Improvement Reinforcement Learning for Solving Multisensor Collaborative Scheduling Problems

The essence of multi-sensor scheduling is a multi-objective optimization problem, which aims to finding the optimal matching relationship between sensors and multiple motion targets within a period time. several literatures proposed multi-sensor scheduling schemes with various reactive algorithms, but the most of them overlook the changes brought to the scheduling scheme by dynamic environments. To address this issue, first an evaluation module between sensor detection advantage, sensor load, and scheduling timeliness are established in dynamic environments. Second, a reinforcement learning algorithm is investigated to improve the collaborative Scheduling. Furthermore, an improvement Actor-Critic algorithm is proposed, which create a centralized evaluation network, that is built to the requirements of task demand. This effectively enhancement the interaction between sensors and the external environment. Finally, simulations show that the models and algorithm proposed in this paper. It strengthens cooperation between sensors, achieve better convergence, reinforce the impact of dynamic environments on sensors, and generates batter multi-sensor scheduling schemes.