Forecasting the usage of appliances of shared use: an analysis of simplicity over complexity


The current revolution of the Internet of Things around the world goes far beyond the goal of simply interconnecting devices and retrieving data from them. New challenges are appearing in relation to the data-flow architectures where Cloud-based initiatives are declining over the Edge computing paradigm. In this latter approach, the resources (i.e. devices involved) are optimized by processing the data as close as possible to the source of them. However, constrained devices still struggle embedding the computation that servers carry out in the Cloud. The case of load and devices usage forecasting is a particular example of this issue where efforts are being made to simplify the processing and device architecture towards reducing energy consumption in data flow and computation. The presented research focuses on the analysis of four typical forecasting models with different levels of complexity that predict the usage of several electrical coffee machines of shared use in office buildings. The results obtained draw on shedding light on the feasibility of embedding simple yet accurate probabilistic models on constrained devices with the aim of saving energy and costs on network infrastructure.