Skip to content

Dynamic defense strategies for cyber-physical systems using Stackelberg games and deep reinforcement learning in discrete and continuous time

Sep 2026 · International Journal of Information Security · Vol 25 · 0 citations · 48 references
Smart Grid Security and Resilience

TL;DR

This study presents a dual-domain defense framework that combines Stackelberg game theory with Deep Reinforcement Learning that offers a robust, adaptable toolset for CPS defense by clarifying the trade-offs between temporal abstractions in nonlinear and dynamic environments.

Abstract

As cyber threats to power grid infrastructures escalate, the urgency of understanding how to protect cyber-physical systems (CPS) has never been greater. These systems, which integrate physical processes with digital control, are increasingly susceptible to sophisticated cyberattacks that can lead to widespread disruption. While most existing defense models function within either discrete or continuous-time frameworks, this research addresses an important limitation in the literature: the limited comparative treatment of both temporal domains within a common strategic framework. This study presents a dual-domain defense framework that combines Stackelberg game theory with Deep Reinforcement Learning (DRL). Rather than merging discrete-time and continuous-time dynamics into a single hybrid model, the two temporal formulations are evaluated as alternative representations of the same CPS security interaction under identical attacker–defender scenarios and metrics, enabling a systematic comparison of strategic decision-making and physical system behavior. The objective of this framework is to facilitate proactive defense decisions that can anticipate and respond to attacks with strategic precision. We conducted extensive simulations using Python to assess the proposed model in both discrete and continuous-time scenarios. Our approach was evaluated through extensive simulations under realistic adversarial conditions to confirm its resilience and cost-effectiveness. Key findings indicate that defender-first strategies in discrete time effectively minimize system damage and alleviate computational burdens, while continuous-time responses, although immediate, demand significantly higher resource investment. This dual-domain solution offers a robust, adaptable toolset for CPS defense by clarifying the trade-offs between temporal abstractions in nonlinear and dynamic environments.

Read PDF

Similar papers

Conference Aug 2026

Bridging the Simulation-to-Emulation Gap in Autonomous Cyber Defence: A Hybrid PPO and LLM-Based Framework

Autonomous cyber defense has become an important research area as modern networks increase in complexity, size, and vulnerability to rapid multi-stage attacks. Although reinforcement learning shows potential for making defensive decisions, many current studies focus only on simulations. This leaves questions about how...

L. K. Suresh Kumar, Mohammed Yaseen, Ravi Uyyala et al. · 0 citations
Review Open access Sep 2026

A Survey on Defensive Resource Allocation for Cyber-Physical Power SystemsA Survey on Defensive Resource Allocation for Cyber-Physical Power Systems

Cyber-Physical Power Systems (CPPSs) are one of the highly visible, sophisticated, and beneficial trends of the modern energy infrastructure. CPPSs connect Information and Communication Technologies (ICTs) to physical power assets to enhance operational capabilities by leveraging bidirectional data flow. When CPPSs are...

A. S. Shatam, Ahmed A. Alabdel Abass · 0 citations
#artificial intelligence Preprint Sep 2026

CyberWorld: World Models for Sample-Efficient Autonomous Cyber Defense

This work introduces CyberWorld, a Dreamer-style world modeling framework that learns latent cyber dynamics from vector, graph, textual, and multimodal representations of the defended network, and identifies world representation as a central design axis for robustness and scalability.

Ryozo Masukawa, Sanggeon Yun, Raheeb Hassan et al. · 0 citations

STOCHASTIC ATTACK–DEFENSE CONFRONTATION MODEL FOR WEB

The model uses an aggregated representation of system components as an aggregation of an attack graph built with the Meta Attack Language together with a stochastic initialization of initial conditions, which reduces the dimensionality of the state space and ensures the computational tractability of the analysis.

Prytula Andrii, Kupershtein Leonid · 0 citations

Related blog posts

MIT News · Artificial Intelligence Nov 17, 2026

Learning the language of machine learning

By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.

Microsoft Research Blog Nov 16, 2026

TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient

AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.