In a smart workplace, creating the ideal room temperature in an energy efficient way is about more than just knowing how many people are in a room at the time. Occupancy based heating and cooling has improved efficiency by reducing energy wasted on empty rooms but uses too much energy adjusting in real-time to dynamic internal and external conditions. To truly be efficient, smart workplaces must not just sense occupants in the present, they should also see into the future.

That’s the aim of an innovative new project at the SMART Infrastructure Facility of the University of Wollongong, Australia, which hopes to create efficiencies by better incorporating forecasting into smart building environmental control. The Building Energy Monitoring Project, part of SMART’s Digital Living Lab, is led by Senior research fellow Dr. Rohan Wickramasuriya, an expert in Geomatics Engineering with over 8 years of experience in the discipline.

"The project’s focus is to increase the efficiency of building environments," Dr. Wickramasuriya clarified. "For instance, by forecasting room temperatures as a function of external and internal conditions, we expect to find that it is more efficient to pump cool night air into a building, rather than turning off the system at 6pm and allowing rooms to heat up due to lack of ventilation."

The goals of the research are to develop a wireless sensor network to monitor the indoor environmental conditions, such as temperature and occupancy, in addition to developing a web-based platform for building control and analytics. This multi-protocol building sensor system will detect occupancy using AI-enabled image recognition, then incorporate forecasts of room temperatures based on historical analysis, outdoor weather, solar radiation on windows and HVAC equipment health.

The traditional method used in commercial buildings “is to start your cooling at 6am and run it until 6pm on a constant flow,” said Dr Wickramasuriya. “There are several obvious problems with this, however. The first is that not many people get in at 6am, the second is that it takes no account of whether the room is occupied, or how many people in the room. The regimes of cooling and heating flows are usually set by an engineer when the building is commissioned and then set in stone,” Wickramasuriya explained.

Accurate estimation of building occupancy is a prerequisite to optimizing both HVAC systems and space utilization. Current systems used for this purpose are only about 66% accurate, while the new image recognition-based system developed in this research is as much as 93% accurate. The researchers expect to finish the project with: a readily deployable, accurate and IoT compliant people counter; an accurate indoor temperature forecasting algorithm; and a prototype vibration sensor.

Occupant privacy concerns are reduced by the project’s internal algorithm, meaning footage never leaves the edge. Furthermore, their video-based analytics system has an accuracy of 92%, beating their own 80% target, and well ahead of the 60% accuracy of traditional laser systems. The true innovation comes from combining real-time sensing with forecasting, however. By predicting how room occupancy and outdoor conditions will change throughout the day, the building can avoid adjusting the temperature if the sun will do the work for it, or if occupants are about to leave the space, for example.

The long short-term memory-based deep neural network algorithms developed in this research are capable of forecasting room temperatures at a much higher accuracy, compared to traditional time series forecasting algorithms. The team is also developing a vibration sensor prototype to identify initial deviations from the standard for any moving part, like a fan motor for example. This will alert building managers to equipment malfunction, long before it becomes a crisis, and all of this information will be available on a web interface for the data that is easily accessible by the user.

The project has attracted funding from Grosvenor Engineering Group and Enviro Building Services, both leading building services providers in Australia, as well as the NSW Government, through the Department of Industry. Anonymized actual building data is collected from equipment maintained by Grosvenor, while Enviro’s office spaces are providing the image data that will be used to train deep neural networks.

“We are expecting to further enhance and innovate within the built environment from the insights gained from this research,” says Rod Kington, national sustainability manager for Grosvenor. “Our main purpose is to make buildings operate more efficiently by driving down inputs from labor, energy, water, and carbon dioxide. The research into deep neural networks moves us down the pathway toward machine learning and artificial intelligence, which will be significant future drivers of value when maintaining building assets.”