Machine Learning for the Low-Voltage Grid
From Smart Meter Signal to Network Decisions
A practical book for energy analytics and ML practitioners
Machine Learning for the Low-Voltage Grid
From raw meter signal to business decision, through the ML foundations, disaggregation, and grid-edge value that make the closing forecast worth trusting, built to be run, not just read.
Meter Signal
Data & ML
Business Value
The arc every chapter follows
The book, in eight parts
Signal Foundations
ML Foundations
Your learning path
Reading a meter signal, general event and change detection, and feature-engineering concepts every later part reuses.
- 01Reading a meter signalSoon
Chapters land after Part 8 ships.
The mathematical and statistical theory behind every model this book uses as a tool: learning theory, optimization, tree ensembles, deep architectures, and the uncertainty-quantification foundations Part 8’s own forecasts depend on.
- 01What is machine learning, preciselySoon
Chapters land after Part 8 ships.
Non-intrusive load monitoring: from the aggregate signal to appliance-level insight.
- 01Non-Intrusive Load Monitoring (in preparation)
- 02Appliance Recognition (in preparation)
- 03Multi-Appliance Recognition (in preparation)
- 04Multi-Task Deep Learning (in preparation)
Modeling the low-voltage network, simulating DER time series on it, and identifying phase from smart-meter data alone, all with open-source power-flow simulation the rest of the book reuses.
- 01Modeling the Low-Voltage Network (in preparation)
- 02Time-Series Simulation and DER Modeling (in preparation)
- 03Phase Identification from Smart-Meter Data (in preparation)
Foundations of load-profile clustering, a pipeline validated on three utilities, the reliability of customer archetypes, and what clustering contributes to network risk under rooftop PV and EV charging.
- 01Foundations of Load-Profile Clustering
- 02Validating a Clustering Pipeline
- 03Customer Archetypes and Their Reliability
- 04Clustering for Network Risk
- 05Segmentation for Network Studies
- 06Representations that Retain Magnitude (in preparation)
- 07Feeder-Level Clustering (in preparation)
Retrieval, ranking, and case-based recommendation for LV management under DER, checked against real, OpenDSS-simulated per-customer outcomes.
- 01Ranking and Recommendation under DER (in preparation)
Voltage-violation and power-quality anomaly detection combining smart-meter data and LV topology, beyond what a single meter can see alone.
- 01Anomaly Detection Beyond the Meter (in preparation)
Household-scale load, PV, and EV forecastability and uncertainty, built on Twiga. The book’s closing part, feeding its own forecasts back into Part 4’s LV network model.
- 01Household Smart-Meter Forecastability (in preparation)
- 02Scaling to Hundreds of Meters (in preparation)
- 03What a Probabilistic Forecast Is Actually Worth (in preparation)
- 04Do Foundation Models Fail Like Everything Else Does? (in preparation)
- 05What the Grid Actually Faces When These Forecasts Come True (in preparation)
- 06Can Smart-Meter Readings Fill In What the Grid Cannot See (in preparation)
5 chapters shipped so far, each with a runnable notebook. Follow progress on GitHub.