Shape-Adaptive RBF Neural Network for Model-Based Nonlinear Controlling Applications

Abstract

Radial basis function neural networks (RBF-NNs) are simple in structure and popular among other NNs. RBF-NNs are capable of fast learning proving their applicability in developing deep learning applications. Its basic form with center states or the means and standard deviations with weight adaptation makes limited variability and complex in tuning when such embedded to the model. Dynamics systems are nonlinear, especially behavior is uncertain and unpredictable, and complete mathematical modeling or model-based controlling has limited applicability for stability and accurate control. Shape-adaptive RBF-NN presented in the paper theoretically proved for stability control using the Lyapunov analysis. The autonomous surface vessel controlling selected for the numerical simulation consists of a mathematical model developed using marine hydrodynamics for a prototype vessel and classical proportional-derivative (PD) controller. Results indicated that shape adaptive RBF-NN blended controlling is more accurate and has a fast learning ability in intelligent transportation vessel development.