Research On Self Adaptive Sliding Mode Control Based On RBF Network

The sliding mode control has the merits of high response and th invariablity to systemic parameters and external disturbance, futhermore, it algorithm is simple and is easy to be realized engineering. So it gets a majo advance in sloving the sliding mode control problem of nonlinear system. How t(?) combine sliding mode control with neural network to enhance the performance o sliding mode control is studied in the paper.First, from the point of view of raising the converging speed of neura network, a novel online traming algorithm for RBF neural network is presented The algorithm use a linear-least-squares method to train the network weight but (?) momentum backpropagation to train the center and width. It is shown by simulations that the algorithm is more accurate than the gradient desen backpropagation.For a class of linear uncertain system, a method of equal control is used in the paper. The equal control input is calculated based on the certain part o(?) system, and a RBF network is used to apply the swtich control input by conpensating the uncertainties. The simulated results show that the twitter using the method presented in the paper is smaller than the stategy using fixed control gain.And it does not need to know the upper bound of the uncertainties.Recent years, the method of reaching law has been widly used in sliding mode control. This paper uses exponential reaching law as an example to analyze the quantity of the twitter. And an exponential reaching law which has a variable parameter is presented in the paper.At last , a self adaptive sliding mode control method based on the systemic indentification is proposed in the paper. It uses a RBF network to indentify the systemic parameters and then uses the method of exponential reaching law to design the sliding mode controller. It can slove the sliding mode control problem of the complex nonlinear system. The simulation used of an inverted pendulum indicates that, compared with the strategy of direct neural network sliding mode control, the method proposed in the paper has high tracking performance and the twitter is becoming more smaller obviously, therefore , the systemic trobustnes is more stronger and it also enhaces the utility of the sliding mode control.