HVAC Systems Optimization
Updated: Nov 8, 2021
Have you ever considered the impact of your building on global climate change? Buildings account for roughly 40% of global energy consumption and 30% of greenhouse gas emissions in the world, with most of the energy consumption being due to heating, ventilation, and air conditioning (HVAC) systems operation. While buildings meet human needs and offer numerous benefits to society, they have also had negative environmental consequences during the last decades.
Considering that we live in the age of data and communication, where almost everything is connected by technology, Bandora uses a data-driven approach to reduce building energy consumption by optimizing the HVAC systems operation.
The weather is one of the most significant factors affecting HVAC systems consumption. Thus, knowing what the weather will be like in the future helps us to anticipate behaviors and take more appropriate actions in the present.
Our algorithm focuses on finding the best combinations of HVAC setpoints based on the current indoor temperature and weather forecast. The main aim is to find the combination that uses the least amount of energy while maintaining a comfortable indoor temperature.
The above requires the use of two distinct predictive models as well as a combinatorial optimization algorithm. Predictive models will be used to forecast indoor temperature and HVAC consumption based on the weather forecast, HVAC setpoints, and indoor conditions. The HVAC consumption will be the objective function of the combinatorial optimization algorithm, and the indoor temperature will be a constraint.
We use optimization algorithms such as hill-climbing, simulated annealing, and evolutionary algorithms to reduce the search space since the computational time required to search all of the different combinations is far superior to the one that we have. Even though these algorithms do not guarantee convergence to the global minimum, the results obtained indicate that we are on the right track.
It is considered a comfortable indoor temperature between 19ºC and 22ºC. Thus, when the weather forecast reveals high temperatures, the model automatically adjusts the HVAC setpoints and the periods of time in which will be working for the ones that guarantee an indoor temperature below or equal to 22ºC, with the least possible energy consumption.
We've been testing this methodology on a fast-food chain building. During the training phase, the models learn to reason about new concepts and experiences based on examples from historical data. This translates into the ability to face a new concept, understand its structure, and then generate reasonable alternative variations of the concept. Following the training phase, we assess the results by comparing them to the originals.
As shown in the figures above, using our fine-tune setup, we can reduce cumulative HVAC consumption while maintaining the indoor temperature within the predefined comfortable range.