Next generation fault detection, evaluation, and correction algorithms for HVAC control systems

点击翻译按钮获取中文摘要

Commercial building heating, ventilation, and air conditioning (HVAC) systems in Canada account for 7% of Canada's total energy use and are responsible for 6% of Canada's total GHG emissions. Growing evidence in the literature indicates that about 30% of the energy used by commercial building HVAC systems is wasted due to controls hardware and software faults. In partnership with an established Canadian building smart building energy data analytics software company, CopperTree, this project will eliminate much of this waste through algorithms that will automatically identify, evaluate, and correct these faults. Specifically, the project will develop autoencoder-based fault detection and diagnostics algorithms for sensor faults, surrogate models trained from a large database of energy model simulations for fault evaluation capabilities, and algorithms for auto-correction of faults emerging from inappropriate technician overrides or hardware issue. The algorithms will be developed by using the building performance simulation (BPS) tool EnergyPlus and its Python energy management system environment as a sandbox. Later, they will be tested in a living-lab facility at Carleton University for measurement and verification.Outcomes of the project include a suite of algorithms enhancing data analytics-driven building energy management software solutions. Development and demonstration of the algorithms will be disseminated via journal and conference papers. Three graduate students will be trained with advanced skills in controls and automation of HVAC systems, BPS, and data analytics. The algorithms are expected to reduce heating and cooling energy use in commercial buildings by 20-30%. While these benefits will be immediately available to 2,600 buildings currently using the partner's cloud-based smart analytics software, the benefits will scale to many of Canada's half a million commercial and institutional buildings.