BPNS: A Fast and Accurate Power Estimation Model for Embedded Devices

Abstract

The surge in the intelligent information age has led to a growing need for enhanced computing power in smart mobile devices. However, the accompanying increase in computing power brings about a significant challenge — high power consumption. This poses a notable hurdle for mobile devices that prioritize sustainability performance. Effectively addressing power consumption estimation is crucial for overcoming this sustainability issue. In this paper, we propose using a branching parallel neural networks (BPNS) that consists of FCNN-LSTM and a attention mechanism to estimate the power consumption of a device. In addition, a data stabilization module is incorporated into the framework to improve the power estimation accuracy. We implement and assess the performance of BPNS for online estimation across diverse smart mobile devices, conducting a comparative analysis with commonly used neural networks. Notably, our approach achieves a deviation of 0.28W, 0.65W and 0.76W for online estimation on three smart mobile devices, and outperforms other methods in terms of estimation accuracy and computational cost.