A Machine Learning Radar Scheduling Method Based on the EST Algorithm

A machine learning radar scheduling method is proposed based on the earliest start time (EST) algorithm. In this method, the EST algorithm is used to find an initial schedule, and a reinforcement learning approach is conducted to reduce the cost of the initial schedule. In search for a better starting point, the start time of all the tasks are randomly shifted within their allowed time ranges, the shifted tasks are scheduled with the EST again. Then the gradient descent algorithm is applied to further shift the tasks' start times, in order to find an enhanced solution. The procedure is repeated several times. The schedule with the minimal cost is the final solution. The performance of the proposed method is evaluated numerically, showing 1.3 to 10.5 times less cost than the EST, depending on the scenario. In addition, a full cycle of scheduling takes a few tens of milliseconds thus the method could be considered in real radar systems.