Abstract: To further improve the prediction accuracy of thermal errors in machine tools and to mitigate the impact on the machining process, a thermal error prediction model based on LSTM-TCN-A is proposed for the motorized spindle, which is the most severe heat-generating component in CNC machine tools. A specific type of horizontal lathe was selected as the experimental object. Experiments at various rotational speeds were designed, during which the temperatures at critical positions of the motorized spindle and the thermal elongation errors at the shaft end were measured and collected. Based on the hierarchical clustering and grey relational analysis, three temperature-sensitive points were identified. Furthermore, a comparative study was conducted with commonly used methods. It is demonstrated by the results that, compared to other algorithms, higher accuracy in forecasting the thermal errors of the spindle is achieved by the LSTM-TCN-A.