Reinforcement learning-based event-driven optimal prevention control strategy for citrus huanglongbing model

Abstract

Citrus Huanglongbing (HLB) is an infectious disease transmitted by Asian citrus psyllids (ACP), which leads to serious economic losses in the citrus industry. Therefore, it is crucial to investigate the prevention and control strategy of citrus HLB. In this paper, the dynamics of HLB propagation between citrus trees and ACP is considered. By applying reinforcement learning (RL) technique, an event-driven optimal prevention control (EDOPC) strategy is developed to ensure the HLB propagation model state converges to a disease-free equilibrium point. Initially, in order to address the challenge of obtaining precise models in practice, a radial basis function-based event-driven observer is built by adopting system input-output data to obtain the approximate HLB propagation model. Subsequently, an EDOPC strategy is devised, which updates only at triggering times to reduce management costs. Additionally, a single critic network structure is constructed to obtain the solution of the Hamilton-Jacobi-Bellman equation, thereby deriving an approximate EDOPC strategy. To align with real-world scenarios, the weights of the observer and the critic network are updated only at event occurrence times. Moreover, by employing the Lyapunov stability principle, the critic network weight error is proved to be uniformly ultimately bounded under the novel event-driven weight adjusting law. Finally, simulation experiments confirm the efficacy of the present RL-based EDOPC strategy.