Aiming to the multi-variable, non-linear, high-order, close coupled, inverted pendulum system, a mathematical model of the inverted pendulum was established. The LQR controller was put forwards according to this model. In the LQR controller, the selection of the weight matrix Q and R directly impact on structural dynamic response and controlling force. How to determine the optimal Q and R to obtain the global optimal control is still a problem. The matrix Q and R were optimized based on genetic algorithm with different objective functions and the results were verified here. The simulation results show that the optimal preference weighting coefficient were obtained by genetic algorithm and the satisfactory factors can be selected by the designer according to the preference, while the subjectivity and blindness of the conventional method have been overcame.