Building Design Model for Benchmarking
Energy Efficiency Optimization

1. Introduction

Residential and commercial buildings are responsible for about 40% of primary energy consumption in the United States, and reducing this consumption will play an important role in taking practical steps toward a sustainable society. In particular, the design of the building has a major impact on its energy footprint. That is why design decisions should be optimized during the conceptual stage of a project, since any attempts to improve energy efficiency in the later stages could be more costly or impossible.

In practice, the design of buildings often involves many material-based and geometrical design variables, which impact performance measures such as energy efficiency, cost, comfort, etc. Our goal is to create a realistic benchmark building model with a large design space that can be used to evaluate approaches to energy consumption prediction, as well as approaches to automatic building design optimization.

2. Dataset Description

We aim to create a benchmark building model with considerable number of decision variables to match closely near real-case design scenarios. Our benchmark is based on an open-plan side-lit building plan (OD), shown in Figure above, developed as one of four UK office buildings archetypal models analyzed in [1]. The major drawback of the original model is that many of the design parameters were applied uniformly to the whole building. In practice, designers often assign different building materials, window types, etc to different building parts in order to meet project requirements. In our benchmark, different parameters were assigned to each of the four individual exterior faces. The parameters and their associated values are presented in Table [1]. Moreover, the model has several fixed variables which are listed in Table [2].

Table 1: Parameter list and values for the benchmark building model





Heating Set Point


Cooling Set Point


Building Fabric: Floors


Building Fabric: Roof


Building Fabric: Wall 1


Building Fabric: Wall 2


Building Fabric: Wall 3


Building Fabric: Wall 4


Glazing Ratio: Wall 1


Glazing Ratio: Wall 2


Glazing Ratio: Wall 3


Glazing Ratio: Wall 4


Glazing Coating: Wall 1

{Non-Reflective, Reflective}

Glazing Coating: Wall 2

{Non-Reflective, Reflective}

Glazing Coating: Wall 3

{Non-Reflective, Reflective}

Glazing Coating: Wall 4

{Non-Reflective, Reflective}

Table 2: Fixed parameters for the benchmark building model



Office Occupant Density


Delight Control




Fresh Air Rate


Lighting Power Density


Office Equipment Power Density


All the other parameters in the building model have been fixed to default values. For the meanings and the details of values, please refer to the description in [1]. Our benchmark model has a total of 2,916,000,000 possible designs, forming a very large search space. The model can be downloaded in the Download Section. To evaluate different predictive models for energy consumption in our benchmark, we generate two datasets of size 5000 (train) and 1000 (test) samples using Latin hypercube sampling (LHS), which could be downloaded in the Download Section.

2.1. Objective Functions

In this dataset, two performance measures have been considered to allow multi-objective optimization of the building model and to have a more realistic design scenarios. Although, we have only considered the two conflicting objectives of cost and energy consumption, other objectives such as comfort can be also added to the model with a little bit of effort:

2.1.1. Energy Consumption

Energy consumption is measured as the sum of two components: the cumulative heating load and the cumulative cooling load of the building for the simulation period. In the provided training and testing datasets in the Download Section, these data can be found in the last two columns of the csv files. Moreover, a Matlab script has been provided to get the simulation result of energy consumption for any new design combination.

2.1.2. Cost

The cost for material and labor associated with each option is also an important factor to be considered during the design development cycles. The unit costs for selected parameters were identified based on a widely used construction cost database ( While cost database was not available for many of the materials, approximated unit costs were obtained by interpolating them, as follows:

3. Download


[1] Korolija, I., Marjanovic-Halburd, L., Zhang, Y. and Hanby, V.I. 2013. UK office buildings archetypal model as methodological approach in development of regression models for predicting building energy consumption from heating and cooling demands. Energy and Buildings 60, pp. 152–162.