Orchestrating Autonomic Software-Defined Networks for Resilience in Critical Energy Infrastructure

Abstract

Critical energy infrastructure is becoming increasingly digitalized as the world transitions toward cleaner and more sustainable energy systems. Offshore wind power plants and other distributed energy systems rely on industrial communication networks to connect turbines, substations, control centres, and edge computing platforms. These networks support continuous monitoring, control, and coordination, making them essential for reliable system operation. However, their growing complexity and connectivity also expose them to equipment failures, changing operating conditions, and increasingly sophisticated cyber threats. Existing resilience approaches mainly depend on static redundancy and predefined recovery mechanisms, which are often unable to respond effectively to dynamic and unpredictable disruptions. This creates a need for industrial networks that can continuously monitor their environment, adapt to changing conditions, and recover while maintaining critical services. This thesis addresses this challenge by proposing that resilience should be treated as a continuous runtime capability rather than a fixed design-time property. Inspired by the adaptive behaviour of the human immune system, the research introduces the **Orchestrated Multi-Agent Resilience Framework (OMARF)**, a biomimetic and autonomic framework for software-defined industrial networks. The framework enables distributed autonomous agents to monitor network conditions, detect abnormal behaviour, make coordinated decisions, and adapt network operations without requiring constant human intervention. By embedding intelligence within the network's knowledge plane, OMARF allows resilience to emerge through continuous observation, learning, and coordinated action. To realise this vision, the thesis designs, develops, and integrates three complementary autonomic agents that address key dimensions of network resilience. The first is a component recovery self-healing agent that detects and recovers from failures affecting both the control plane and the data plane. The second is a threshold-triggered Deep Q-Network self-healing agent that maintains network performance by dynamically adapting routing and resource allocation under congestion and thermally induced performance degradation. The third is an event-driven self-defense agent based on OpenFlow Random Host Mutation that implements a moving target defence strategy to reduce the effectiveness of multi-stage cyber attacks while preserving normal network operation. These agents are deployed within a microservices-based architecture and orchestrated to operate collaboratively as a unified resilience system. The proposed framework is evaluated using a reproducible software-defined networking testbed that combines virtualized industrial network environments, distributed SDN controllers, and containerized agent deployments. In addition, stochastic Petri nets, continuous-time Markov models, and reinforcement learning techniques are employed to analyse resilience behaviour, quantify system performance, and evaluate adaptation under uncertainty. The experimental results demonstrate that coordinated autonomic agents significantly improve resilience by reducing recovery time, maintaining service continuity, stabilizing network performance, and mitigating cyber threats under diverse operating conditions. This research advances the ongoing discourse on resilience by presenting a new perspective in which resilience emerges from orchestrated autonomous behavior rather than relying solely on redundancy or predefined recovery policies. The findings contribute to the development of next-generation critical energy infrastructure that is better equipped to operate reliably, securely, and sustainably in the presence of uncertainty.