Structured Differentiable Optimization for Efficient Decision-focused Learning in Power Systems
2026arXiv preprint arXiv:2608.04189

Research Associate · Imperial College London
I develop learning and optimisation methods for efficient, robust, and secure power-system operation under high penetrations of renewable energy.
Learning-augmented optimisation Decision-focused learning Power-system security System dynamics
Recent work spanning learning-augmented decision-making including efficient differentiable optimization for power system decision-focused learning and constraint learning for stability-constrained power system optimizations.
Structured Differentiable Optimization for Efficient Decision-focused Learning in Power Systems
2026Wangkun Xu and Fei Teng
arXiv preprint arXiv:2608.04189
Wangkun Xu, Hongyang Jia, Yi Wang, Ning Zhang, and Fei Teng
IEEE Transactions on Power Systems
Le Fang, Wangkun Xu*, and Fei Teng
* Corresponding author
arXiv preprint arXiv:2603.20472
Wangkun Xu, Zhongda Chu, and Fei Teng
IEEE Transactions on Power Systems
Delivered the guest lecture An Open-Source Toolbox for Embedding Neural Stability Surrogates in Stability-Constrained Optimization at the University of Strathclyde, Glasgow. Course details.
Open-sourced NCET, a toolbox for embedding neural stability surrogates in stability-constrained optimization. NCET package.
Presented Input Convex Neural Network as a Surrogate in Stability-Constrained Optimization for IBR-dominated Power Systems.
Presented Structured Differentiable Optimization for Efficient Decision-Focused Learning in Power Systems at the EPICS Bimonthly Meeting, covering theory, software, and IEEE 118-bus evidence. Slides.
Reusable scientific packages, research implementations, and open teaching resources.
NCET (Neural-network Constraint Embedding Toolkit) converts supported PyTorch neural networks into exact mixed-integer linear constraints compiled by CVXPY.
Neural stability surrogates · Stability-constrained optimization
Differentiable PyTorch layers for affine-parametric quadratic and linear programs expressed directly in CVXPY.
Differentiable optimisation · Decision-focused learning
Build optimisation-ready power-system cases with profile dataset from PYPOWER or MATPOWER grids, YAML configurations, and bus-level time-series data.
Power-system testbeds · Data pipelines · CVXPY
Hands-on examples covering SVMs, PCA, k-means, and zero-shot energy-price forecasting with Amazon Chronos.
Python tutorial · MSc teaching
ELEC97077 · Smart Grid Technology
Machine Learning and Data Science for Power Systems
MSc Future Power Networks · Department of Electrical and Electronic Engineering, Imperial College London
Guest Lecturer · Lecturing, coursework, demonstrations, tutorials, and marking · 5 hours of lectures + 1-hour tutorial
MECH70039 · Data Science and Digitalisation
Introduction to Machine Learning for Energy Systems
MSc Sustainable Energy Futures · Energy Futures Lab, Imperial College London
Guest Lecturer · Lecture and hands-on tutorial · 2-hour lecture + 2-hour tutorial