Development of a First-Generation Community-Resilience Systems Model

Objective
Develop an efficient, interactive computer-based tool, Alternatives for Resilient Communities (ARC), to facilitate exploration of decision alternatives at the community scale for infrastructure systems (buildings, water, power, transportation) for a range of natural hazard events and evaluate its use by a community. 
 

Technical Idea
This project will develop a new systems model and tool called NIST ARC (Alternatives for Resilient Communities) to help communities identify viable solutions that can serve as starting points for further refinement and testing. This will be accomplished with application of operations research (OR), specifically mathematical programming, which is uniquely suited to the solution of large-scale problems. Mathematical programming, which has a storied history dating from World War II, and which has application in every major industry, has grown to exploit computing advances such as high performance and cloud-based computing, and, now, AI infrastructure (e.g., GPUs), enabling its application to ever larger problems. The project research that will be pursued involves specifying the community resilience problem in a particular form (“mathematical program”) so as to be solvable within known measurement precision using well-established solution methods. Getting the problem into a solvable form requires simplifications and therefore it is recommended that resulting solutions are tested with simulation models not requiring such simplifications (for which solvers with reasonable run times and known numerical precision are not available). In this sense it is a screening tool for identifying solutions. Users of the tool will not be exposed to the complexity of the mathematical programming models but instead will access its capabilities via simpler calls from Jupyter notebooks; Jupyter notebooks are a general-purpose, more user-friendly computational platform with extensive data analysis and visualization capabilities. Notebook-based case studies will be distributed for demonstration of ARC to a range of hazards including tornadoes, hurricanes, inland and coastal flooding, and wildland-urban-interface (WUI) fires. The flexibility and clarity of Jupyter notebooks make it easy to use and extend by consultancy firms, communities with in-house technical capability, the research community, and software developers. 
 

Research Plan
The research team for the project consists of operations researchers (OR) interacting with other NIST EL and CoE social scientists, engineers, and economists depending on the nature of the research efforts for the planning year.

Much of the foundational work for this project has been completed, resulting in beta versions of the model for several hazards. Progress of the project has reflected a gradual development of the decision support tool. In FY17, FY18, and FY19 the focus was on, respectively, verification of the math programming optimization approach (“proof-of-concept”), development of a MATLAB-based software prototype, and extension of the underlying math programs. In FY20, the model was ported to a much more flexible, powerful computational environment (a Jupyter Hub with multi-core solver licenses) for use by NIST researchers, and in FY21 a publicly released beta version of NIST ARC was available for download at https://www.nist.gov/services-resources/software/nist-arc-nist-alternatives-resilient-communities-tool.

NIST ARC has since been continually extended. In FY22, the underlying math programs were extended to include additional parts of the community resilience system (e.g., distributed networks and service areas) and initial steps were taken toward coordination with the AEO-developed EDGe$ software. In FY23, as a riverine flooding case study was used for the original testing, ARC was demonstrated for a new hazard, seismic. In FY24, its application was further extended with adaptation for tornados (ARC-T) and work towards a fire version (ARC-F). Also, in FY24, a freely available version of a key component the tool (an optimization modeling environment, “AMPL”) was successfully tested; combined with an open-source or free-for-noncommercial-use solver (e.g., SCIP, COIN-OR CBC, HIGHS), this represents a free option for communities. In FY25, the model was extended to include a broader system, one that includes consideration of access to health care services.

A summary of the research plan consists of roughly two-year development cycles:

1. Participate in field studies/ review literature on disaster science/ interact with communities. The purpose is to gain information to guide modeling efforts for community resilience in this rapidly evolving and relatively new field. For example, through these activities mitigation options, empirical relationships describing recovery, and the decision-making processes of infrastructure operators are learned.
2. Identify new extension/new functionality of ARC. The extension may address a new hazard, incorporate new science, or incorporate a new system (e.g., schools), or some combination thereof. Also, it may include new functionality, e.g., a new method for uncertainty analysis or connection to other NIST-funded Community Resilience tools.
3. Identify a case study. The extension, and the case study are a coordinated decision as the case study will be used in the testing of the extension. Data availability is also among the considerations.
4. Make changes to the mathematical programming model. This likely will require changes to the base model.
5. Test the new extension. Use the case study to evaluate the correctness of the model and the computational time required to run it.
6. Publish the science behind the new version. Submit the research behind the model development to a peer reviewed journal to gain scientific approval.
7. Make the case study-based Jupyter notebook publicly available. This requires approval in the NIST MIDAS system.
8. Gain feedback from the community on the value of the notebooks.

This research intersects with that of the other projects in and outside of the NIST Community Resilience Thrust Areas, including the Community Assessment Methodology, and the Economic Decision Guide projects. The determination of the math program objectives, which include resilience and cost, will be informed by the metrics and economics research. Further, the solutions found with ARC can be evaluated from a resilience metrics and economic perspective. For this reason, ARC is being developed in close coordination with these projects. In addition, the software will be developed to interact with other tools, including NIST-funded tools such as the EDGe$ online tool EDGe$ online tool, TraCR TraCR, and CoE’s IN-CORE.

In addition, research is conducted that has merit independent of ARC. The main emphasis of this kind of research is the development of algorithms for characterizing and improving the resiliency of distributed infrastructure networks (e.g., transportation, power, water). To date, this has included the development of computationally tractable methods for prioritizing resilience-enhancing investments in large-scale distributed infrastructure systems (“criticality analysis”); in FY23, methods based on machine learning, an artificial intelligence technique, and network embedding, a means of generating lower-dimension, and therefore more tractable, approximate representations of networks, were developed. In FY23, new Monte Carlo based methods for more efficient sampling of distribution network failure were tested. In FY25, analytic work and programs for incentivizing upgrades was conducted to support the modeling of bringing buildings up to current code.