Data Prediction-Based Energy-Efficient Architecture for Industrial IoT

Abstract

This paper presents an energy-efficient industrial Internet of Things (IIoT) architecture that minimizes the data transmission process based on sensor data prediction. While current IIoT network implementations aim to improve lifetime and reduce maintenance costs, existing data prediction studies have primarily focused on prediction performance, disregarding computing time and energy efficiency. In this paper, we propose a data prediction approach on the base station side to maximize the energy efficiency of sensor nodes. A fast deep learning (DL) model is required to achieve low-latency network communications. Therefore, we exploited a DL-based multi-layer perceptron (MLP) with a deep concatenation method called DC-MLP to ensure data prediction reliability and fast computing time. To demonstrate its robustness, we evaluate the proposed DC-MLP model using six performance metrics with k-fold cross-validation. We varied the sampling rate for data prediction to demonstrate the effectiveness of prediction accuracy and energy efficiency. The performance evaluation results revealed that the proposed architecture successfully reduced energy consumption by up to 33% compared with traditional data transmission while maintaining reliable sensor data and achieving an 81% faster prediction time than existing DL models. Based on these findings, the application of the proposed DC-MLP has the potential to increase the sensor lifetime while satisfying the rigorous requirements of the industrial sector, such as fast prediction times, energy efficiency, and reliable prediction results.