A Recurrent Neural Network Approach for Identification and Control of Nonlinear Objects

The Neural Network (NN) modelling and application to system identification, prediction and control was discussed for many authors [15]. Mainly, two types of NN models are used: Feedforward (FFNN) and Recurrent (RNN). The main problem here is the use of different NN mathematical descriptions and control schemes, according to the structure of the object model. For example, N a r e n d r a and P a r t h a s a r a t h y [1, 2], applied FFNN for system identification and direct model reference adaptive control of various non-linear objects. They considered four object models with a given structure and supposed that the order of the object dynamics is known. Y i p and P a o [3] solved control and prediction problems by means of a flat-type functional FFNN, used for direct inverse model learning control. P h a m [4] applied Jordan RNN for robot control. S a s t r y [5] introduced two types of neurones  Network Neurones and Memory Neurones to solve identification and adaptive control problems, considering that the object model is also autoregressive one. In [7], some schemes of NN and RNN applications to control, especially of direct model reference adaptive control, are surveyed. All drawbacks of the described in the literature NN models could be summarised as follows: there exists a great variety of NN models and a universality is missing [15]; all NN models are sequential in nature as implemented for systems identification. (The FFNN model uses one or two tap-delays in the input, [1, 2] and RNN models usually are based on the autoregressive model [5], which is one-layer sequential one); some of the applied RNN models are not trainable, others are not trainable in the feedback part [4]. Most of them are dedicated to a SISO and not to a MIMO applications [3]; in more of the cases, the stability of the RNN is not considered, [4], especially during the learning; in the case of FFNN application for systems identification, the object is given in one of the four described in [1] object models, the linear part of the object model, especially the system order, has to be known and the FFNN approximates БЪЛГАРСКА АКАДЕМИЯ НА НАУКИТЕ . BULGARIAN ACADEMY OF SCIENCES