Design of a novel robust recurrent neural network for the identification of complex nonlinear dynamical systems

• Published in: Soft computing (Berlin, Germany) Vol. 28; no. 3; pp. 2737 - 2751
• Main Authors: Shobana, R., Jaint, Bhavnesh, Kumar, Rajesh
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01-02-2024; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Back propagation networks; Computational Intelligence; Control; Design; Dynamical systems; Engineering; Fuzzy logic; Identification; Mathematical Logic and Foundations; Mechatronics; Neural Networks; Nonlinear dynamics; Nonlinear systems; Recurrent neural networks; Robotics; Stability analysis; Recurrent neural networks; Disturbance rejection; Feed-forward neural network; Identification
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-023-09187-5

A novel fully connected recurrent neural network (FCRNN) structure is proposed for the identification of unknown dynamics of nonlinear systems. The proposed recurrent structure consists of internal feedback layers of adjustable weights which impart necessary memory property to the structure and improves its ability in handling the dynamical systems. The back-propagation algorithm (BP) is used to derive the weight update equations of the proposed model. The convergence of the proposed approach is proven in the sense of Lyapunov-stability analysis. A total of three examples are considered and the performance of the proposed structure is evaluated by comparing it with the results obtained from other popular neural network models such as feed-forward neural network (FFNN), Elman neural network (ENN), Jordan neural network (JNN), and the locally recurrent neural networks (LRNN). Experimental results obtained show that the FCRNN model has outperformed the other neural models in terms of identification accuracy and robustness.