maintenance.