Machine Learning Creates Controls While Computing at a Fast Pace





According to Gregory Pavlak, assistant professor of architectural engineering, control mechanisms for heating, ventilation and air conditioning in buildings follow set parameters to make conditions in a building more comfortable, but what they save on time can reduce efficiency and increase energy costs.

Machine Learning Creates Controls While Computing at a Fast Pace.
Penn State researchers developed a framework for building HVAC controls that could balance efficiency, comfort and cost without requiring significant computation time. Image Credit: iStock/Maxiphoto.

Highly advanced control models, called model predictive controllers, have the ability to improve multiple variables to save on energy, operating charges and carbon emissions but they need a lot more time to find suitable solutions.

The scientists at Penn State have come up with a method that leverages machine learning to make controls that balance building energy cost, comfort and efficiency while computing at a rapid pace. The study outcomes have been reported in the journal energy

Detailed model predictive controllers may not be able to compute solutions fast enough for real-time operations in some buildings. We used machine learning to generate a simple, easily interpretable set of rules for reducing building cooling energy and operating costs—without needing to run model predictive controllers in real time

Gregory Pavlak, Assistant Professor, Architectural Engineering, Penn State

To determine shortcuts to the optimal solutions facilitated by model predictive controllers, the researchers made use of the controllers and gathered data from them. The researchers targeted patterns that constantly resulted in a good performance.

As soon as these patterns were found, the researchers filtered and classified them into categories of control strategies. These categories offered examples of high-performance control strategies that could be utilized to train a machine learning model. Classification trees, producing a set of decision rules, were utilized for the machine learning model to identify the best times of day to cool a building.

Pavlak feels that these rules were very simple to use with basic control hardware, and were easy to interpret.

Once these rules have been generated, they work extremely fast and can be implemented in low-cost controllers and standard building automation equipment. They’re also very interpretable. Operators and engineers can read the rules and easily understand how the systems will behave

Gregory Pavlak, Assistant Professor, Architectural Engineering, Penn State

The rules that were based on the data from model predictive controllers acquired energy efficiency and energy cost levels relative to those of a model predictive controller in action. The best rule sets achieved 95% to 97% of the energy savings and 89% to 92% of the cost objective savings of the elaborate model predictive controller.

These values ​​were obtained with the help of considerably quicker computations: To plan a control strategy for one day, the original model predictive controller needed several hours of computation, while the researchers’ technique could finish the task in less than a second.

The goal of the scientists is to allow highly advanced explorations of this method with future research, such as investigations on a range of buildings, operating conditions and HVAC systems.

The first author of the study was Min Yu, an architectural engineering doctoral candidate.

Journal Reference:

Yum MG, et al† (2022) Extracting interpretable building control rules from multi-objective model predictive control data sets. energy† doi.org/10.1016/j.energy.2021.122691.

Source: https://www.psu.edu/




Leave a Comment

x