Projects

The Argonne Resilience AI Assistant (ARAIA) project is designed to support decision-making in the face of natural hazards and infrastructure challenges. ARAIA is the culmination of a series of strategic partnerships and research funding initiatives that began with support from the Department of Homeland Security (DHS) and has evolved through support from the Department of Energy (DOE) Grid Deployment Office (GDO). This project brings together experts in artificial intelligence and machine learning, operations research, and infrastructure analysis to create an innovative tool that leverages artificial intelligence for improved resilience planning.

ARAIA employs a retrieval-augmented generation (RAG)–based, multi-agent large language model (LLM) architecture designed for scientific reasoning over heterogeneous data sources. The system integrates structured and unstructured inputs from peer-reviewed literature, physics-based simulations, and observational datasets via specialized retrieval, grounding, and reasoning agents. These agents coordinate to perform context-aware inference, uncertainty-aware synthesis, and cross-modal alignment. The initial proof of concept, WildfireGPT, operationalizes this architecture to assess wildfire risk and cascading impacts on public safety and critical infrastructure. Through a natural-language interface, users can query model-grounded metrics (e.g., fire weather indices) and receive decision-support outputs for hazard mitigation, infrastructure risk modeling, and end-to-end wildfire impact assessment.

California traffic has drawn the attention of city planners and researchers for decades. This trend, along with California's dense and complex traffic, makes the city ideal for energy-saving, vehicle-centered traffic management studies.

Our team at Argonne National Laboratory (ANL) accurately forecast large-scale traffic patterns under the umbrella of a larger project on the design and planning of mobility systems led by collaborators at DOE’s Lawrence Berkeley National Laboratory (LBNL).

We developed graph-partitioning-based Diffusion Convolutional Recurrent Neural Network (DCRNN) approach for large scale traffic forecasting. The highway traffic network is represented as a directed graph whose nodes are sensor locations (loop detectors that capture the traffic speed and flow over time), and edge weights are the driving distance between the pairs of sensor locations. The algorithm partitions a graph (large highway network) into subgraphs and trains a DCRNN for each subgraph independently. We demonstrated the efficacy of the graph-partitioning-based DCRNN approach to model the traffic on a large California highway network with 11,160 sensor locations. We developed an overlapping nodes approach for the graph-partitioning-based DCRNN to include sensor locations from other partitions that are geographically close to a given partition.

Furthermore, we demonstrated that the DCRNN model can be used to forecast the speed and flow simultaneously and that the forecasted values preserve fundamental traffic flow dynamics. Our approach of developing DCRNN models that represent large highway networks can be a potential core capability in advance highway traffic monitoring systems. The forecasted results of a trained DCRNN model at all sensor locations can be used to adjust traffic management strategies proactively.

Large-scale highway traffic forecasting approaches are critical for intelligent transportation systems. Recently, deep learning-based traffic forecasting methods have emerged as promising approaches for a wide range of traffic forecasting tasks. These methods are specific to a given traffic network, however, and consequently they cannot be used for forecasting traffic on an unseen traffic network. Previous work has identified diffusion convolutional recurrent neural networks, (DCRNN), as a state-of-the-art method for highway traffic forecasting. It models the complex spatial and temporal dynamics of a highway network using a graph-based diffusion convolution operation within a recurrent neural network. Currently, DCRNN cannot perform transfer learning because it learns location-specific traffic patterns, which cannot be used for unseen regions of a network or new geographic locations.

To that end, we develop TL-DCRNN, a new transfer learning approach for DCRNN, where a single model trained on a highway network can be used to forecast traffic on unseen highway networks. Given a traffic network with a large amount of traffic data, our approach consists of partitioning the traffic network into a number of subgraphs and using a new training scheme that utilizes subgraphs to marginalize the location-specific information, thus learning the traffic as a function of network connectivity and temporal patterns alone. The resulting trained model can be used to forecast traffic on unseen networks. We demonstrate that TL-DCRNN can learn from San Francisco regional traffic data and can forecast traffic on the Los Angeles region and vice versa.

TL-DCRNN outperformed all state-of-the-art traffic forecasting methods such as such as ARIMA, SVR, FNN, LSTM, spatiotemporal graph convolutional network. Allowing practitioners to apply data-driven methods trained on datasets collected elsewhere is a transformative capability, enabling a wide range of transportation system operations and functions to operate more efficiently and sustainably through improved forecasting at reduced infrastructure development and data acquisition costs.

Wide area networking infrastructures (WANs), particularly science and research WANs, are the backbone for moving large volumes of scientific data between experimental facilities and data centers. With demands growing at exponential rates, these networks are struggling to cope with large data volumes, real-time responses, and overall network performance. Network operators are increasingly looking for innovative ways to manage the limited underlying network resources. Forecasting network traffic is a critical capability for proactive resource management, congestion mitigation, and dedicated transfer provisioning. To this end, we propose a non autoregressive graph-based neural network for multistep network traffic forecasting. Specifically, we develop a dynamic variant of diffusion convolutional recurrent neural networks to forecast traffic in research WANs. We evaluate the efficacy of our approach on real traffic from ESnet, the U.S. Department of Energy's dedicated science network. Our results show that compared to classical forecasting methods, our approach explicitly learns the dynamic nature of spatiotemporal traffic patterns, showing significant improvements in forecasting accuracy. Our technique can surpass existing statistical and deep learning approaches by achieving ~20% mean absolute percentage error for multiple hours of forecasts despite dynamic network traffic settings. 

We develop an NLP-driven ML analytic capability to help identify emerging issues of concern for infrastructure assessments, analysis, and planning. The process involved in this capability would include 4 major components:

• Corpus development: COVID-related documents can be extracted from a wide range of sources. As a first step, we focus on extracting the documents from the open-access repository of electronic preprints. In particular, we develop APIs and modules that automatically download the preprints from arXiv, semantic scholar, and other public resources that continuously curate large bodies of research for the scientific and engineering communities. We also explore options for scraping news publications and legal/policy sources in order to expand the domain expertise in the corpus and ingest additional resources.
• Document Classification : Develop machine learning-based multi-label document classifier which categorize documents through supervised learning method. ML model can classify large numbers of unlabeled documents by leverages TF-IDF feature embedding. 
• Active Learning (AL) Pipeline : The efficacy of classification depend on the quantity of labeled data. To that end, we develop an AL pipeline that iteratively identifies documents for user to label.
• Topic Modeling: Generating topics from the categories allows to quickly identify key themes present within each category. To accomplish this, we develop Latent Dirichlet Allocation (LDA) based topic modeling pipeline.