Abstract: This study addresses the difficulty in measuring and modeling thermal deformation in a five-axis horizontal machining center's spindle system. A new method for analyzing, measuring, and modeling the multi-component temperature field and thermal deformation is proposed. First, an equivalent geometric model of key components was created. This model allowed the thermal error at 3 tool center points to be decomposed into 9 error components from the spindle, saddle, and column. A dedicated measurement device was then designed based on finite element simulation results. This device simultaneously captured temperature and deformation data of multiple components under various working conditions. Using the collected data, a long short-term memory (LSTM) network was employed to model the relationship between influencing factors (temperature, speed, coordinates) and each error component. An error compensation experiment validated the model's real-time performance and accuracy. The steady-state errors at the tool tip were reduced by 4.44 μrad (55%) in the A-axis, 5.83 μm (73%) in the Y-axis, and 47.92 μm (74%) in the Z-axis. This research provides a practical reference for tracing and modeling thermal deformation in complex machine tool spindle systems.