Graph Neural Network-based Task Migration Strategy for Battery Run Time Optimization of Smart Medical Device Cluster

The interconnection of smart medical devices is becoming an important trend in healthcare. As the demand for health monitoring continues to grow, the battery run time on which these devices primarily rely is challenged. Due to the differences in task requirements and power reserves of smart medical device clusters, if some devices run out of energy prematurely, it may lead to a decrease in the performance and reliability of the whole system, which makes it particularly important to globally optimize the battery run time state of smart medical device clusters. This paper proposes a task migration strategy for the battery run time collaborative optimization of smart medical device clusters. Firstly, a feature factor of device run time is proposed to evaluate the difference in the distribution of remaining power and pending tasks in the device cluster. Secondly, each device regulates the local scheduling according to the feature factor, so the dominant device reserves more free time for task migration. Finally, a reinforcement learning-based task migration strategy is proposed, where a graph neural network embeds graph features on the feature factor and its associated device features and sub-task features. The task migration actions are selected based on the graph features for collaborative optimization of battery run time for smart medical device clusters. Validated in the distributed medical device lithium-ion battery drive hardware-in-the-loop experimental platform, the results show that in the cluster with balanced battery run time, the proposed method effectively extends the overall battery run time of the cluster by up to 4.35%. In the cluster with unbalanced battery run time, the proposed method can effectively extend the battery run time of the cluster by more than 19.5%.