ESDN: edge computing task scheduling strategy based on dilated convolutional neural network and quasi-newton algorithm

Edge computing systems face the challenges of node heterogeneity and task diversity, and their efficient operation relies on suitable task scheduling methods. Existing task scheduling methods have the problems of difficult application conditions and poor universality because they usually require comprehensive system information and precise system modeling to make excellent scheduling decisions. To address the above problems, this paper proposes Edge Computing Task Scheduling Strategy Based on Dilated Convolutional Neural Network and Quasi-Newton Algorithm (ESDN), which aims to improve the energy efficiency of the edge computing system. The strategy first uses prediction models based on dilated convolutional neural networks to predict the future computing resources requirements of the tasks. The scheduling decisions are then evaluated using a pre-trained deep learning technique based surrogate model. After obtaining the evaluation result, the scheduling decisions are optimized using the Limited-memory BFGS optimization algorithm to obtain better scheduling decisions. This evaluation and optimization process is performed iteratively in a loop to obtain the final scheduling decision with higher quality. ESDN has the advantages of low requirements for information perfection of edge computing system and does not depend on accurate modeling of the system, showing good universality. Simulation experiments based on real data show that ESDN can improve the task completion number of the edge computing system by 27.39%, and reduce the average energy consumption by 18.72% compared with best baseline algorithms in the case of heterogeneous and limited communication resources.