Abstract: Based on in-situ measured data of hydraulic generator vibration, an improved genetic algorithm is adopted to identify the vibration load of hydraulic generator. Unlike the traditional inverse algorithm, the genetic algorithm randomly searches the inverse problem domain by the objective function itself other than follows the gradient direction of objective function. Combined with simulated annealing, a mixed genetic algorithm is developed to enhance the diversity of populations in the evolution process and effectively protect from premature problem of simple genetic algorithm. Combined with the gradient search, a hybrid genetic algorithm is formed to overcome the local minimum of the gradient search and the slowness of convergence rate of simple genetic algorithms. The practical application of the proposed inverse algorithm shows that the prediction of vibration responses of hydraulic generator is highly accurate according to the identified loading parameters.