INFORMS 2021 Program Book

INFORMS Anaheim 2021

TB01

2 - A Continuous-time Linear Programming Formulation for Resource-constrained Project Scheduling with Multiple Sites Norbert Trautmann, Professor, University of Bern, Bern, Switzerland, Mario Gnaegi We present a continuous-time mixed-integer linear programming formulation for scheduling the activities of a multi-site project subject to precedence and renewable-resource constraints. As a consequence of the distribution of the renewable-resource units among the multiple sites, transportation times must be considered for moving some mobile resource units or the output of some precedence-related activities. 3 - A Revised PERT Model Using Log-Normal Activity Times Eric Logan Huggins, Professor of Management, Fort Lewis College, Durango, CO, United States, Ivan G. Guardiola The standard PERT model assumes that activity times follow Beta distributions and that completion time for the entire project is Normally distributed. We consider several revisions to this model; specifically, we assume that the activity times follow Log-Normal distributions instead which we believe may improve the model. While the Log-Normal can be similarly skewed like the Beta, it has an infinite right tail which puts no limit on how long an activity can be delayed; further, our model allows estimating the worst case scenario with a given confidence level. 4 - Linear Relaxations for Mixed Integer Nonlinear Programs in Natural Gas Transportation Networks Sai Krishna Kanth Hari, Los Alamos National Laboratory, Los Alamos, NM, United States Efficient and profitable transportation of natural gas along pipeline networks requires solving challenging Mixed Integer Nonlinear Programs (MINLPs) as the gas flow is governed by nonlinear, non-convex physics. Obtaining tight bounds on the objective value of these MINLPs using convex relaxations is of significant interest in the research community. Here, we utilize the recent advancements in the literature of polyhedral relaxations for univariate and bilinear functions to develop Linear Programming and Mixed Integer Linear Programming relaxations for the MINLP. 5 - How To Quantify Outcome Functions of Linear Programs with Interval Right-hand Sides Mohsen Mohammadi, University of Louisville, Louisville, KY, United States, Monica Gentili, Milan Hladik, Raffaele Cerulli An outcome function is an extra function of interest associated with the set of optimal solutions of an optimization problem. In this talk, we consider the problem of finding the range of a given outcome function over the set of all possible optimal solutions of a linear program with interval right-hand sides. We show the relevance of the problem in practice, address its computational complexity, and discuss some of its theoretical properties. Moreover, we propose several heuristics to solve the problem and analyze their quality and efficiency.” TB01 CC Ballroom A / Virtual Theater 1 Hybrid Dynamic Data Driven Application Systems Sponsored: Simulation Society Sponsored Session Chair: Kevin Jin, 1 - Social Media Networks as Dynamic Data Driven Applications Systems Conrad Tucker, Carnegie Mellon University, Pittsburgh, PA, United States, Sakthi Prakash This research aims to transform large-scale social media networks (SMNs) into dynamic data-driven application systems (DDDAS) that model and predict real world events/national security threats. The conventional wisdom has been that complex, high fidelity sensors are needed to accurately model real-world events. The concept of large-scale SMNs serving as dynamic sensing systems is a departure from traditional perceptions of SMNs simply being platforms for disseminating content and connecting individuals. However, the advent of AI that are capable of generating hyper-realistic data, threatens to significantly degrade the veracity of SMN data and hence, their ability to serve as DDDAS. This work proposes algorithms that enable SMNs to serve as DDDAS, as well as algorithms that protect SMNs against adversarial attacks seeking to degrade their veracity. Tuesday, 7:45AM-9:15AM

2 - Identification and Adaptive Control of Markov Jump Systems: Sample Complexity and Regret Bounds Laura Balzano, University of Michigan, Ann Arbor, MI, 48109, United States Learning how to effectively control unknown dynamical systems is crucial for intelligent autonomous systems. This task becomes a significant challenge when the underlying dynamics are changing with time. Motivated by this challenge, this work considers the problem of controlling an unknown Markov jump linear system (MJS) to optimize a quadratic objective. By taking a model-based perspective, we consider identification-based adaptive control for MJSs. We first provide a system identification algorithm for MJS to learn the dynamics in each mode as well as the Markov transition matrix, underlying the evolution of the mode switches, from a single trajectory of the system states, inputs, and modes. Through mixing-time arguments, sample complexity of this algorithm is shown to be O(1/sqrt(T). We then propose an adaptive control scheme that performs system identification together with certainty equivalent control to adapt the controllers in an episodic fashion. Combining our sample complexity results with recent perturbation results for certainty equivalent control, we prove that when the episode lengths are appropriately chosen, the proposed adaptive control scheme achieves O(sqrt(T)) regret which can be improved to O(log(T)) with partial knowledge of the system. This is work with Yahya Sattar, Zhe Du, Davoud Ataee Tarzanagh, Samet Oymak, and Necmiye Ozay. 3 - Using Saliency Map to Interpret EO and Passive RF Sensor Fusion Asad Vakil, Oakland University, Rochester, MI, United States, Robert Ewing, Erik Blasch, Jia Li This study applies saliency map to find the impact of individual modalities in EO and P-RF sensor fusion using the AFRL’s ESCAPE dataset. When saliency maps are applied to P-RF data alone, the resulting maps are mostly blank due to insignificant changes. To prevent the AI models from focusing solely on the EO input, where dense optical flow (DOF) makes the presence of vehicle targets obvious, the P-RF histogram was overlayed on top of DOF-EO to generate a single input. The experimental results reveal three broad categories of heatmaps, heatmaps focused on vehicles when the target is visible, heatmaps focused on the P-RF histogram when the target is out of sight, and heatmaps focused on both which occur when multiple targets are moving. While the insights provided by heatmaps are more intuitive than quantitative, they highlight and confirm the impact of P-RF data in various scenes. 4 - Physics-informed Statistical Learning of Scientific and Engineering Processes Xiao Liu, University of Arkansas, Fayetteville, AR, 72701, United States, Xinchao Liu, Guanzhou Wei, Mohammadmahdi Hajiha This talk will focus on a physics-informed statistical learning framework for engineering processes governed by fundamental physical laws. Two examples will be discussed. The first example is related to the modeling and prediction of natural processes (smoke, sea surface temperature, aerosol depth, etc.), and the second example deals with how statistical learning can be used to learn the outputs from computer simulation models (such as Finite Element Analysis), and make predictions under new scenarios. 5 - Thermospheric Density Modeling Using Dynamicmode Decomposition with Control and Machine Learning Richard Linares, MIT Thermospheric density is the main source of uncertainties for satellites’ orbit prediction in Low Earth Orbit. It is highly dynamic and driven by external drivers, such as solar and geomagnetic activity. A dynamic reduced order thermospheric density model is developed for real-time density prediction using density data from Global Ionosphere Thermosphere Model (GITM). A deep convolution autoencoder neural network is constructed to capture a high-level abstraction of the thermospheric density in a reduced order space. A linear model is then fitted onto the reduced-order states using Dynamical Mode Decomposition with Control. Data assimilation is then used to improve prediction capabilities. 6 - DDDAS for Robust Reconfiguration of Self-healing Microgrid Clusters Cluster of microgrids are formed of loads, distributed generation units (DGs) and energy storage units that operate in coordination to supply electricity in a reliable and low-cost manner while supporting each other against the effects of prevalent faults, with or without support from the host grid. Here, we propose a reconfiguration method for such microgrid clusters using the Dynamic Data Driven Applications Systems (DDDAS) framework to manage the stochasticity of the non-dispatchable DGs and loads under different system anomalies, such as extreme weather events. Self-healing capabilities are facilitated via intelligent exploitation of real-time data and, in return, adjustment of the measurement process to guide new data collection. Abdurrahman Yavuz, University of Miami, Miami, FL, United States, Nurcin Celik, Jie Xu, Chun-Hung Chen

99