Dynamic Task Merging Scheduling Method for Intermittent Computing System

With the development of the Internet of Things, intermittent systems that rely on energy harvesting devices as power supply equipment have become a research hotspot. Intermittent computing systems face challenges such as frequent power failures and high state-saving overheads due to the instability of energy harvesting. Traditional task-based programming models, while reducing state backup costs compared to checkpointing techniques, still experience significant performance impacts from non-volatile storage operations during task transitions. To address these issues, this paper proposes a dynamic task merging scheduling strategy based on environmental energy prediction. The method utilizes a lightweight Long Short-Term Memory (LSTM) model to forecast environmental energy levels and dynamically adjusts task merging scales based on historical task execution data. Experimental results on the MSP430FR5994 platform demonstrate that, compared to the InK system, the proposed method shortens the execution time of benchmark programs and reduces system power failures. By dynamically optimizing the task merging strategy, the system reduces read and write operations on non-volatile memory, significantly enhancing task completion efficiency under limited energy conditions in intermittent computing devices.