Selected Publications

MEBA: AI-powered precise building monthly energy benchmarking approach (Published)

Authors: Tian Li∗, Haipei Bie, Yi Lu, Azadeh Omidfar Sawyer, Vivian Loftness

Monthly energy benchmarking supports identifying trends, improving energy efficiency, and conducting cost management for building owners, managers, and policymakers better than annual or hourly benchmarking. Annual data cannot fully reflect operation utility status, and hourly data poses the issue of high-cost data mining and incomparability due to its minor scale. However, the primary challenges of monthly energy benchmarking are data limitation, “black-box” barrier, and building classification uncertainty. This study proposes a novel AI-powered Monthly Energy Benchmarking Approach (MEBA) to better assess building energy use patterns, benchmark end-use loads, and track utility bills. MEBA addresses two scenarios: (1) predict complete year-round monthly energy using partial monthly energy data; (2) estimate monthly energy loads from annual total energy data. The study collects monthly electricity and natural gas energy use from two U.S. cities. For the first scenario, the entire dataset is clustered into two primary groups by Gaussian Mixture Model (GMM). Then, the two groups are divided by Self-Organizing Map (SOM) models into five subclusters via energy use patterns. For the second scenario, an additional step is needed to locate the subcluster labels with advanced Light Gradient Boosting Machine (LGBM) classifications. All five subclusters have high prediction performance with an average accuracy of >95%. Both scenarios require the last stage to predict monthly electricity and natural gas by LGBM regressions. MEBA's prediction performance achieves R2s ranging from 0.50 to 0.73, with RMSEs between 0.15 and 2.35, outperforming the state-of-the-art XGBoost model. Each subcluster exhibits distinct energy use patterns, with EUIs, electricity loads, and year built as the most significant attributes.

Generalized Building Energy and Carbon Emissions Benchmarking with Post-Prediction Analysis (Published)

Author: Tian Li*, Tianqi Liu, Azadeh Omidfar Sawyer, Pingbo Tang, Vivian Loftness, Yi Lu, Jiarong Xie

Developing a generalized building energy and carbon emissions benchmarking tool that can be applied nationwide is critical for building performance, policy-making, and realistic decarbonization goals. However, the lack of enacted energy benchmarking programs in the majority of U.S. cities hinders progress in establishing effective benchmarking. Meanwhile, advanced machine learning models pose challenges in revealing the performance and associated building or climate attributes. This study proposes a generalizable building energy and carbon emissions benchmarking approach that is applicable to any contiguous U.S. city. The method is based on eleven years of real-world data from twelve U.S. cities across seven ASHRAE climate zones with comprehensive weather data. The collected data is fed into an advanced tree-boosting ensemble learning model to predict the building site, source energy, as well as carbon emissions. The generalizability performance is validated by isolating seven climate zone separately with R2s varying from 0.3 to 0.9. Additionally, a post-prediction approach is first proposed by this study to gain insights into the prediction performance. The actual and predicted values are classified into well-estimated, underestimated, and overestimated groups by an 85% statistical acceptance interval. This method assesses prediction performance associated with energy, building types, and climate zones. The results show that a reliable generalized benchmarking tool can be achieved. Furthermore, it's been observed that energy display for most underestimated groups tends to be higher than that for their overestimated counterparts. More than 70% of the buildings perform within the well-estimated group, and some building types outperform significantly than others.

VOD: Vision-Based Building Energy Data Outlier Detection (Published)

Author: Jinzhao Tian, Tianya Zhao, Zhuorui Li, Tian Li, Haipei Bie, Vivian Loftness

Outlier detection plays a critical role in building operation optimization and data quality maintenance. However, existing methods often struggle with the complexity and variability of building energy data, leading to poorly generalized and explainable results. To address the gap, this study introduces a novel Vision-based Outlier Detection (VOD) approach, leveraging computer vision models to spot outliers in the building energy records. The models are trained to identify outliers by analyzing the load shapes in 2D time series plots derived from the energy data. The VOD approach is tested on four years of workday time-series electricity consumption data from 290 commercial buildings in the United States. Two distinct models are developed for different usage purposes, namely a classification model for broad-level outlier detection and an object detection model for the demands of precise pinpointing of outliers. The classification model is also interpreted via Grad-CAM to enhance its usage reliability. The classification model achieves an F1 score of 0.88, and the object detection model achieves an Average Precision (AP) of 0.84. VOD is a very efficient path to identifying energy consumption outliers in building operations, paving the way for the enhancement of building energy data quality, operation efficiency, and energy savings.

An Innovative Building Energy Use Analysis by Unsupervised Classification and Supervised Regression Models (Published)

Author: Tian Li*, Jiarong Xie, Tianqi Liu, Yi Lu, Azadeh Omidfar Sawyer

The traditional building energy use analysis classification is typically conducted according to the primary uses. However, some buildings with different primary uses have similar energy use patterns. This is because multiple attributes, other than building types, impact energy. Also, the same building type may have significantly different energy use patterns due to factors such as building age, size, equipment, and climate conditions. In order to better improve the performance of building energy benchmarking, this study employs the unsupervised K-Means++ model to classify the building energy benchmarking with comprehensive weather data from four cities in ASHRAE climate zones 3A, 4A, 5A, and 6A under five years from 2016 to 2020. During the classification process, the Elbow and Calinski Harabasz metrics are conducted to optimize the number of clusters. After classifying the total dataset, this study splits the data of each cluster into training and test sub-datasets for training and testing. Each test dataset is isolated by the same year in each climate zone. Then, LGBM tree-boosting ensemble regression model is applied to evaluate the model performance for each cluster. During the training process of each cluster, the Grid-search with Cross-validation is operated to optimize the hyperparameters and improve the running efficiency. The results show that each classified model by K-Means++ performs well, all achieving more than 80% R2s for actual and predicted values. Furthermore, the building clusters are significantly different from the primary use classification. In addition, the LGBM model features positive and negative impacts on the building energy use intensity of each cluster is also investigated, giving a deeper scope in understanding the model performance for future studies.

Subjective Impression of an Office with Biophilic Design and Blue Lighting: A Pilot Study (Published)

Author: Jiarong Xie*, Siqing Ge, Azadeh Omidfar Sawyer, Tian Li

This paper investigates and compares people’s subjective impression of an office with a biophilic design and blue lighting. Existing studies have examined their influence on perception separately, but how they compare is unclear. Additionally, only a few studies have used an office setting as a case study. To address this research gap, this study collected people’s ratings and rankings of four simulated interior scenes of a private office using an online survey. The scenes include blue lighting, a biophilic design with daylight and view, a biophilic design with indoor plants, and a non-biophilic baseline with conventional white lighting. A total of 284 complete responses were collected and analyzed using a mixed-effect model. It was found that the two biophilic designs improved people’s perception of the office compared to the base case. The biophilic design with access to daylight and view outperformed the space with indoor plants in all the examined perceptual categories, specifically how the office space was perceived by participants as brighter, more comfortable, and spacious. On the contrary, the space with blue lighting decreased people’s ratings in most perceptual attributes in comparison to the baseline. The negative influence was notably significant in how lively, comfortable, bright, and appealing the space was perceived as being by participants. Subjects’ preference rankings of the four simulated office spaces showed a similar pattern.

DUAL-FEED: AI-Driven Building End-Use Energy Benchmarking Classification and Forecasting (Under-Review)

Authors: Tian Li∗, Jiarong Xie, Tianqi Liu, Yi Lu, Azadeh Omidfar Sawyer, Vivian Loftness, Pingbo Tang

Data-driven building energy benchmarking is one of the most efficient ways to save energy, reduce carbon emissions, and improve energy management. Data limitation, especially for the end-use loads, for most areas in the United States hits the first challenge. Also, current benchmarking classifications are typically conducted by building types or climates. However, multiple factors impact energy consumption, leading to the identical building type may have significantly various energy use patterns. Consequently, this study proposes a DUAL-FEED approach to develop an innovative energy benchmarking classification and end-use energy forecasting approach nationwide…

A Review of Hybrid Mode of Inpatient Care and Homecare Design Based on IoMT Technology (Published)

Author: Tian Li*, Yi Lu

Since 2014, the Internet of medical things (IoMT) has been fast developed, which carried the following healthcare industry revolves around the world. This revolution will not only improve healthcare quality tremendously for both patients and givers but also advance the interrelationship between human behavior and the built environment. The United States owns the most advanced healthcare system worldwide, however, it has been facing many tough issues, such as facility limitation, healthcare insurance, and healthcare accessibility, etc. With the help of IoMT, the healthcare system is transferring from inpatient hospital-centered mode to hybrid mode (home care and inpatient care). With this process, the healthcare system can be greatly improved, where healthcare givers and patients are supported by their health assistant devices to facilitate a healthier life, a sustainable environment, and good behaviors.

Thesis from Washington University: Abandoned House Rethinking: The Soul of Burnt Violin

Author: Tian Li

Abandoned houses are more than 600 in some north City neighborhoods of St. Louis. The population loss is more than 20% over the past 5 years in these areas and abandoned houses are growing. Residents passed away due to bad wellness status and their inheritors had already moved out to live somewhere else. According to research by Virginia Commonwealth University (VCU) in 2016, People who live in different parts of north St. Louis may have a 12-year difference in how long they can expect to live. Many residents with low perceived education of health care believe illness will improve with time or on its own. Guiding everyone to access hospitals when they feel sick is difficult because of low perceived educational background, but it is much easier to foster the younger generation's healthy behavior, which could have a good influence on the community. This project aims to provide primary education and healthcare service program to foster individual health and help communities thrive again.