Direct and Indirect Adaptive Neural Control of Nonlinear Systems

Abstract

Abstract. A comparative study of various control systems using neural networks is done. The paper proposes to use a Recurrent Trainable Neural Network (RTNN) identifier with backpropagation method of learning. Two methods of adaptive neural control with integral plus state action are applied – an indirect and a direct trajectory tracking control. The first one is the indirect Sliding Mode Control (SMC) with I-term where the SMC is resolved using states and parameters identified by RTNN. The second one is the direct adaptive control with I-term where the adaptive control is re-solved by a RTNN controller. The good tracking abilities of both methods are confirmed by simulation results obtained using a MIMO mechanical plant and a 1-DOF mechanical system with friction plant model. The re-sults show that both control schemes could compensate constant offsets and that - without I- term did not. 1 Introduction Recent advances in understanding of the working principles of artificial neural networks has given a tremendous boost to identification and control tools of nonlinear systems, [1], [2], [3]. Most of the current applications rely on the classical NARMA approach, where a feed-forward neural net-work is used to synthesize the nonlinear map, [4], [5]. This approach has some disadvantages, [2], like that: the network inputs are a number of past system inputs and outputs, so to find out the optimum number of past val-ues, a trial and error must be carried on; the model is naturally formulated in discrete time with fixed sampling period, so if the sampling period is changed the network, must be trained again; problems associated with sta-bility, convergence and rate of convergence of this networks are not