Home
Services
Case Studies
Contact us
Product | Cascade intelligence
VolteGrid™
Real-time cascade intelligence for DER dominated networks

VolteGrid™ predicts, quantifies and helps prevent cascading failures in DER dominated networks in real time. It fuses live telemetry with physics-informed stochastic models and a graph neural network cascade surrogate, resolved into a continuously updated reliability index.

Live operations cockpit below
Heterodyne | Grid Intelligence
VolteGrid™

Real-time cascade intelligence for DER dominated distribution networks.

VolteGrid™ is an AI-native grid intelligence platform that predicts, quantifies and helps prevent cascading failures in DER dominated power systems, in real time. It fuses live network telemetry with physics-informed stochastic modelling and a graph neural network cascade surrogate, delivering a continuously updated index for every operational decision.

Live · Operations cockpit

One view. Every feeder. Real time.

Operations CockpitSevernside / Avonmouth feeder group
Network status0.793â–˛
LIVE
06:47:35NETWORK TIME
Network overview · Geographic feeder map Status Low → High
Priority watchlist

A live view of the VolteGrid™ operator console. Network status shifts as local disturbances develop and clear across the feeder group.

Network status score
0.793
Elevated risk
60-frame trend
Monitoring signals
Asset health0.79
Capacity margin0.66
Network activity0.92
Data coverage0.93
Stability outlook0.77
Activity trend1.87 index
Watch nodes
Max coverage
A single view spanning time, location, asset class and failure mode
ms scale
Inverter-speed dynamics captured at the distribution edge
10³×
Cascade scoring vs direct physical simulation, via GNN surrogate
GB data
Validated on UKPN, National Grid and SP Energy Networks open data
The problem

Cascades that scalar metrics cannot see.

Distribution networks were engineered for one-way power flow, predictable load and abundant rotational inertia. That grid is gone. Rooftop solar, batteries, EVs and heat pumps have turned the LV and MV network into a fast, bidirectional, low-inertia system, where disturbances propagate in ways legacy protection and planning tools were never designed to see.

Transient instability at the edge.

Inverter-based resources respond in milliseconds and mass-disconnect on frequency and voltage excursions.

Contagion through coupling.

A feeder trip, a protection maloperation or a correlated DER disconnection propagates through electrical, protection and control couplings into system-scale loss.

Blind averages.

SAIDI and SAIFI describe the past. They say nothing about the probability, geometry or severity of the next failure sequence.

Scalar reliability metricsNo temporal resolution
SAIDI42.3min per customer-year · constant over reporting year
SAIFI0.82interruptions per year · constant over reporting year
CAIDI51.6min per interruption · constant over reporting year
ENS118MWh per year · constant over reporting year
LOLE2.4h per year · constant over reporting year
ANNUAL SCALARS · BACKWARD LOOKING · SINGLE NUMBERS
What VolteGrid™ does

From telemetry to foresight in one operational layer.

VolteGrid™ ingests the network as it actually is: topology, live measurements, DER behaviour, protection configuration. It answers three questions continuously.

Q1 · Initiation

Where could a cascade start?

Stochastic point-process models, running in real time, learn the self-exciting structure of network events: how one disturbance raises the probability of the next, where, and on what timescale. This captures the clustered, bursty character of real failure data that classical independent-failure models miss.

Q2 · Propagation

How would it propagate?

A graph neural network cascade surrogate, trained against high-fidelity physical simulation, evaluates propagation across the network graph orders of magnitude faster than direct simulation: fast enough to score thousands of contingency and disturbance scenarios inside the operational decision window.

Q3 · Consequence

What does it mean right now?

The outputs resolve into a single multi-dimensional reliability index: a live measure of network health that decomposes across time, location, asset class and failure mode. Designed to sit alongside, and ultimately supersede, the scalar averages the industry has relied on for forty years.

Architecture

Physics first. AI where it earns its place.

Sense01
Fuses SCADA, smart-meter, PMU-class and DER telemetry with network topology into a coherent live state, including principled treatment of the unmonitored LV network.
Model02
Physics-informed stochastic point-process engine estimating event intensity and inter-event coupling across the network, updated in real time.
Accelerate03
GNN cascade surrogate delivering simulation-grade propagation estimates at operational speed, with uncertainty quantification on every output.
Decide04
Multi-dimensional reliability index, ranked intervention options and explainable evidence trails, exposed via console, API and operator integrations.

Every AI component is anchored to network physics and validated against full physical simulation. VolteGrid™ is not a black box scoring engine. It is a physically grounded inference system with machine learning used precisely where it buys speed and coverage.

Validation

Built on real networks. Tested on real data.

Models tuned on synthetic benchmarks routinely fail on the messiness of live networks: missing telemetry, asymmetric feeders, correlated weather-driven events. VolteGrid™ was built the other way round: from real network data outward, using openly published operational datasets from Great Britain's network operators.

UK Power Networks
Open data

Real distribution topologies, event records and DER penetration profiles across London, the South East and East of England.

National Grid
Open data

System-level operational datasets grounding the transmission-distribution interface and frequency-event context.

SP Energy Networks
Open data

Network and outage datasets extending validation coverage across distinct geography, weather exposure and network construction.

Provenance

Oxford engineering science research - productised.

VolteGrid™ originates from research in Engineering Science at the University of Oxford on cascading failure dynamics and reliability engineering for DER dominated distribution grids. That research lineage runs through the product: the stochastic modelling framework and the cascade surrogate methodology are the direct descendants of peer-oriented academic work, hardened into an operational platform.

Born in the Engineering Science laboratories of the University of Oxford.
Cascading failure dynamics · Reliability engineering · DER dominated grids

Heterodyne is a member of the NVIDIA Inception Program
Deploy

The grid is changing faster than the tools that run it.

VolteGrid™ is working with forward-leaning network operators to deploy live cascade intelligence on real networks.