Among spindle dynamic errors, thermal error is an important factor affecting machining precision, the most convenient and economical solution for which is dynamic compensation of thermal error. In view of defects of complicated system, complex debugging, poor stability, high cost, etc., in the existing compensation approaches, thermal error compensation technology is investigated with compensation model established to complete the system design. The system is tested on the spindle of precise horizontal machining center. In the experiment, multiple linear regression, BP neural network, etc. are taken for technical analysis with dynamic feedforward model established. The experimental results indicate that the compensation system under study has dynamic compensation rate of over 90%, while the spindle has greatly shorter warm-up time, which can operate normally before entering the thermal steady state. The finally completed feedforward model-based intelligent compensation control system has good steady-state and dynamic compensation performance. In particular, its self-correction function greatly improves system adaptability and simplifies operation in use.