Neural Network-Based Parameter Estimation in Dynamical Systems

Abstract

Mathematical models are designed to assist decision-making processes across various scientific fields. These models typically contain numerous parameters, the values’ estimation of which often comes under analysis when evaluating the strength of these models as management tools. Advanced artificial intelligence software has proven to be highly effective in estimating these parameters. In this research work, we use the Lotka–Volterra model to describe the dynamics of a telecommunication sector in Greece, and then we propose a methodology that employs a feed-forward neural network (NN). The NN is used to estimate the parameter’s values of the Lotka–Volterra system, which are later applied to solve the system using a fourth-algebraic-order Runge–Kutta method. The application of the proposed architecture to the specific case study reveals that the model fits well to the experiential data. Furthermore, the results of our method surpassed the other three methods used for comparison, demonstrating its higher accuracy and effectiveness. The implementation of the proposed feed-forward neural network and the fourth-algebraic-order Runge–Kutta method was accomplished using MATLAB.