Abstract: To analyze the temperature field of the linear feed system of CNC machine tools, it is necessary to arrange a certain number of temperature measurement points for temperature data collection. However, the research results will be directly affected by the location and number of measurement points. To achieve accurate placement of temperature measurement points, an improved Canopy FCM-GRA temperature measurement point optimization model based on statistical theory is proposed in this paper. Taking the X-axis linear feed axis of a certain CNC machine tool as an example, the pre clustering number is first determined based on the measured experimental data, and then the corresponding temperature sensitive points are selected through fuzzy matrix and grey correlation coefficient. Finally, based on SVR theory, temperature thermal error prediction models are established before and after temperature measurement point optimization. The effectiveness of temperature measurement point optimization is verified by comparing the accuracy of the two models. The results show that the optimization effect of temperature measurement points is good, and the optimized temperature measurement points can accurately explore the thermal characteristics of the feed system.