Dynamic Optimization of Police Dog Task Allocation Based on Reinforcement Learning: An Ethical Intervention Strategy for Stress Reduction

Abstract

This paper presents a dynamic optimization model for police dog task allocation based on reinforcement learning. This model can be regarded as a Markov decision process (MDP), which formally Outlines how the environment and the dog's state are interrelated, as well as the sequence of stress accumulation and relief, and uses state vectors to dynamically adjust the allocation plan. The experimental results show that the proposed method achieves a task completion rate of 94.3%, reduces the required rest intervention by 46-57%, and is an active stress prevention measure rather than a passive threshold enforcement mechanism. Even when the estimation error of the stress-sensitive parameter was 20%, the completion rate was still 92.7%, and the average stress was 0.42. The research provides a feasible approach to balance operational requirements and ethical responsibilities in the management of working animals.