Abstract
The continuous introduction of emerging technologies such as Artificial Intelligence (AI) in smart grids has improved the reliability of operations and communication among components. However, such technologies can become unsustainable and further introduce security vulnerabilities. Any potential impact of security attacks on a smart grid component such as a sensor, either through internal or external adversaries, may lead to a breakdown of operations. Additionally, the large number of components along with their dependencies on emerging technologies may hinder the ability of grid operators to sustainably identify security flaws. In this paper, we propose a sustainable dispositional and situational security awareness model through the utilization of a Fuzzy Cognitive Map (FCM) to effectively reveal and mitigate security vulnerabilities in smart grids without overdependence on high-overhead emerging technologies.
Introduction
As smart grids increasingly adopt Artificial Intelligence (AI) and Internet of Things (IoT) components, they become more efficient but also more vulnerable to cyber-attacks. A critical gap exists in current security models: many are "unsustainable," requiring massive computational resources or constant manual intervention. Furthermore, existing solutions often focus solely on situational awareness (external factors like network timing) while ignoring dispositional awareness (internal factors like data packet size). Most models can identify a threat but lack integrated control mechanisms to mitigate it and return the system to a healthy state.
Methodology: The Four-Stage Security Pipeline
The proposed model utilizes Fuzzy Cognitive Maps (FCM), a graph-based modeling technique where nodes represent grid components and edges represent causal relationships. This approach is computationally lightweight, making it suitable for deployment on edge devices like Intelligent Electronic Devices (IEDs).
The security pipeline consists of four integrated stages:
- Security Disposition: Captures internal component attributes (e.g., data size).
- Security Situation: Captures external environmental attributes (e.g., data availability time).
- Security Modeling: Combines dispositional and situational data to identify state drifts indicating potential attacks.
- Security Control: Automatically resets compromised values to a "healthy" state.
To adapt to variations in grid events with minimal lag, the model employs Hebbian learning to fine-tune causal weights between components. Components are classified based on their security state: Healthy (0.25 to 1.0), Unhealthy (0 to 0.25), and Failure (-1 to 0).
Case Study: SCADA System Simulation
The model was evaluated using a simulated Supervisory Control and Data Acquisition (SCADA) system environment in MATLAB, involving a Master Terminal Unit (MTU), Remote Terminal Units (RTUs), and a Human Machine Interface (HMI).
Key Results
- Normal Operations: Under healthy conditions, components maintained security awareness scores around 0.80.
- Integrity Attack Mitigation: A failure was simulated by manually dropping the MTU's state to -1. While this initially caused the HMI and RTUs to drop to an "Unhealthy" state (0), the Security Control function successfully intervened. In the subsequent simulation state, the MTU recovered to 0.8035 and the HMI to 0.4034, demonstrating the model's self-healing capability.
Conclusion and Future Work
This research demonstrates that monitoring "dispositional" attributes allows for the detection of sophisticated data manipulation attacks that situational metrics might miss. By providing automated recovery, the model moves beyond simple intrusion detection toward active resilience. Future work will focus on expanding the set of dispositional and situational attributes and addressing non-security failures, such as physical equipment wear and tear.
Summary updated