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

Area of Science:

• Agricultural Science
• Control Theory
• Machine Learning

Background:

• Citrus Huanglongbing (HLB) poses a significant threat to the citrus industry, causing substantial economic losses.
• The disease is transmitted by Asian citrus psyllids (ACP), necessitating effective prevention and control strategies.

Purpose of the Study:

• To develop an event-driven optimal prevention control (EDOPC) strategy for citrus HLB using reinforcement learning (RL).
• To ensure the HLB propagation model converges to a disease-free equilibrium point.

Main Methods:

• An approximate HLB propagation model was created using a radial basis function-based event-driven observer with system input-output data.
• An EDOPC strategy was devised, updating at triggering times to minimize management costs.
• A single critic network structure was employed to solve the Hamilton-Jacobi-Bellman equation for an approximate EDOPC strategy.

Main Results:

• Observer and critic network weights were updated at event occurrence times, aligning with real-world conditions.
• The Lyapunov stability principle was used to prove uniform ultimate boundedness of the critic network weight error under the event-driven weight adjusting law.
• Simulation experiments validated the effectiveness of the RL-based EDOPC strategy.

Conclusions:

• The developed RL-based EDOPC strategy is effective in controlling citrus HLB propagation.
• The event-driven approach offers a cost-efficient method for managing infectious diseases in agriculture.
• This research provides a novel framework for applying advanced control techniques to agricultural pest and disease management.