Energy scheduling for DoS attack over multi-hop networks: Deep reinforcement learning approach

Area of Science:

• Cybersecurity
• Network Security
• Control Systems

Background:

• Remote state estimation in multi-hop networks is vulnerable to Denial-of-Service (DoS) attacks.
• Limited sensor communication range necessitates relay nodes, creating complex network topologies.
• DoS attacks aim to degrade estimation accuracy by manipulating energy levels on communication channels.

Purpose of the Study:

• To develop an optimal energy scheduling strategy for DoS attacks in multi-hop networks.
• To maximize the estimation error covariance under energy constraints.
• To reduce the computational complexity of finding the optimal attack policy.

Main Methods:

• Formulation of the energy scheduling problem as a Markov decision process (MDP).
• Proof of the existence of an optimal deterministic and stationary policy (DSP).
• Derivation of a simplified threshold structure for the optimal policy.
• Application of the dueling double Q-network (D3QN) deep reinforcement learning algorithm to approximate the optimal policy.

Main Results:

• The existence of an optimal deterministic and stationary policy for the DoS attacker is proven.
• A threshold structure significantly simplifies the optimal policy and reduces computational complexity.
• The D3QN algorithm effectively approximates the optimal energy scheduling strategy.
• Simulations validate the effectiveness of the proposed approach.

Conclusions:

• The study provides a robust framework for understanding and optimizing DoS attack energy scheduling in multi-hop networks.
• The combination of MDP and D3QN offers an efficient solution for complex network security scenarios.
• The findings contribute to the development of more resilient remote state estimation systems.