BLEMAT: Data Analytics and Machine Learning for Smart Building Occupancy Detection and Prediction

Abstract

Running costs of buildings represent a significant outlay for all businesses, thus finding a way to run facilities as efficiently as possible is vital. IoT-enabled Building Management Systems provide means for process and resource usage automation leading to overall efficiency improvements. Inferring spatial and temporal occupancy in all its forms (binary, numerical or continuous) is one of the key contextual inputs required for smart building management systems. In this work, we showcase design, implementation and experimental validation of a smart building occupancy detection and forecasting solution. The presented solution comprises three main building blocks: (1) A fog computing indoor positioning system (BLEMAT — Bluetooth Low Energy Microlocation Asset Tracking) which, combined with wireless access network monitoring processes, produces indoor location information in a semi-unsupervised manner; (2) Data analysis and pattern searching pipelines responsible for fusing data coming from different smart building and networking systems and deriving information on temporal and spatial occupancy patterns; (3) Long short-term memory (LSTM) neural networks trained to predict occupancy patterns in different areas of a smart building. Data analysis and neural network training are conducted on real-world smart building dataset which authors provide in public online repository. Experimental validation confirms that the proposed solution can provide actionable occupancy detection and prediction information, required by smart building management systems.