Joint Forecasting-Scheduling for the Internet of Things

Abstract

We present a joint forecasting-scheduling (JFS) system to be implemented at an IoT Gateway in order to alleviate the Massive Access Problem of the Internet of Things. The existing proposals to solve the Massive Access Problem model the traffic generation pattern of each IoT device via random arrivals. In contrast our JFS system forecasts the traffic generation pattern of each IoT device and schedules the transmissions of these devices in advance. The comparison of the network throughput of Autoregressive Integrated Moving Average (ARIMA) Multi-Layer Perceptron (MLP) and Long-Short Term Memory (LSTM) forecasting models reveals that the optimal choice of the forecasting model for JFS depends heavily on the proportions of distinct IoT device classes that are present in the network. Simulations show that our JFS system scales up to 1000 devices while achieving a total execution time under 1 second. This work opens the way to the design of scalable joint forecasting-scheduling solutions at IoT Gateways. © 2020 Elsevier B.V. All rights reserved.