News - Construction News

Artificial Intelligence and building performance



An integrative machine learning methodology for occupants’ behaviour prediction; an attempt to minimise the energy performance gap.

Building energy simulation models are imperative for the planning, optimisation, and energy performance prediction of buildings. Due to the high level of complexity involved in the process and the pervasiveness of large number of interdependent factors and constraints, as well as the need for advanced mathematical knowledge, achieving a truthful simulation of real-world building performance tends to be challenging.

This challenge also lies in the limitation of present building energy simulation models to include occupant behaviours thoroughly. Most predictive models for building energy performance take occupants into account by prefixed values, deterministic scenarios, and predefined schedules. Consequently, limitations in the ability to predict occupants’ behaviour accurately is partially responsible for the building energy performance gap between predicted and actual building performance. For this reason, it is important, when estimating the building energy performance, to take into account occupant behaviour.

Bridging the energy performance gap is a viable key for energy, design, and construction firms’ decision-making mechanisms, which support exploring various energy options and establishing more reliable energy predictions.

A need arises to predict energy performance with consideration to occupant behaviour. The research aims to address this need and the challenge of the energy performance gap, through providing a better understanding of occupants’ behaviour, and attempting to predict its impact on building energy performance.

The research proposes a novel methodology based on a probabilistic model built on the basis of knowledge derived from occupant behaviour. By employing machine learning techniques, which are capable of handling complex and non-linear problems, more accurate predictions of occupants’ behaviour can be obtained using prior knowledge of occupants to provide insights and defined relationships. Different machine learning algorithms are employed to predict different human actions (i.e opening and closing of windows, shades control, use of appliances, HVAC and thermostat control, movement). This allows the inclusion of behavioural trends, patterns and their impact on energy consumption, which can be further employed as occupant behaviour-specific inputs into constructing the proposed predictive model. The predictive model is then used for co-simulating to predict building energy performance, which eventually leads to more accurate prediction of the building energy performance, thus minimising energy performance gap.

The research is comprised of the following setups:

Theoretical Setup:

This stage provides insights to the occupant-related parameters affecting energy performance identified through previously conducted research and case studies. The stage is summarised as follows:

  1. Identifying the uncertainties and implications of occupants’ behaviour on energy performance prediction.
  2. Defining the inputs related to occupants’ behaviours “occupants’ behaviour markers”.
  3. Evaluating the machine learning algorithms for each of the defined occupants’ behaviour markers by reviewing their application in building energy performance prediction, and which algorithms and techniques can be assessed to determine the best fit to each behavioural marker.

The output of this stage is the input for next stage in which the model will be built.

Practical Setup:

Data will be will be acquired, which will be used to build the Knowledge Base. Selected educational buildings will be used as case study for our research. This stage can be summarised as follows:

  1. Advanced evaluation techniques will be employed to identify behavioural trends and patterns.
  2. Different machine learning algorithms are employed to predict different occupant behaviour markers. This allows the inclusion of the behavioural trends, and their impact on the energy consumption, which will be further employed as occupant behaviour-specific inputs into constructing the proposed predictive model.
  3. A probabilistic modelling tool to co-simulate current building energy simulation tools will be developed.
  4. Model is validated using the empirical data set from the case studies; the model is then tested and compared to the real case scenarios.

The research attempts to provide quantifiable benefits in terms of energy savings. By providing more accurate predictions, studying different alternative design, and operation scenarios, optimized solutions and improved decisions can be made. This promotes energy efficiency in buildings and guide occupants to more responsible behaviour. Thus, a step towards meeting energy efficiency targets. By using machine learning techniques, we will be able to develop a tool that will quantify the impact of occupant behaviour on energy consumption, thus providing more accurate predictions and consequently, bridging the energy performance gap.

This research will be further developed by constructing an expert system that can be used by designers, architects, construction companies and other decision makers, in which a user will provide building-specific data, and (current or prospective) occupants’ characteristics, and the system will return an assessment report highlighting future energy demand, parameters that significantly affect energy consumption, and possible intervention strategies that might reduce the energy consumption.

As a conclusion, this research offers an integrative machine learning approach that aims to provide more reliable energy performance simulation through the inclusion of occupants’ behaviour impact on the energy performance in the simulation model.

Research is being conducted by Rima Alaaeddine, PhD Researcher, School of Art, Design and Architecture at the University of Huddersfield.

If you would like to read more articles like this then please click here.

  •