Aiming at the difficulties of knowledge acquisition of training data in the neural network direct inverse control, method for generalizing design method of neuro-controllers was proposed. After analyzing the Method of Complex (MC) and Particle Swarm Optimization (PSO),a novel algorithm called Complex Particle Swarm Optimization (CPSO) was deduced based on matching the present best point with the worst point, and taking advantage of median point objective value to judge which part that the better point would be on the line of the best point and worst point, and also learning from present best point. Based on CPSO, we optimized control inputs of dynamic systems, and then trained neuro-controller with the obtained desirable response trajectory and control signals that produce it as training data. The synchronous machine was employed as a test-bed to demonstrate the effectiveness of the proposed design method and the simulation results are given at the end of paper..