Energy Consumption Modelling in IoT-surveyed Peacekeeping Missions

Abstract

The Internet of Things (IoT) framework enables the monitoring of power consumption in electrical devices. Different Machine Learning (ML) techniques can be leveraged in this context to perform energy usage prediction. This work presents three main temporal and spatial methods —Long Short-Term Memory (LSTM), Autoregressive Integrated Moving Average (ARIMA), and Besag-York-Mollié (BYM)— to build forecast models for electricity consumption in IoT-surveyed peacekeeping mission camps. The adoption of Deep Learning (DL) to perform forecast tasks has recently dominated the literature in the IoT context. We built the proposed models with a baseline LSTM approach. However, further insights were extracted with other classical methods coming from Statistics. In particular, this work uses the Gaussian adaptation of the BYM model to estimate residuals. With this technique, we aim to enhance accuracy and interpretability. Moreover, we studied the role of neighbourhood features among sensors in increasing the models’ effectiveness and stability. One of the main challenges in this setting is dealing with noisy readings coming from local network issues or user manipulation. This project presents different data transformations to improve data quality, clean incoming outliers, and tailor it for prediction.