Artificial Intelligence for Optimisation and Demand Side Response of Built Environment Management
The project aims to improve aspects of Mitie Energy's existing Building Analytics and enhance its Demand Side Response capability. Both areas are seen as the key part of the Mitie's energy strategy to deliver not only energy-saving solutions, but also a fully-connected workplace in Building Management Systems (BMS).
The project will carry out R&D in two aspects:
- Building analytics (BA) - A big problem with BA is the initial data acquisition and translation of the BMS data flow into the rules of the BA engine. This work today is undertaken manually by skilled operators. This project will develop automated solutions to replace this manual element with mathematical techniques and self-learning algorithms. The project will also investigate the enhancement of the existing database rules with new energy-saving rules that extract the information/knowledge from the noisy data and optimise BMS energy use. This aspect constitutes the majority of the workload, and it comprises 3 main challenges:
- Automatic BMS equipment group clustering and individual equipment consumption profiles.
- Automatic BMS point tagging system using Natural Language Processing approach.
- Advanced BMS rules for reducing energy consumption.
- Demand Side Response (DSR) - this is seen as a high revenue growth area for Mitie Energy, as addressing energy demand under the National Grid DSR programmes is well incentivised. This commercial challenge will be investigated together with the BA challenges. The initial stage of the DSR challenge has been completed and implemented in the Mitie’s infrastructure, based on the paper entitled “Modelling Energy Demand Response Using Long-Short Term Memory Neural Networks” recently submitted for publication.
The PhD project will continue to study both challenges and develop a series of advanced analytic techniques to deliver the satisfactory solutions and system optimisation.
Paper “Modelling Energy Demand Response Using Long-Short Term Memory Neural Networks” submitted to journal.