Written by: Sanam Dabirian1 and Ursula Eicker1,2
1 Concordia University, Montreal Canada
2 Canada Excellence Research Chair, Montreal Canada
Overview
The accuracy of urban building energy simulations is highly sensitive to the diverse usage patterns of occupants within buildings. Traditional energy models for urban areas often rely on static schedules for occupants, which do not account for the real variations in occupancy, resulting in unreliable energy demand estimates. A comprehensive model that encapsulates the complexity of occupant behavior is essential for accurate energy demand predictions in urban building energy modeling (UBEM). This study suggests a method for creating occupant profiles that accurately represent this behavior by analyzing time-series data and considering the unpredictable nature of human activity. This article is sourced from “Stochastic-based Occupant-Centric Building Archetype Modelling Using Plug Loads,” Build. Simul. Conf. Proc., vol. 18, pp. 1635–1642, 2023, doi: 10.26868/25222708.2023.1381.
Introduction
In pursuit of a Green Economy, nations are targeting zero emissions by 2050, with Canada aiming for a significant reduction by 2030. Buildings play a crucial role in this effort. UBEM is vital for planning how to reduce emissions and manage energy in urban areas, yet it is limited by its reliance on fixed occupant schedules, which do not mirror the true range of behaviors, resulting in a gap between simulated and actual energy usage.
Building archetypes, which classify urban buildings to support energy policies and improvements, typically assume a set pattern of occupant behavior. However, by incorporating variability and unpredictability into these archetypes, energy simulations can better reflect actual district energy demands. Past research has not fully integrated the variability of occupant behavior into energy models for diverse urban districts.
Therefore, this article introduces a framework to derive representative occupant behavior profiles from time-series data in mixed-use areas, accounting for the randomness of human behavior. This approach improves the accuracy of urban energy models and assists in forecasting energy demand. By factoring in the stochastic nature of occupant behavior, the model enables more precise energy management strategies to be developed for urban neighborhoods, thus underpinning efforts for energy efficiency and informed energy policy-making.
Stochastic Modeling in Urban Energy Simulations
Stochastic modeling of occupant schedules offers a truer reflection of the variability in occupant behavior. Many stochastic models, such as Markov Chain processes, use statistical techniques to predict occupancy behavior. Other approaches have also been applied, like the Monte Carlo and Markov Chain Monte Carlo (MCMC) methods. These methods simulate the variety of inputs needed for urban energy modeling, such as occupant density and appliance usage. The MCMC method is particularly useful for approximating distributions and finding the best-fit curves by sampling from a probability distribution and using random values to approximate real probability distributions.
Methodological Approach
The methodological approach outlined in the study involves a sequence of steps to extract representative occupant-related schedules, such as for plug loads, from electricity consumption data. It allows users to input diverse time-series data and derive profiles useful for simulating building performance. The methodology encompasses data-driven modeling of plug loads and the creation of a stochastic model to generate plug load schedules for a range of building types across different climate zones.
The initial data preprocessing step in data-driven modeling is crucial for analyzing and refining the dataset. Collected raw data often have discrepancies, outliers, and extraneous information. Preprocessing streamlines and enhances the data, ensuring high-quality inputs for further analysis. In this research, preprocessing involves tasks such as cleansing and filling in gaps in the data, reducing its complexity, normalizing the scale of the data, selecting or engineering relevant features, combining datasets from various sources, and categorizing the data into distinct segments.
The study employs k-means clustering to discern patterns and representative occupant behaviors. This clustering method aims to minimize the distance between the data points and their central points in the cluster. The clustering result identifies representative values for occupant characteristics and sorts the profiles into categories based on their similarity to these central points, specific to each type of building within a given climate zone. The cluster outcomes are used to identify typical plug load patterns and compute the transition matrix necessary for constructing the MCMC model. The transition matrix reflects the likelihood of moving from one cluster state to another within the plug load schedules.
The study applies the MCMC algorithm to hourly plug load data. It begins by testing various probability distributions against the hourly data to select the most fitting one, using measures like the Kolmogorov-Smirnov test, Akaike information criterion, and Bayesian information criterion to assess fit quality. The chosen distribution samples hourly plug loads and constructs a daily usage profile. The predicted value for the next hour is influenced by the previous hour’s data, adhering to the Markov Chain transition probabilities. The stochastic values for plug loads are derived by applying the transition matrix to the probability distribution samples, repeating the process to form a complete daily profile.
The accuracy of the developed model is measured using the Root Mean Square Error (RMSE) and R-squared values through a sensitivity analysis, which considers stochastic plug load profiles generated for five selected consecutive weekdays. The Root Mean Square Error (RMSE) assesses the precision of the stochastic plug load profiles in reflecting the true plug load data. The R-squared (R²) value is a statistical indicator that shows how much of the variance in the plug load (the dependent variable) is accounted for by the generated profiles (the independent variable).
Findings
The methodology was tested on office buildings in climate zone 5A. Sensitivity analysis showed RMSE values ranging from 1.15 to 1.45 and R² values from 0.7 to 0.82, indicating a substantial correlation between the model’s output and real data. This implies that while there is some discrepancy in the model’s plug load predictions, it remains within an acceptable margin, and the model explains a significant portion of the actual data’s variance, indicating a substantial correlation between the model’s output and real plug load figures. However, the methodology’s reliance on a single year’s data, with significant fluctuations in data points, poses a risk of creating pronounced fluctuations between consecutive hours in the plug load profiles. Therefore, careful selection of data subsets, such as seasonal or monthly data, for the Monte Carlo simulation inputs is critical.
Conclusion
Accurate UBEM is heavily impacted by occupant-related factors. Relying on standard and static occupant-related schedules in UBEMs leads to a considerable discrepancy between simulated and actual data, emphasizing the need for a more nuanced approach to occupant modeling. The study introduces a stochastic model to create occupant-centric archetypes, tested on office buildings in climate zone 5A. The improved simulation results inform more effective energy efficiency measures and energy savings. The methodology provides a way to infer plug load data from general electricity consumption, offering representative occupant-related profiles for UBEM platforms. This approach represents a significant step towards more accurate urban building energy modeling, better sizing of energy systems, informing policymaking in urban environments, and reducing consumption and greenhouse gas emissions. The extraction of representative plug load profiles across different archetypes and climate zones can also enhance national building codes and standards.
Acknowledgment
This research was undertaken, in part, thanks to funding from the Canada Excellence Research Chairs Program with grant number CERC-2018-00005.