The invention relates to a PID control parameter off-line tuning method based on a hidden Markov model. The method comprises the steps of acquiring a PID control parameter observation sequence, judging whether the length of the acquired observation sequence is greater than 1, calculating the initial state probability distribution of the observation sequence, carrying out hidden Markov model training with use of the obtained probability distribution to obtain the state probability of the observation sequence after training, sending the trained control parameter observation sequence to a PID control system for verification, and judging whether the set of control parameters meets the precision requirement of the PID control system. By adopting the method of the invention, deterministic calculation of a PID control parameter sequence vector is realized, an offline PID control parameter optimization adjustment method is proposed, and the difficulty in parameter debugging caused by the randomness of parameter determination is avoided.