Towards optimal control of HPV model using safe reinforcement learning with actor–critic neural networks

Abstract

This paper proposes a novel approach that applies state-of-the-art concepts in reinforcement learning (RL) to the optimal control of human papillomavirus (HPV) infection. The methodology transforms the nonlinear optimal control problem into a constrained nonlinear programming problem, thus allowing effective application of the RL algorithms. This approach combines Hamilton–Jacobi–Bellman (HJB) equations with actor–critic neural networks and control barrier functions to obtain an adaptive strategy for optimal vaccination and screening against HPV infection. A key innovation is the Sophia optimizer with experience replay, addressing the critical need for online data application in infectious disease control. Unlike the traditional methods that rely on the accumulation of extensive data, this approach utilizes experience replay to learn and adapt continuously, hence giving practical solutions for diseases like HPV where waiting for data is not practical or desirable. Experience replay helps to store and reuse past experience, hence improving the learning efficiency and stability of the system. This is an important feature for online applications to make sure that an RL model responds quickly enough to changing epidemiological conditions. Numerical simulations demonstrate the effectiveness of this approach in minimizing HPV prevalence and optimizing resource allocation. This research offers significant insights into the application of advanced control strategies in infectious disease management, highlighting the potential of RL to address complex epidemiological challenges. The ability to apply these techniques to online underscores the importance of adaptive and responsive strategies in public health.