说明：  首先对非线性模型在控制目标附近进行线性化得到相应的状态空间表达式，再对线性模型设计LQR控制器，进而用此控制器控制非线性倒立摆，取得训练神经网络控制器的数据，最后用这些数据训练得到神经网络控制器，最终基于神经网络的（一、二、三级）倒立摆控制均取得了很好的控制效果；
(First non-linear model in the vicinity of the control objectives of corresponding linear state-space expression for the linear model re-design LQR controller, and then use this non-linear inverted pendulum controller to obtain the training neural network controller data, and finally using these data have been trained neural network controller, and ultimately based on neural network (I, II, 3) inverted pendulum control are achieved very good control effect )