About

Farrokh Jazizadeh is an Associate Professor in the Charles E. Via, Jr. Department of Civil & Environmental Engineering at Virginia Tech. His areas of interest include data-driven urban infrastructure management, infrastructure operational analytics, infrastructure-oriented data sensing and analytics, and the development of responsive and adaptive built environments. His research focuses on applying machine learning and statistical inference to cyber-physical systems, emphasizing sustainability and innovative sensing techniques. He holds multiple advanced degrees from the University of Southern California, including a Ph.D. in Civil Engineering and two M.Sc. degrees in Computer Science and Civil Engineering, obtained in 2015, 2013, and 2011 respectively. Additionally, he earned a M.Sc. in Civil Engineering from Amirkabir University of Technology in Tehran, Iran, in 2004. Jazizadeh is actively involved in teaching courses such as Computer Applications in Civil & Environmental Engineering and Applied Data Sensing and Management for Built Environment, among others. His scholarly work is documented on his Google Scholar profile and his research lab website, INFORM Laboratory.

Research topics

• Computer Science
• Artificial Intelligence
• Engineering
• Architectural engineering
• Human–computer interaction
• Knowledge management
• Management science
• Thermodynamics
• Statistics
• Mechanical engineering
• Psychology
• Mathematics
• Industrial engineering
• Physics

Selected publications

• Cross-Modal Adaptation of Large Language Models for Building Energy Forecasting

  2026-01-28

  articleSenior author

  Energy forecasting research has traditionally focused on individual buildings or clusters of similar facilities with limited progress on generalized models applicable across diverse building types. However, growing demands and greater variability in energy use patterns call for scalable and efficient solutions to enhance energy management. Barriers to more scalable solutions lie in the limited availability of large-scale, high-quality datasets, the heterogeneity of data attributes across different sources, and modeling efficiency. To this end, this study investigates the application of transfer learning through the cross-modal adaptation of pretrained large language models (LLMs), specifically GPT-2, for building energy forecasting. By fine-tuning GPT-2, the model is adapted to handle multivariate time series data and predict energy consumption across a dataset of 100 commercial buildings. The impact of input sequence length on forecasting performance was also evaluated, with results indicating that a 168-h (1-week) input length yields better performance. Moreover, comparative analyses reveal that the GPT-2-based approach achieves performance comparable to or exceeding that of state-of-the-art energy forecasting models while maintaining computational efficiency. These findings highlight the potential of cross-modality adaptations of pretrained large models, such as GPT-2, in advancing predictive energy management for the built environment as scalable and adaptable solutions.

  PublisherDOI

• LLM-Based Building Energy Management Assistants

  2025-12-11

  articleSenior authorCorresponding

  This study introduces a novel framework for Large Language Model (LLM)-based Building Energy Management Assistants (BEM Assistants), designed to address the limitations of traditional energy management interfaces by harnessing the advanced capabilities of AI, particularly through the use of LLMs, such as the Generative Pre-trained Transformer (GPT). This framework integrates historical energy consumption data analysis and smart building IoT control by leveraging connections with external devices to enhance operational efficiency and responsiveness in energy management systems. A prototype of a BEM Assistant for residential energy management was developed to demonstrate its practical application and evaluate it through simulation using 95 example user queries included in three categories: information retrieval, data analysis for recommendation, and external device interactions. The results highlight the framework’s feasibility in the effective processing of energy information queries, offering energy consumption insights and executing smart building controls. This study introduces the application of the OpenAI Assistants API in evaluating an Assistant’s capability in building energy management, showcasing the potential of LLMs to improve building energy efficiency and management practices. This research lays a framework for future exploration into AI-powered building energy management, with promising results showing the potential to bridge the gap between traditional management techniques and the evolving advancements of AI and smart building technologies.

  PublisherDOI

• Toward Large Energy Models: A comparative study of Transformers’ efficacy for energy forecasting

  Applied Energy · 2025-01-29 · 4 citations

  articleCorresponding
  PublisherDOI

• Toward Generalized CO ₂ Prediction Modeling in Educational Buildings

  2025-12-11

  articleSenior authorCorresponding

  Indoor air quality (IAQ) is crucial for occupants’ well-being, health, and productivity, especially within educational buildings. One of the critical factors in indoor air quality is CO2 concentration. A healthy level of CO2 enhances students’ concentration, cognitive function, and productivity. Sensing the temporal variations of the CO2 concentration in educational buildings could help design effective mitigating strategies, but it could be costly. Evaluating the feasibility of estimating CO2 concentrations based on room characteristics, HVAC operational attributes, and occupancy states could aid in reducing sensing costs. Accordingly, this paper seeks to address two main objectives: (1) feasibility assessment of predicting CO2 concentration based on space/room’s characteristics, states, and HVAC operational attributes; and (2) feasibility assessment of training models on a limited number of indoor spaces while applying them to CO2 level estimating in unseen environments. To address the objectives, we utilized a publicly available data set collected in April 2022 on a university campus, including variables such as occupancy schedules, supply air rates, and CO2 concentrations in various spaces, such as classrooms, offices, and lecture halls. Our results demonstrate the feasibility of our objectives, with ensemble learning models, specifically random forest and gradient boosting, outperforming other methods and showing promising accuracy with a mean absolute error (MAE) level of 19 ppm. Finally, we evaluated the generalizability of a trained model (on a limited number of data points) to other spaces with promising performance measured by MAE compared to the ground truth. These findings can inform scalable Co2 modeling for devising mitigating strategies in balancing indoor air dynamics by using a range of solutions from mechanical ventilation to utilizing indoor plants.

  PublisherDOI

• Context-aware LLM-based AI Agents for Human-centered Energy Management Systems in Smart Buildings

  arXiv (Cornell University) · 2025-12-31

  preprintOpen accessSenior author

  This study presents a conceptual framework and a prototype assessment for Large Language Model (LLM)-based Building Energy Management System (BEMS) AI agents to facilitate context-aware energy management in smart buildings through natural language interaction. The proposed framework comprises three modules: perception (sensing), central control (brain), and action (actuation and user interaction), forming a closed feedback loop that captures, analyzes, and interprets energy data to respond intelligently to user queries and manage connected appliances. By leveraging the autonomous data analytics capabilities of LLMs, the BEMS AI agent seeks to offer context-aware insights into energy consumption, cost prediction, and device scheduling, thereby addressing limitations in existing energy management systems. The prototype's performance was evaluated using 120 user queries across four distinct real-world residential energy datasets and different evaluation metrics, including latency, functionality, capability, accuracy, and cost-effectiveness. The generalizability of the framework was demonstrated using ANOVA tests. The results revealed promising performance, measured by response accuracy in device control (86%), memory-related tasks (97%), scheduling and automation (74%), and energy analysis (77%), while more complex cost estimation tasks highlighted areas for improvement with an accuracy of 49%. This benchmarking study moves toward formalizing the assessment of LLM-based BEMS AI agents and identifying future research directions, emphasizing the trade-off between response accuracy and computational efficiency.

  PublisherDOI

• Context-aware LLM-based AI Agents for Human-centered Energy Management Systems in Smart Buildings

  ArXiv.org · 2025-12-31

  articleOpen accessSenior author

  This study presents a conceptual framework and a prototype assessment for Large Language Model (LLM)-based Building Energy Management System (BEMS) AI agents to facilitate context-aware energy management in smart buildings through natural language interaction. The proposed framework comprises three modules: perception (sensing), central control (brain), and action (actuation and user interaction), forming a closed feedback loop that captures, analyzes, and interprets energy data to respond intelligently to user queries and manage connected appliances. By leveraging the autonomous data analytics capabilities of LLMs, the BEMS AI agent seeks to offer context-aware insights into energy consumption, cost prediction, and device scheduling, thereby addressing limitations in existing energy management systems. The prototype's performance was evaluated using 120 user queries across four distinct real-world residential energy datasets and different evaluation metrics, including latency, functionality, capability, accuracy, and cost-effectiveness. The generalizability of the framework was demonstrated using ANOVA tests. The results revealed promising performance, measured by response accuracy in device control (86%), memory-related tasks (97%), scheduling and automation (74%), and energy analysis (77%), while more complex cost estimation tasks highlighted areas for improvement with an accuracy of 49%. This benchmarking study moves toward formalizing the assessment of LLM-based BEMS AI agents and identifying future research directions, emphasizing the trade-off between response accuracy and computational efficiency.

  PublisherOA PDF

• Are Transformers Effective for Time Series Anomaly Detection? A Case Study in Building Energy Management

  2025-12-11

  articleSenior authorCorresponding

  Analyzing smart meter data to identify irregularities in buildings’ operating conditions has the potential to lead to substantial energy savings. Prediction-based anomaly detection, a self-supervised machine learning method, proves effective in automating this process. This methodology involves training a model on historical data, enabling it to predict the future state of a given time series. Anomalies are detected when observed data points deviate from these predictions. Transformers have shown remarkable performance in long-range sequential data processing, credited to their attention mechanisms and parallel computing paradigm. However, there has been limited comparative research conducted on prediction-based anomaly detection relying on Transformers in the domain of energy management. To address this gap and assess efficiency and effectiveness, a case study was conducted using a public data set on energy anomalies in buildings. Following a discussion on the impact of data contamination and sequence size on Transformer-based anomaly detection, a comparative analysis was conducted using the data from 200 buildings. The findings revealed that a Long Short-Term Memory (LSTM) Recurrent Neural Network achieved a marginally higher F1 score compared to a Transformer model. However, the Transformer model demonstrated significantly faster processing, consuming merely one-ninth of the time required by the LSTM model. The superior advantage of the Transformer model in efficiency indicates its future potential for real-world practice in addressing energy management problems with large models built upon massive web-scale data.

  PublisherDOI

• Digital Twin-Assisted Anomaly Detection for Smarter Built Environment Management: A Literature Review

  2024-03-18 · 1 citations

  reviewSenior authorCorresponding

  Improving the reliability and efficiency of anomaly detection is important for establishing and maintaining a resilient built environment. Meanwhile, Digital Twins (DT) have attracted increasing attention in the built environment management. In this study, through a structured literature review, we compiled and categorized existing DT-assisted anomaly detection methods applied in built environment based on three anomaly detection criteria (scope, target, approach) and three key attributes of DT (integration level, fidelity level, and decision speed). In doing so, we presented the current research trends and analyzed how DTs contribute to the built environment management with different anomaly detection approaches. This study lays the foundation for establishing DT-assisted anomaly detection frameworks utilizing Digital Twins of different integration and fidelity levels to tackle anomaly detection problems with different approaches. Moreover, it shows the efficacy of Digital Shadows (one directional DTs) in built environment anomaly detection and calls for future research endeavors in high-integration-level DT applications and DT-assisted hybrid anomaly detection methods.

  PublisherDOI

• A longitudinal study of IAQ metrics and the efficacy of default HVAC ventilation

  Building and Environment · 2024-02-29 · 17 citations

  articleSenior author
  PublisherDOI

• Trust in Human-AI Interaction: Review of Empirical Research on Trust in AI-Powered Smart Home Ecosystems

  2024-01-25 · 2 citations

  articleSenior authorCorresponding

  The growing socio-technical influence of artificial intelligence (AI) is increasing in various dimensions of our daily lives. The monitoring and decision-making role of smart home AI agents could go beyond simple operations, and a lack of users’ trust could lead to the underuse or misuse of their capabilities. As such, lying at the intersection of trust in human-AI interactions and smart home ecosystems, this paper attempts to explore the driving factors of building trustworthy AI-powered smart home ecosystems. To this end, through a systematic literature review, we have identified the characteristics of human-AI trust in intelligent environments, factors that can influence users’ trust in AI, features that can be implemented to improve users’ trust, and challenges and opportunities that need to be considered. The findings can help develop a pathway for building long-term users’ trust in AI-powered smart homes for integrated human-AI interactions.

  PublisherDOI

Frequent coauthors

• Burçin Becerik-Gerber

  32 shared

• Milad Afzalan

  Impact Technology Development (United States)

  17 shared

• Arsalan Heydarian

  Engineering Systems (United States)

  17 shared

• Wooyoung Jung

  University of Arizona

  14 shared

• Geoffrey Kavulya

  University of Southern California

  11 shared

• Mahsa Pahlavikhah Varnosfaderani

  11 shared

• Milind Tambe

  8 shared

• Laura Klein

  University of California, San Francisco

  8 shared

Education

• Ph.D., Astani Department of Civil and Environmental Engineering

  University of Southern California

  2015

• MS.C., Department of Computer Science

  University of Southern California

  2013