The code of third-party tool software such as MATLAB cannot be run in the hardware of general compensators, which leads to the fact that most models cannot be applied to the actual compensation of the thermal error of machine tools. A real-time compensation method for thermal error of static-pressure turntable was proposed, in order to improve the efficiency of error modeling and reduce the hardware requirements of the compensation system. The compensation method is based on the support vector machine (SVM). The fish population algorithm and the wolf group algorithm are used to optimize the core parameters of SVM in advance and later, and the modeling efficiency is improved under the premise of ensuring prediction accuracy. By the offline training of MATLAB, the support vector is filtered and imported into the developed compensation software, and object linking and embedding for process control (OPC) is used to implement real-time online compensation for thermal errors. Compared with the traditional multiple linear regression modeling method, the proposed model is better in accuracy and efficiency. The results of compensation experiment showed that the axial error of the turntable was reduced from the original maximum of 40 μm to about 10 μm. The machining accuracy of the turntable was improved by 75%, which verifies the effectiveness of the proposed compensation method.