In the edge computing environment, when existing task scheduling algorithms allocate resources for tasks, the central host of edge computing consumes more energy and processes fewer tasks successfully. To solve this problem, this paper proposes an edge computing task scheduling mechanism based on multi-dimensional feature extraction and attention fusion (MFEAF). MFEAF achieves efficient fault prediction and fault-tolerant scheduling optimization by integrating graph attention network and temporal network modeling. In order to capture the dynamic dependency relationships between hosts, this paper adopts a multi-level graph neural network architecture that integrates graph convolution and graph attention mechanisms to extract features from the scheduling decisions and time state data of hosts. By dynamically adjusting the learning rate and cosine annealing strategy, redundant transfer is reduced and convergence efficiency is improved. The experimental results show that in terms of fault prediction performance, the F1 score of MFEAF reaches 0.9328. Compared with the latest method, the F1 score of our method has increased by 5.83%, the accuracy has improved by 9.27%, and the recall rate has increased by 2.46%. In terms of energy efficiency and task processing capability, the average energy consumption decreased by 5.0%, and the number of completed tasks increased by 12.0%. In terms of migration efficiency, the average migration time has been reduced by 50%, with a total migration time of only 19.79 seconds, a decrease of 51.3% compared to the suboptimal model. In terms of cost and fairness, containers have the lowest cost and the highest fairness index, reflecting the balance of resource allocation and high cost-effectiveness. In conclusion, MFEAF provides an efficient and adaptive solution for dynamic fault tolerance in edge computing environment.
Zhang et al. (Tue,) studied this question.