A Federated Learning System for Optimised Environmental Control of Consecutive Areas

Abstract

Building management systems have led to artificially controlled environmental conditions. While new infrastructure will lead to a lower carbon footprint and a better working environment, the costs cannot justify the rewards at this present time. Ambient temperature regulation has the potential to mitigate excessive energy consumption. This work proposes an externally influenced environmental control Artificial Neural Network (ANN) implementation to optimise ambient temperature for a given specific area whether that be internal or external to the building. The approach uses a multi-aspect ANN. Two architectural components are introduced, an Agent ANN (A-ANN) and a Coordinating ANN (C-ANN). The Agent ANNs (A-ANN) are deployed to provide temperature control at the extremities of the open plan area. The A-ANN operates with a degree of autonomy. A Coordinating ANN (C-ANN) considers the optimal ambient temperature of the room and consequently effects the surrounding area(s). These A-ANNs have internal and external factors acting as effectors to the system such as outdoor environmental conditions and internally located effectors such as adjacent rooms. Results are presented which diagnose the effort applied by A-ANN instances in varying environmental conditions both internally and externally.

Privacy preserved and decentralized thermal comfort prediction model for smart buildings using federated learning

“…A novel use of the federated learning framework is provided in research Li et al (2020) , Perry & Fallon (2019) , with an intelligent building energy forecast scenario as the focal point. Deep neural networks are employed with data from the ASHRAE to implement and assess this design.…”