Abstract: The local search ability of the classical genetic algorithm in the complex search space is weak, and easy to drop into premature convergence. When close to the optimal solution the search is inefficient, due to less pressure of optimization. In view of problems above, Lamarckian learning mechanism is introduced into the population evolution of the traditional genetic algorithm, the local search operator based on the Lamarckian learning mechanism is designed, and hybrid genetic algorithm model is constructed, which can make the advantage of learning fully, enhance the local depth search capability and accelerate the rate of global convergence. The application in experiment modal parameter identification for continuous miner speed reducer proves the effectiveness and accuracy of the hybrid genetic algorithm.

Key words: Experiment modal, Genetic algorithm, Lamarckian learning, Parameter identification, Powell search method, Speed reducer