Power Estimation Models for Edge Computing Devices

The increasing demand for energy-efficient solutions in IoT devices and edge computing calls for novel methodologies to generate accurate power models for diverse devices, enabling sustainable growth and optimized performance. This paper presents a methodology for creating power models for edge devices and their embedded components. The proposed methodology collects power and resource utilization measurements from the edge device and generates both additive and regression models. The methodology is evaluated on a Raspberry Pi 4 device using a smart plug for power monitoring and various benchmarking tools for CPU and network sub-components. The evaluation shows that the generated models achieve low error, demonstrating the effectiveness of the proposed approach. Our methodology can be applied to any edge device, providing insights into the most efficient power consumption model. The heterogeneity of edge devices poses a challenge to creating a global power model, and our approach provides a solution for developing device-specific power models. Our results indicate that the generated models for Raspberry Pi 4 scored a maximum of 8% MAPE.