A self-learning temperature precise control method is provided. Aiming at the fact that a temperature control system has the characteristics of non-linearity, hysteresis and time variation, the temperature control system is adopted by the invention to automatically adjust and determine the proportion gain P, the integration gain I, the differential gain D and the output power OPH of each temperature control point automatically according to a temperature control error e and a temperature change rate ec, so that the precise control with +/- 0.005 DEG C deviation is realized for each temperature control point. According to the invention, the automatic learning balance of the PID control parameters and the output power OPH are ingeniously utilized to realize the precise control of temperature. The invention is the innovated exploration and try of the temperature precise control method, the high-speed processing capability of an embedded ARM processor and intelligent control parameter learning are used to replace a normal PID control adjusting technology, so that the temperature control precision is improved, and the self-learning temperature precise control method is simple, reliable and effective.