Design of a Novel Recurrent Neural Network Model for Identification and Adaptive Control of Dynamical Systems

ABSTRACT This work introduced a novel approach to identify and control nonlinear systems using an artificial neural network (ANN). In this study, we have designed a modified diagonal recurrent neural network (MDRNN)‐based identifier and controller. The ANN‐based controller is designed using an adaptive indirect control scheme. The conventional diagonal RNN model is expanded to the proposed architecture of the MDRNN. The architecture of the proposed MDRNN model contains an additional self‐feedback loop on the output neuron. Its output layer receives direct input from the initial layer with delay. The hidden layer of the proposed model can capture the dynamic behavior of the nonlinear plant under study since it contains self‐recurrent neurons. The backpropagation optimization method is the most widely used learning technique for updating the parameters of the suggested model. The effectiveness of the proposed neural network is compared with different ANN models, including the feedforward, Jordan, and diagonal neural networks. One benchmark system, a single‐link rotary inverted pendulum, and a second, a nonlinear dynamic plant are considered for the performance analysis of complex dynamic systems. The simulation results demonstrate that the suggested model performs better than the alternative neural network models. Additionally, the fluctuation of the parameters and the disturbance signal are considered to assess the robustness of the MDRNN‐based controller.