Elsevier

Applied Energy

Volume 277, 1 November 2020, 115584
Applied Energy

Automatic and rapid calibration of urban building energy models by learning from energy performance database

https://doi.org/10.1016/j.apenergy.2020.115584Get rights and content

Highlights

  • An automatic and rapid urban building energy model calibration method was developed.

  • 72 large office buildings with annual usage data were calibrated to test the method.

  • The calibration method adjusted 17 model parameters.

  • It took less than four simulation runs on average to calibrate each model.

  • Full-scale, 72,000, EnergyPlus simulations were performed to validate the algorithm.

Abstract

Urban building energy modeling (UBEM) is attracting increasing attention in the energy modeling filed. Unlike modeling a single building using detailed building systems information, UBEM generally uses limited high-level building stock data to infer default assumptions about building characteristics and operations. This practice inherently brings uncertainty to UBEM. This study introduced a novel method of automatic and rapid calibration of UBEM based on the annual electricity and natural gas energy use data by learning the correlations between crucial model input parameters and the building energy use from the reference building models. A case study was presented to calibrate 72 large office buildings built before 1978 in San Francisco. Seventeen model parameters were selected and Monte Carlo sampling was used to create 1000 samples that reasonably represent the parameter space. Then 1000 simulations were performed for the reference building model to create an energy performance database. The results showed that by learning from the energy performance database, it took less than four simulation runs on average to calibrate a building model. After the calibration, the distributions of each parameter were obtained to replace their single predefined default values. For example, the default lighting power density of 21.39 W/m2 was calibrated to be 7.50 W/m2 on average. The case study successfully demonstrated the effectiveness of the novel calibration method for UBEM in the mild climate. The method will be further tested in future for other climate zones and other building types.

Introduction

Today’s cities consume more than two-thirds of the world’s primary energy and account for more than 70% of global greenhouse gas (GHG) emissions. The building sector in cities can consume up to 75% of the total primary energy. In the United States, half of the commercial buildings [1] and 55% of the residential buildings [2] were built before 1980, while 82% of both commercial and residential buildings were built before 2000. Retrofitting the existing building stock to improve energy efficiency and reduce energy use is a crucial strategy for cities to reduce GHG emissions and mitigate climate change [3].

There are many building energy conservation measures (ECMs), such as enhancing the performance of envelope system (e.g., adding thermal insulation or using high-performance windows), installing photovoltaics on roof [4], improving the energy efficiency of equipment, using natural ventilation, optimizing building operation and controls [5], and promoting energy-saving occupant behavior [6]. The energy saving potential of some ECMs is apparent and easy to evaluate, e.g., replacing the existing lighting system with light-emitting diode (LED) lamps [7]. However, the energy saving potential of some ECMs may significantly vary by climate zones. For example, the external wall thermal insulation performs better in a cold climate than a hot and mild climate [8]. Some ECMs may not save energy when the system is not well designed and operated. For example, if the district cooling system is oversized, the system may operate under inefficient low-load conditions and thus may not save energy. It is, therefore, essential to evaluate the performance of ECMs before implementing them, mainly when the resources are limited [9].

Urban Building Energy Modeling (UBEM) refers to the application of bottom-up physics-based building energy models to predict operational energy use as well as indoor and outdoor environmental conditions for groups of buildings in an urban context [10]. UBEM is an excellent tool to explore opportunities for ECMs when applying to a large group of buildings in the urban context [11]. UBEM can also be used to evaluate the district-scale technologies [12]. Unlike modeling a single building, which will use detailed information, UBEM generally uses existing building stock data consisting of high-level building information. The details of building systems are rarely available for a large number of buildings at the city scale. Typically, the building systems and their efficiency values are determined based on the building type, vintage, size, and climate zone, which represent the average conditions among peer groups [13].

The use of archetype may lead to smaller ranges of site and source energy use intensities (EUIs) distribution for simulated results compared with measured data [14]. Many cities in the United States begin to enforce energy ordinance for existing buildings and provide open data via web portals [15]. Those contained energy data typically at the annual resolution, can be used to calibrate the UBEM and reduce the discrepancy between simulated results and measured data. The model calibration is commonly defined as an inverse approximation because of the need for tuning necessary inputs to reconcile the outputs by a simulation program, as closely as possible to the measured energy data [16]. It is impossible to calibrate the model to be exact as the real building. However, after the calibration, the results are believed to be better than without calibration [17].

There are several automatic calibration methods developed to calibrate individual building, including optimization-based methods [18], pattern or rule-based methods [19], and Bayesian calibration methods [20]. Those calibration methods iteratively adjust the values of selected parameters until the difference between the simulated and measured energy use meets the calibration criteria, which typically requires numerous simulation runs and can be computationally intensive. Those methods were well demonstrated for individual buildings. Although they can be directly applied to calibrate UBEM, the number of simulations required to calibrate the UBEM is proportional to the number of buildings; thus, the total number of simulation runs can be enormous. When simplified reduced-order models (e.g., resistance–capacitance network models) were used in UBEM, each simulation run may only take several milliseconds. So sophisticated calibration methods such as Bayesian calibration can be applied [21]. However, for detailed physics-based energy models, it may take several minutes or more to run a simulation. It is therefore very important to reduce the number of simulations when calibrating physic-based energy models at a large scale (e.g., with hundreds or thousands of buildings).

There are limited studies on UBEM calibration. Nagpal et al. [22] showed a methodology for auto-calibrating UBEM using surrogate modeling techniques. For each building, about 200 EnergyPlus simulations were performed to generate samples for training the surrogate models. The surrogate models can then generate new representative data samples without the need for computationally expensive energy simulations. The surrogate models can be combined with optimization algorithms to determine the combinations of unknown parameters. Sokol et al. [23] presented a Bayesian-based calibration method for defining residential archetypes in UBEM. Three hundred ninety-nine homes with monthly electricity and gas data were used as a training set to generate the probability distributions of uncertain parameters, then applied to a broader test set of 2263 homes to validate the models. Each building was simulated with 1000 samples from the posterior distribution.

Moreover, Cerezo et al. [24] proposed a Bayesian calibration technique to determine the frequency distributions for occupancy, lighting power density, plug load density, and cooling temperature setpoints. This calibration method was applied to 164 residential building with annual energy data, then validated by 159 metered buildings in the same district. Each building was simulated using 100 samples by a Latin Hyper Cube approach. Santos et al. [25] described an automated UBEM calibration process to calibrate five parameters using a genetic algorithm for 56 buildings with monthly electricity data. The parameters were converged after 4680 iterations using the reference rural weather file. When the local weather file generated by the Urban Weather Generator [26] was used, the models took 2650 iterations to converge. Most of the current UBEM calibration studies performed the calibration method for each building separately. However, buildings in the UBEM are usually similar in characteristics and operations. The UBEM calibration should not be the simplified scale-up of the methods used for calibrating individual buildings, which require computationally intensive iterations and monthly or more detailed energy use data usually not available for a large group of buildings.

This paper introduces a novel UBEM calibration method by learning from a pre-simulated performance database of the reference building models to automatically and rapidly calibrate 72 energy models of large office buildings built before 1978 in San Francisco with the annual electricity and natural gas data.

Section snippets

Method

Fig. 1 shows the main steps to calibrate the UBEM in this study. First, we collected the building information of a district in San Francisco. Seventy-two buildings with the annual electricity and natural gas consumption data were selected as target buildings for the UBEM calibration. Secondly, we created the UEBM for the buildings using City Building Energy Saver (CityBES), an open data and computing platform for urban buildings [27]. The geometries of the building energy models are created

Results

The simulation results of the minimum and maximum cases for the updated prototype models and parameter ranges are shown in Fig. 8. The following section shows the simulation results of the reference building model, the number of candidate solutions per building, and the validated solutions per building.

Discussion

As shown in Fig. 11, most of the buildings have multiple candidate models. For each calibration run, different candidate models may be selected, which may result in different parameter combinations after calibration. The problem will be worse when calibrating only a few buildings. In the future, we will perform multiple calibration runs to check whether the parameter distribution can converge to a unique set.

For UBEM calibration, the first step is to adjust the prototype models, so the

Conclusions

The case study successfully demonstrated the effectiveness of the novel urban building energy model (UBEM) calibration method for a group of large office buildings in the mild climate. The main findings of this study are summarized as follows:

  • For UBEM calibration, it is important to perform two simulations with upper and lower limits of key parameters for each building to determine whether the prototype models can represent the target buildings. If not, an adjustment of the prototype models is

CRediT authorship contribution statement

Yixing Chen: Writing - original draft, Conceptualization. Zhang Deng: Resources, Investigation. Tianzhen Hong: Writing - review & editing, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

This paper is supported by the National Natural Science Foundation of China (NSFC) through Grant No. 51908204 and the Natural Science Foundation of Hunan Province of China through Grant No. 2020JJ3008. The work was also supported by Lawrence Berkeley National Laboratory through the Laboratory Directed Research and Development Program and the Assistant Secretary for Energy Efficiency and Renewable Energy, the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.

References (35)

Cited by (41)

View all citing articles on Scopus
View full text