Deep Reinforcement Learning with Transfer Learning for Efficient Solution of Fractional Optimal Control Problems

Abstract

This paper introduces a novel approach for solving Fractional Optimal Control Problems (FOCPs) using Deep Reinforcement Learning (DRL). Fractional-order systems, involving derivatives of arbitrary order, have shown great potential for accurately modeling complex, real-world systems that cannot be effectively captured by traditional integer-order models. The inherent nonlinearity and complexity of FOCPs often make them difficult to solve using conventional methods. In this work, we leverage DRL to learn optimal control strategies for fractional systems by interacting with the environment. Additionally, Transfer Learning (TL) is incorporated to accelerate the learning process and enhance model efficiency by utilizing pre-trained models. The proposed method provides significant improvements in computational efficiency, accuracy, and the ability to handle highly nonlinear dynamics compared to traditional approaches. Numerical simulations validate the effectiveness of our approach, demonstrating its potential for broader applications in fractional control systems.