*Article* **Energy-Efficient Offloading Based on Efficient Cognitive Energy Management Scheme in Edge Computing Device with Energy Optimization**

**Vishnu Kumar Kaliappan 1, Aravind Babu Lalpet Ranganathan 2, Selvaraju Periasamy 3, Padmapriya Thirumalai 4, Tuan Anh Nguyen 1, Sangwoo Jeon 5, Dugki Min 5,\* and Enumi Choi <sup>6</sup>**


**Abstract:** Edge devices and their associated computing techniques require energy efficiency to improve sustainability over time. The operating edge devices are timed to swap between different states to achieve stabilized energy efficiency. This article introduces a Cognitive Energy Management Scheme (CEMS) by considering the offloading and computational states for energy efficacy. The proposed scheme employs state learning for swapping the computing intervals for scheduling or offloading depending on the load. The edge devices are distributed at the time of scheduling and organized for first come, first serve for offloading features. In state learning, the reward is allocated for successful scheduling over offloading to prevent device exhaustion. The computation is therefore swapped for energy-reserved scheduling or offloading based on the previous computed reward. This cognitive management induces device allocation based on energy availability and computing time to prevent energy convergence. Cognitive management is limited in recent works due to non-linear swapping and missing features. The proposed CEMS addresses this issue through precise scheduling and earlier device exhaustion identification. The convergence issue is addressed using rewards assigned to post the state transitions. In the transition process, multiple device energy levels are considered. This consideration prevents early detection of exhaustive devices, unlike conventional wireless networks. The proposed scheme's performance is compared using the metrics computing rate and time, energy efficacy, offloading ratio, and scheduling failures. The experimental results show that this scheme improves the computing rate and energy efficacy by 7.2% and 9.32%, respectively, for the varying edge devices. It reduces the offloading ratio, scheduling failures, and computing time by 14.97%, 7.27%, and 14.48%, respectively.

**Keywords:** edge computing; energy efficiency; reward function; state learning
