Existing machine tool thermal error mitigation relying on passive structural optimization and conventional feedforward PID decoupling poorly addresses strong multi-temperature-field coupling, large time delays, and nonlinear thermal characteristics in large precision horizontal machining centers. These methods lack predictive optimization, fail to suppress the long-term temperature drift of key structural components, and cannot realize active thermal intervention, leaving a clear research gap. This paper develops a three-layer closed-loop active thermal control framework with temperature sensing, numerical decoupling, and executive regulation. S-shaped hollow aluminum temperature control plates are optimally arranged on the bed, column, and beam, and a multi-temperature zone coupling transfer function model is established. A hybrid control strategy integrating feedforward decoupling, MPC prediction, and PID steady-state compensation is proposed; MPC is introduced to handle multivariable coupling, time lag, and actuator constraints beyond the capability of traditional PID. Comparative experiments show that the MPC-based scheme reduces key point temperature variation rates by 31.47%, 14.56%, 16.06% and 44.86%. This study focuses on temperature stabilization (rather than the direct measurement of the spindle drift or geometric deformation). The proposed method provides an effective active temperature balance solution for large precision machine tools.