AI and Machine Learning for Energy
For utility CIOs, market participants, and policymakers managing the AI buildout. How AI and ML reshape both sides of the grid - improving forecasting, anomaly detection, and asset health on the supply side, while data-center loads add unprecedented demand on the consumption side. Includes my U.S. patent in ML data preprocessing, published work on anomaly detection in California prices, and physics-informed neural networks for maritime energy.
What this frontier is about
AI and machine learning are no longer experimental in the electricity sector - they're operational. Forecasting, anomaly detection, capacity planning, and resource adequacy assessment all increasingly depend on ML pipelines, while the rise of AI workloads themselves is the single largest new source of grid load growth in decades. The work in this frontier sits at both ends of that loop: building AI methods that make the grid more reliable, and helping the grid absorb the data centers training and serving those AI models.
This is also the frontier where my U.S. patent in machine-learning data preprocessing lives - the technical foundation under most of the applied research I do across the other three frontiers.
How I work in this area
Physics-informed AI for power systems. Pure data-driven models often violate physical laws — voltages, power flows, and machine dynamics that any operator knows can't actually happen. Physics-informed approaches embed those constraints directly into the model, producing forecasts and simulations that are both more accurate and more trustworthy. My recent work applies this to electrical machine modeling and maritime energy systems.
Anomaly detection and forecasting in wholesale electricity markets. Wholesale electricity price signals carry information about congestion, scarcity, renewable curtailment, and operator behavior - but only if you can separate signal from noise. My work on the California ISO market combines anomaly-detection techniques with price-forecasting models to give traders, planners, and regulators sharper visibility into what the market is actually doing.
Machine-learning data preprocessing at scale. The accuracy of any ML system depends more on data preparation than on model choice. My U.S. patent covers methods for automated outlier detection and feature transformation — techniques developed against messy utility datasets where missing values, sensor errors, and distributional shifts are the norm rather than the exception.
AI as load: the data center frontier. Hyperscale AI training and inference are reshaping load forecasting at every ISO in North America. Single sites that draw multiple gigawatts and operate at unprecedented thermal-cycling profiles change interconnection timelines, transmission planning, capacity adequacy, and tariff design. This is increasingly the operational core of my work at Energy Exemplar — helping utilities, regulators, and hyperscalers turn questions like "where can we site the next gigawatt of demand?" into specific, defensible answers.
Stranded-asset analytics using ML and AI. Building on my SAS-era research, I use ML and AI to identify, quantify, and price the risk of stranded electricity generation and storage assets in a decarbonizing economy.
My selected publications and thought leadership in this frontier
Patent and books
U.S. Patent No. 12,254,001 — Automated Outlier Detection and Feature Transformation in Machine Learning Models — 2025
Tackling the Risk of Stranded Electricity Assets with Machine Learning and Artificial Intelligence — IntechOpen, 2021
Introductory Chapter: Sustainable Energy Investment and the Transition to Renewable Energy-Powered Futures — IntechOpen, 2021
Recent peer-reviewed research
Advancing Electrical Machine Modeling with Physics-Informed AI — 2026
Physics-Informed AI for Maritime Energy and the Blue Economy — 2025
Anomaly Detection in California Electricity Price Forecasting — 2025
Recognition
For the full archive, see Publications →
What this means for utilities, System Operators, and large-load customers
If you're a utility analytics team, the question isn't whether to adopt ML — it's how to do it without inheriting brittle pipelines, hallucinating forecasts, or violations of physical constraints that put operators at risk. If you're an ISO planner, the question is how AI-driven decision intelligence reshapes resource adequacy and capacity expansion under deep uncertainty. And if you're a large-load customer — particularly a hyperscaler siting new compute — the question is how the grid you're plugging into actually works, how fast it can scale, and what you can do to make interconnection faster.
