{"id":{"repo_id":"vt","oai_identifier":"oai:vtechworks.lib.vt.edu:10919/137808"},"canonical_url":"https://search.dev.ndltd.org/etd/vt/oai:vtechworks.lib.vt.edu:10919/137808","repository":{"repo_id":"vt","name":"Virginia Tech","base_url":"https://vtechworks.lib.vt.edu/oai/request"},"display":{"title":"Deep Learning Methods for Built Environment Operational Management","abstract":"This dissertation investigated the development of efficient, reliable, and scalable time series (TS) deep learning (DL) frameworks toward enhancing operational management in the built environment, with case studies on (i) reliable infrastructure anomaly detection (AD) and (ii) scalable energy forecasting. An unsupervised, univariate probabilistic anomaly detection framework—DEGAN: Density Estimation-based Generative Adversarial Networks (GANs)—was studied to enhance detection accuracy, with an emphasis on balancing the recall-precision trade-off, using a real-world case study of railroad track monitoring. By leveraging repeated inspection data, employing standalone discriminator models trained solely on normal time series samples, and using kernel density estimation for probabilistic AD, DEGAN achieved a balanced F1 score of 0.83 (R = 0.8 | P = 0.86) and outperformed classical unsupervised machine learning baseline methods. The findings demonstrated the potential of DL architectures to effectively encode domain-specific human knowledge in infrastruc- ture monitoring tasks. The second study extended the univariate DEGAN framework for effective and efficient multivariate time series AD. A flexible framework was introduced to support both one-dimensional (1D) and two-dimensional (2D) DL architectures, including Autoencoders (AEs and VAEs) and GANs. Using this framework, 14 combinations of data embedding techniques (ensemble, reshaping, stacking, TS-to-image conversion) and model types (1D and 2D DL models) were evaluated. Using multi-channel railroad track inspection data, a 2D convolutional AE with channel stacking and a 1D convolutional GAN with reshaping (flattening multi-channel sequences into vectors) were identified as the best-performing models. Both achieved an F1 score of 0.86 and demonstrated higher computational efficiency than classical ML models. Expanding the scope beyond context-specific models, the third study addressed the scalability and generalizability of DL models. Given the need for large and heterogeneous datasets, scalable DL models were studied in the context of energy forecasting tasks through the lens of foundation models (FMs)—large models trained on such datasets. A comprehensive literature synthesis was first conducted on Time Series Foundation Models (TSFMs), which represent promising alternatives to specialist energy forecasting models. The synthesis covered general-purpose TSFMs, including native TSFMs (trained exclusively on TS data) and large language model (LLM)-adapted variants. Using data from more than 1,000 buildings, a comprehensive comparative study was then conducted and showed that GEM (a dedicated FM trained solely on the large energy dataset) and a representative TSFM fine-tuned on the large energy dataset (TimesFM2.0-E) consistently outperformed baseline DL models trained on individual buildings, with zero-shot mean absolute error (MAE) improvements ranging from 16.3% to 7.3% across 24h to 168h horizons. Building-level fine-tuning of these two FMs further increased gains to 17.8%–8.5%, with adaptation times reduced to 11–35 seconds, compared to 301–963 seconds for baselines. Although general-purpose TSFMs exhibited weaker zero-shot performance, all of their building-level fine-tuned variants outperformed baselines. These findings demonstrate the effectiveness of TSFMs—particularly energy-pretrained or domain-adapted models—as scalable and high-performing solutions for building energy forecasting. Together, these studies offer insights into achieving reliable and scalable deep learning in infrastructure operational management, advancing the use of generative artificial intelligence and foundation models in real-world, data-driven built environment management.","abstract_html":"This dissertation investigated the development of efficient, reliable, and scalable time series (TS) deep learning (DL) frameworks toward enhancing operational management in the built environment, with case studies on (i) reliable infrastructure anomaly detection (AD) and (ii) scalable energy forecasting. An unsupervised, univariate probabilistic anomaly detection framework—DEGAN: Density Estimation-based Generative Adversarial Networks (GANs)—was studied to enhance detection accuracy, with an emphasis on balancing the recall-precision trade-off, using a real-world case study of railroad track monitoring. By leveraging repeated inspection data, employing standalone discriminator models trained solely on normal time series samples, and using kernel density estimation for probabilistic AD, DEGAN achieved a balanced F1 score of 0.83 (R = 0.8 | P = 0.86) and outperformed classical unsupervised machine learning baseline methods. The findings demonstrated the potential of DL architectures to effectively encode domain-specific human knowledge in infrastruc- ture monitoring tasks. The second study extended the univariate DEGAN framework for effective and efficient multivariate time series AD. A flexible framework was introduced to support both one-dimensional (1D) and two-dimensional (2D) DL architectures, including Autoencoders (AEs and VAEs) and GANs. Using this framework, 14 combinations of data embedding techniques (ensemble, reshaping, stacking, TS-to-image conversion) and model types (1D and 2D DL models) were evaluated. Using multi-channel railroad track inspection data, a 2D convolutional AE with channel stacking and a 1D convolutional GAN with reshaping (flattening multi-channel sequences into vectors) were identified as the best-performing models. Both achieved an F1 score of 0.86 and demonstrated higher computational efficiency than classical ML models. Expanding the scope beyond context-specific models, the third study addressed the scalability and generalizability of DL models. Given the need for large and heterogeneous datasets, scalable DL models were studied in the context of energy forecasting tasks through the lens of foundation models (FMs)—large models trained on such datasets. A comprehensive literature synthesis was first conducted on Time Series Foundation Models (TSFMs), which represent promising alternatives to specialist energy forecasting models. The synthesis covered general-purpose TSFMs, including native TSFMs (trained exclusively on TS data) and large language model (LLM)-adapted variants. Using data from more than 1,000 buildings, a comprehensive comparative study was then conducted and showed that GEM (a dedicated FM trained solely on the large energy dataset) and a representative TSFM fine-tuned on the large energy dataset (TimesFM2.0-E) consistently outperformed baseline DL models trained on individual buildings, with zero-shot mean absolute error (MAE) improvements ranging from 16.3% to 7.3% across 24h to 168h horizons. Building-level fine-tuning of these two FMs further increased gains to 17.8%–8.5%, with adaptation times reduced to 11–35 seconds, compared to 301–963 seconds for baselines. Although general-purpose TSFMs exhibited weaker zero-shot performance, all of their building-level fine-tuned variants outperformed baselines. These findings demonstrate the effectiveness of TSFMs—particularly energy-pretrained or domain-adapted models—as scalable and high-performing solutions for building energy forecasting. Together, these studies offer insights into achieving reliable and scalable deep learning in infrastructure operational management, advancing the use of generative artificial intelligence and foundation models in real-world, data-driven built environment management.","abstract_has_math":false,"creators":["Gu, Yueyan"],"institution":"Virginia Tech","degree_name":"Doctor of Philosophy","degree_level":"doctoral","degree_discipline":"Civil Engineering","degree_department":"Civil and Environmental Engineering","school":null,"contributors":[],"advisors":[],"committee_chairs":["Jazizadeh Karimi, Farrokh"],"committee_members":["Jia, Ruoxi","Garvin, Michael J.","Sarlo, Rodrigo"],"year":2025,"date_issued":"2025-09-19","date_published":"2025-09-19","updated_at":"2026-07-22T22:19:42Z","subjects":["Deep Learning","Infrastructure Operational Management","Energy Predictive Management","Time Series","Foundation Model"],"languages":["en"],"rights":["In Copyright"],"rights_urls":["http://rightsstatements.org/vocab/InC/1.0/"],"identifier_entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:44321"],"render_values":[{"text":"vt_gsexam:44321","href":null,"code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/10919/137808","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.committeechair","label":"Committee Chair","values":["Jazizadeh Karimi, Farrokh"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Jia, Ruoxi","Garvin, Michael J.","Sarlo, Rodrigo"]},{"key":"dc:contributor.department","label":"Department","values":["Civil and Environmental Engineering"]},{"key":"dc:creator","label":"Author","values":["Gu, Yueyan"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-09-20T08:00:37Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-09-20T08:00:37Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-09-19"]},{"key":"dc:publisher","label":"Institution","values":["Virginia Tech"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Civil Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Virginia Polytechnic Institute and State University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Deep Learning","Infrastructure Operational Management","Energy Predictive Management","Time Series","Foundation Model"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://rightsstatements.org/vocab/InC/1.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:44321"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10919/137808"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["This dissertation investigated the development of efficient, reliable, and scalable time series (TS) deep learning (DL) frameworks toward enhancing operational management in the built environment, with case studies on (i) reliable infrastructure anomaly detection (AD) and (ii) scalable energy forecasting. An unsupervised, univariate probabilistic anomaly detection framework—DEGAN: Density Estimation-based Generative Adversarial Networks (GANs)—was studied to enhance detection accuracy, with an emphasis on balancing the recall-precision trade-off, using a real-world case study of railroad track monitoring. By leveraging repeated inspection data, employing standalone discriminator models trained solely on normal time series samples, and using kernel density estimation for probabilistic AD, DEGAN achieved a balanced F1 score of 0.83 (R = 0.8 | P = 0.86) and outperformed classical unsupervised machine learning baseline methods. The findings demonstrated the potential of DL architectures to effectively encode domain-specific human knowledge in infrastruc- ture monitoring tasks. The second study extended the univariate DEGAN framework for effective and efficient multivariate time series AD. A flexible framework was introduced to support both one-dimensional (1D) and two-dimensional (2D) DL architectures, including Autoencoders (AEs and VAEs) and GANs. Using this framework, 14 combinations of data embedding techniques (ensemble, reshaping, stacking, TS-to-image conversion) and model types (1D and 2D DL models) were evaluated. Using multi-channel railroad track inspection data, a 2D convolutional AE with channel stacking and a 1D convolutional GAN with reshaping (flattening multi-channel sequences into vectors) were identified as the best-performing models. Both achieved an F1 score of 0.86 and demonstrated higher computational efficiency than classical ML models. Expanding the scope beyond context-specific models, the third study addressed the scalability and generalizability of DL models. Given the need for large and heterogeneous datasets, scalable DL models were studied in the context of energy forecasting tasks through the lens of foundation models (FMs)—large models trained on such datasets. A comprehensive literature synthesis was first conducted on Time Series Foundation Models (TSFMs), which represent promising alternatives to specialist energy forecasting models. The synthesis covered general-purpose TSFMs, including native TSFMs (trained exclusively on TS data) and large language model (LLM)-adapted variants. Using data from more than 1,000 buildings, a comprehensive comparative study was then conducted and showed that GEM (a dedicated FM trained solely on the large energy dataset) and a representative TSFM fine-tuned on the large energy dataset (TimesFM2.0-E) consistently outperformed baseline DL models trained on individual buildings, with zero-shot mean absolute error (MAE) improvements ranging from 16.3% to 7.3% across 24h to 168h horizons. Building-level fine-tuning of these two FMs further increased gains to 17.8%–8.5%, with adaptation times reduced to 11–35 seconds, compared to 301–963 seconds for baselines. Although general-purpose TSFMs exhibited weaker zero-shot performance, all of their building-level fine-tuned variants outperformed baselines. These findings demonstrate the effectiveness of TSFMs—particularly energy-pretrained or domain-adapted models—as scalable and high-performing solutions for building energy forecasting. Together, these studies offer insights into achieving reliable and scalable deep learning in infrastructure operational management, advancing the use of generative artificial intelligence and foundation models in real-world, data-driven built environment management."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["This dissertation focused on creating effective and scalable artificial intelligence (AI) mod- els that learn from data, specifically sequential data collected from sensors, to help better manage infrastructure systems through two real-world case studies identifying maintenance needs and predicting energy use in buildings using advanced AI models. A framework called DEGAN was proposed and tested to help detect maintenance needs (anomalies) in railroad tracks using only normal data for developing the models. It was designed to accurately iden- tify real anomalies while avoiding too many false alarms. Using repeated inspection data and training the model only on normal patterns, DEGAN reached a high accuracy score of 0.83. It correctly detected 80% of real anomalies and 86% of its alerts were accurate and per- formed better than traditional models with less complex design. These results showed that the proposed advanced computing model can successfully encode expert knowledge (reflected in learning normal data) to help monitor infrastructure systems. In a follow-up study, the research expanded DEGAN to handle more complex datasets with multiple types of mea- surements using different sensor data streams. A comprehensive framework was proposed that could work with different AI models, including ones that are more compatible with sequential data and others that are compatible with images. The study tested 14 different combinations of (i) methods for data organization and precessing (as input to models) and (ii) AI models, using techniques such as combining, reshaping, and converting time series into images. With real inspection data from railroad tracks, two models stood out: one that stacked data from different sensor streams, and another that reshaped the data into flat sequences. Both models reached an accuracy score of 0.86 and were faster and more efficient than conventional machine learning methods. The third study looked at how to make such models work well at a larger scale and in a wider range of situations by using emerging large AI models—called foundation models—that are trained on very large datasets. Because large and varied datasets were needed, this part of the studies focused on a case study of predicting future energy use in buildings, for which large datasets are available. First, the past research on time series foundation models was studied, in which time series data refers to sequentially ordered data. These models could be used as effective alternatives to tradi- tional energy forecasting methods. The review included both time series–focused models and versions of large language models adapted to work with time series or sequential data. Upon learning the trends from the review study, using data from over 1,000 buildings, several rep- resentative foundation models were developed to study their effectiveness in other buildings. The results showed that two foundation models—one trained from scratch only on energy data and a second generic time series model further updated by energy data—worked better than models trained for each building separately. Even without extra adjustments in each building, they improved accuracy between 7.3% and 16.3% when predicting energy use from one to seven days ahead. When these models were slightly adjusted for each building, their accuracy improved even more (reached between 8.5% and 17.8%) and the adjustment took only 11 to 35 seconds, compared to 5 to 16 minutes for conventional models. Even though the general-purpose models did not perform well directly, once customized for each building, they still outperformed the conventional models. These results show that foundation mod- els, especially those trained on energy data or adjusted for the task, are effective, scalable tools for predicting how much energy buildings will use. Together, these studies offer new perspectives to use advanced AI in managing infrastructure systems, showing how smart, data-driven tools can help solve real-world problems reliably at scale."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Doctor of Philosophy"]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["ETD"]},{"key":"dc:title","label":"Title","values":["Deep Learning Methods for Built Environment Operational Management"]}]}],"canonical_facts":{"dc:contributor.committeechair":["Jazizadeh Karimi, Farrokh"],"dc:contributor.committeemember":["Jia, Ruoxi","Garvin, Michael J.","Sarlo, Rodrigo"],"dc:contributor.department":["Civil and Environmental Engineering"],"dc:creator":["Gu, Yueyan"],"dc:date.accessioned":["2025-09-20T08:00:37Z"],"dc:date.available":["2025-09-20T08:00:37Z"],"dc:date.issued":["2025-09-19"],"dc:description.abstract":["This dissertation investigated the development of efficient, reliable, and scalable time series (TS) deep learning (DL) frameworks toward enhancing operational management in the built environment, with case studies on (i) reliable infrastructure anomaly detection (AD) and (ii) scalable energy forecasting. An unsupervised, univariate probabilistic anomaly detection framework—DEGAN: Density Estimation-based Generative Adversarial Networks (GANs)—was studied to enhance detection accuracy, with an emphasis on balancing the recall-precision trade-off, using a real-world case study of railroad track monitoring. By leveraging repeated inspection data, employing standalone discriminator models trained solely on normal time series samples, and using kernel density estimation for probabilistic AD, DEGAN achieved a balanced F1 score of 0.83 (R = 0.8 | P = 0.86) and outperformed classical unsupervised machine learning baseline methods. The findings demonstrated the potential of DL architectures to effectively encode domain-specific human knowledge in infrastruc- ture monitoring tasks. The second study extended the univariate DEGAN framework for effective and efficient multivariate time series AD. A flexible framework was introduced to support both one-dimensional (1D) and two-dimensional (2D) DL architectures, including Autoencoders (AEs and VAEs) and GANs. Using this framework, 14 combinations of data embedding techniques (ensemble, reshaping, stacking, TS-to-image conversion) and model types (1D and 2D DL models) were evaluated. Using multi-channel railroad track inspection data, a 2D convolutional AE with channel stacking and a 1D convolutional GAN with reshaping (flattening multi-channel sequences into vectors) were identified as the best-performing models. Both achieved an F1 score of 0.86 and demonstrated higher computational efficiency than classical ML models. Expanding the scope beyond context-specific models, the third study addressed the scalability and generalizability of DL models. Given the need for large and heterogeneous datasets, scalable DL models were studied in the context of energy forecasting tasks through the lens of foundation models (FMs)—large models trained on such datasets. A comprehensive literature synthesis was first conducted on Time Series Foundation Models (TSFMs), which represent promising alternatives to specialist energy forecasting models. The synthesis covered general-purpose TSFMs, including native TSFMs (trained exclusively on TS data) and large language model (LLM)-adapted variants. Using data from more than 1,000 buildings, a comprehensive comparative study was then conducted and showed that GEM (a dedicated FM trained solely on the large energy dataset) and a representative TSFM fine-tuned on the large energy dataset (TimesFM2.0-E) consistently outperformed baseline DL models trained on individual buildings, with zero-shot mean absolute error (MAE) improvements ranging from 16.3% to 7.3% across 24h to 168h horizons. Building-level fine-tuning of these two FMs further increased gains to 17.8%–8.5%, with adaptation times reduced to 11–35 seconds, compared to 301–963 seconds for baselines. Although general-purpose TSFMs exhibited weaker zero-shot performance, all of their building-level fine-tuned variants outperformed baselines. These findings demonstrate the effectiveness of TSFMs—particularly energy-pretrained or domain-adapted models—as scalable and high-performing solutions for building energy forecasting. Together, these studies offer insights into achieving reliable and scalable deep learning in infrastructure operational management, advancing the use of generative artificial intelligence and foundation models in real-world, data-driven built environment management."],"dc:description.abstractgeneral":["This dissertation focused on creating effective and scalable artificial intelligence (AI) mod- els that learn from data, specifically sequential data collected from sensors, to help better manage infrastructure systems through two real-world case studies identifying maintenance needs and predicting energy use in buildings using advanced AI models. A framework called DEGAN was proposed and tested to help detect maintenance needs (anomalies) in railroad tracks using only normal data for developing the models. It was designed to accurately iden- tify real anomalies while avoiding too many false alarms. Using repeated inspection data and training the model only on normal patterns, DEGAN reached a high accuracy score of 0.83. It correctly detected 80% of real anomalies and 86% of its alerts were accurate and per- formed better than traditional models with less complex design. These results showed that the proposed advanced computing model can successfully encode expert knowledge (reflected in learning normal data) to help monitor infrastructure systems. In a follow-up study, the research expanded DEGAN to handle more complex datasets with multiple types of mea- surements using different sensor data streams. A comprehensive framework was proposed that could work with different AI models, including ones that are more compatible with sequential data and others that are compatible with images. The study tested 14 different combinations of (i) methods for data organization and precessing (as input to models) and (ii) AI models, using techniques such as combining, reshaping, and converting time series into images. With real inspection data from railroad tracks, two models stood out: one that stacked data from different sensor streams, and another that reshaped the data into flat sequences. Both models reached an accuracy score of 0.86 and were faster and more efficient than conventional machine learning methods. The third study looked at how to make such models work well at a larger scale and in a wider range of situations by using emerging large AI models—called foundation models—that are trained on very large datasets. Because large and varied datasets were needed, this part of the studies focused on a case study of predicting future energy use in buildings, for which large datasets are available. First, the past research on time series foundation models was studied, in which time series data refers to sequentially ordered data. These models could be used as effective alternatives to tradi- tional energy forecasting methods. The review included both time series–focused models and versions of large language models adapted to work with time series or sequential data. Upon learning the trends from the review study, using data from over 1,000 buildings, several rep- resentative foundation models were developed to study their effectiveness in other buildings. The results showed that two foundation models—one trained from scratch only on energy data and a second generic time series model further updated by energy data—worked better than models trained for each building separately. Even without extra adjustments in each building, they improved accuracy between 7.3% and 16.3% when predicting energy use from one to seven days ahead. When these models were slightly adjusted for each building, their accuracy improved even more (reached between 8.5% and 17.8%) and the adjustment took only 11 to 35 seconds, compared to 5 to 16 minutes for conventional models. Even though the general-purpose models did not perform well directly, once customized for each building, they still outperformed the conventional models. These results show that foundation mod- els, especially those trained on energy data or adjusted for the task, are effective, scalable tools for predicting how much energy buildings will use. Together, these studies offer new perspectives to use advanced AI in managing infrastructure systems, showing how smart, data-driven tools can help solve real-world problems reliably at scale."],"dc:description.degree":["Doctor of Philosophy"],"dc:format.medium":["ETD"],"dc:identifier.other":["vt_gsexam:44321"],"dc:identifier.uri":["https://hdl.handle.net/10919/137808"],"dc:language.iso":["en"],"dc:publisher":["Virginia Tech"],"dc:rights":["In Copyright"],"dc:rights.uri":["http://rightsstatements.org/vocab/InC/1.0/"],"dc:subject":["Deep Learning","Infrastructure Operational Management","Energy Predictive Management","Time Series","Foundation Model"],"dc:title":["Deep Learning Methods for Built Environment Operational Management"],"dc:type":["Dissertation"],"thesis:degree_discipline":["Civil Engineering"],"thesis:degree_level":["doctoral"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Virginia Polytechnic Institute and State University"]},"updated_at":"2026-07-22T22:19:42Z"}