3 Presentations

• 
  01:00 PM - 01:22 PM

  AI-Accelerated Power Systems Optimization with MIPLearn and UnitCommitment.jl

  • Alinson Santos Xavier, presenter, Argonne National Laboratory

  New grid technologies, such as renewables, energy storage and distributed energy resources, have led to an exponential increase in the complexity of the mathematical models that the power industry relies on to maintain the reliability, resilience and affordability of the power grid. Taking into account that many optimization problems in power systems are solved repeatedly with only slight variations in input data, in this talk we investigate the usage of machine learning (ML) methods to accelerate the performance of state-of-the-art mixed-integer programming (MIP) solvers. We introduce two open-source tools for AI-enhanced power systems optimization: MIPLearn, a general-purpose framework for combining MIP and ML; and UnitCommitment.jl, a learning-enhanced optimization package specifically designed for the Security-Constrained Unit Commitment problem.

• 
  01:22 PM - 01:44 PM

  Data-driven coordination for renewable grid integration

  • Robert Mieth, presenter, Rutgers University

  This talk discusses the benefits of data exchange and task-focused data-driven decision making for renewable grid integration. Here, we focus on the interface between the grid operator and energy resource owners and how a data economy can lead to individual and welfare beneficial improvements of operation cost and safety. Grid owners can benefit from purchasing high-quality data and predictions from renewable and distributed resource owners. We will discuss optimal acquisition and data valuation. Resource operators, on the other hand, can improve their operations if provided access to grid data the extends price-only coordination. We will discuss this focusing on maintenance schedules for offshore wind power systems.

• 
  01:44 PM - 02:06 PM

  Differential Privacy for Regulatory Compliance in Cyberattack Detection on Energy Systems

  • Paritosh Ramanan, presenter, Paritosh Ramanan

  With growing risk of cyberattacks, regulatory compliance requirements are increasing for large scale energy networks comprising multiple utility stakeholders. The primary goal of regulators is to ensure overall system stability with recourse to trustworthy stakeholder attack detection. However, adhering to compliance requirements requires stakeholders to also disclose sensor and control data to regulators raising privacy concerns. In this talk, we present a private cyberattack detection framework that utilizes a differentially private (DP) hypothesis testing framework to address the problem of mutual dependency caused by regulation. The hallmark of our approach is a two phase privacy scheme that protects the privacy of covariance matrices, as well as the associated test statistics computed for detecting alarms. Theoretically, we show that our method induces misclassification error rates comparable to non-DP versions. Using real-world datasets, we demonstrate that our method delivers reliable outcomes for a wide variety of attack scenarios with robust privacy guarantees.