The empirical modeling method based on modern control theory is complex to establish a standard thermal error solution for different production conditions of CNC machine tools. Explore the research on adaptive prediction of thermal errors in CNC machine tools using model-free driving under the digital twin framework. Firstly, a digital twin framework based on the “thermal sensing-mapping fusionoptimization-drive” structure of machine tools is established to achieve the storage and fusion of thermal feature information in the digital twin. Then, based on the assumptions of the MISO system and the dynamic linearization geometric interpretation, a thermal error model free adaptive control (MFAC) method is proposed that is not affected by any structural data of the controlled system. Furthermore, based on the dynamic discovery probability and adaptive step size of the DACS-MFAC algorithm, the system parameters are updated according to a certain period to achieve dynamic optimization of thermal error prediction values in the digital twin system. Experimental result shows that the DACS-MFAC method has advantages such as strong adaptability, high accuracy, and good convergence.