Skip to content
Enrique Tomás Martínez Beltrán
HomeResearchPublicationsTeachingBlog
ENES
Contact
HomeResearchPublicationsTeachingBlog
ENES
Contact

Enrique Tomás Martínez Beltrán

Federated learning, trustworthy AI and cyberdefense research, focused on systems that are robust, privacy-preserving and useful in security operations.

  • Privacy Policy
  • Terms of Service
  • Accessibility Statement
  • GitHubopens in a new tab
  • LinkedInopens in a new tab
  • Google Scholaropens in a new tab
  • RGopens in a new tab
  • ORCIDopens in a new tab
  • Scopusopens in a new tab
  • DBLPopens in a new tab
  • Web of Scienceopens in a new tab

Enrique Tomás Martínez Beltrán. All rights reserved.

Back to top

This site loads optional analytics from Google and external analytics providers only if you accept. You can decline and continue using the site normally.

  1. Home
  2. Research Notes on Federated Learning, Cybersecurity and Cyberdefense
  3. Federated Energy Anomaly Detection for Critical Infrastructure
Industrial CybersecurityCritical InfrastructureDecentralized Federated LearningTrustworthy AI

Federated Energy Anomaly Detection for Critical Infrastructure

Privacy-aware cyberattack detection for smart-grid time series

Enrique Tomás Martínez Beltrán

Ph.D. Researcher in Federated Learning and Cybersecurity

May 25, 20266 min read
  • LinkedInopens in a new tab
  • Xopens in a new tab
Federated Energy Anomaly Detection for Critical Infrastructure

Critical infrastructure operators need anomaly detection, but they also need privacy and operational control. In energy systems, raw telemetry may expose substations, load profiles, operational routines and sensitive fault patterns.

Federated anomaly detection is a way to collaborate without pooling everything.

Why energy systems are different

Smart-grid telemetry is temporal, heterogeneous and context-dependent. The same signal can be normal in one operational state and suspicious in another.

That means detection models need context:

  • threat level,
  • asset criticality,
  • seasonality,
  • model drift,
  • local operating constraints.

Decentralized learning pattern

Each monitoring node can train on local time-series data and share model updates or distilled signals with peers. The goal is to detect cyberattacks and anomalies while preserving control over sensitive industrial data.

This is not just a privacy feature. It is also an availability and resilience feature: the system should keep learning even when a central service is unavailable.

Trustworthy alerts

The alert should explain why a signal matters and whether the model is operating in a familiar regime. If drift is high, the alert should say so. If peers disagree, that disagreement is part of the evidence.

Trustworthy anomaly detection is not only about raising alerts. It is about making alerts useful enough to support mitigation.

For critical infrastructure, alert evidence must remain constrained and auditable: affected asset, anomaly window, peer agreement, drift level, safety constraints and approved operational context.

Key takeaway

In critical infrastructure, federated anomaly detection should prioritize useful, bounded evidence over broad data sharing. The alert must be explainable, auditable and aligned with operational constraints so operators can decide whether to investigate, mitigate or keep observing.

Open research question

How can federated energy systems distinguish true cyber anomalies from local operational changes when raw telemetry cannot be pooled centrally?

Related Research

Byzantine-Resilient Aggregation for Decentralized Federated Learning

May 27, 2026

Byzantine-Resilient Aggregation for Decentralized Federated Learning

A focused research note on median, trimmed mean and trust-weighted aggregation for peer-to-peer federations under poisoning and unreliable clients.

From Monitoring to Mitigation: A DFL Cyberdefense Lifecycle with LLM Explanations

May 30, 2026

From Monitoring to Mitigation: A DFL Cyberdefense Lifecycle with LLM Explanations

A practical note on how distributed monitoring, DFL models, alert evidence and LLM-based support can fit into a cyberdefense workflow.