A Machine Learning Task Selection Method for Radar Resource Management (Poster)

A radar task selection method is proposed through a machine learning approach in order to improve the scheduling performance. The method initially sorts the tasks according to the importance determined by their dwell times and priorities. More important tasks are selected first until the time window for the task execution is full. A set of reward value is defined based on the initial order of the tasks' importance, then the order of the tasks' importance is changed iteratively according to an award-punishment policy in a reinforcement learning process. Finally, the best group of selected tasks is passed to the earliest start time (EST) algorithm for scheduling. By doing so, the performance of the task scheduling is significantly enhanced. The cost of the schedule is about 2.1 to 5.6 times less, under different overloading situations, than the EST. The proposed method is also very time efficient. A full cycle that including the task selection and scheduling only takes less than 15 ms, thus it is practical.