Research

My research interests lie in exploring solutions at the intersection of power systems, optimization, machine learning, and more recently, quantum computing, to address the challenges of modern energy networks, particularly under uncertainty.

Learning-Integrated Power System Operations: The group’s research will focus on developing physics-inspired methodologies that leverage machine learning (ML) to optimize and secure power system operations and planning. We aim to enhance the reliability, efficiency, and resilience of energy networks operating under uncertainty by developing problem-specific ML tools.

Probabilistic and Non-Parametric Methods: Group will focus on advancing probabilistic methods and non-parametric techniques to create robust and interpretable models for Medium to Long Term Planning & Cause-Effect Studies in renewable energy-dominated Indian power grids. By addressing the complexities introduced by renewable integration, we aim to enhance grid reliability and operational efficiency.

Quantum Computing Potential Exploration: Group will investigate the capabilities of quantum computing in solving computationally combursome problems in power grids & other engineering domains. The aim will be to scrutinizing and advance the boundaries of what’s possible using Quantum computing for Our Favourite Problems.

Below is a brief non-rigorous summary and the associated papers related to above mentioned three main research themes.

#1 Closed-form Power Flow & Applications

Closed-form power flow (CFPF) framework aim to provide explicit mathematical expressions for solving AC power flow equations: \(V = f(\mathbf{s})\) (Node voltage as a function of injection vector). The core idea is to bypas iterative numerical techniques like Newton-Raphson and replace these with function evaluation. Unlike linearizations, CFPF is a general framework which allows to tailor complexity of \(f(\mathbf{s})\) based on target application. Desired features of CFPF are:

  • Flexible Forms \(\implies\) Non-linear forms with complexity-accuracy trade-off
  • Easy to Evaluate Forms \(\implies\) Faster numerical calculations
  • Non-parametric Forms \(\implies\) Works within a power injection range or hypercube
  • Differentiability of Forms \(\implies\) Can be fed into optimization problems
  • Interpretability of Forms \(\implies\) Should provide insights into the physical system

The CFPF framework is build using Gaussian Process (GPs) to achieve these desired features. Beyond direct applications, Vertex-Degree-Kernel (VDK) design has been proposed for scaling the GPs to larger grid. Some key innovations involve designing a Network-Swipe Active Learning algorithm, Multi-Task VDK GP and Theoretical Learning Bounds for Risk Analysis.

Related Papers: 1) Basic CFPF Frameworl 2) Vertex-Degree Kernel and Network-Swipe Algorithm for Fast Risk Assessment 3) Multi-Task VDK GP for Network Contingencies 4) Critical Prosumer Identification 5) Privacy-preserving Feasibility Assessment

#2 Optimal Power Flow (OPF) Proxies

Details will be updated soon!

#3 Potential Quantum Advantage Analysis in Power Systems

Details will be updated soon!