Event-Triggered GAT-LSTM Framework for Attack Detection in Heating, Ventilation, and Air Conditioning Systems

TL;DR

HVAC systems with distributed sensors are vulnerable to cyber-physical attacks and raise privacy concerns from cloud data sharing. The authors propose a privacy-preserving two-tier framework: a local Event-Triggered Unit (ETU) performs binary anomaly detection and encrypts data with Fully Homomorphic Encryption (FHE) before sending to a cloud-based Attack Detection Unit (ADU) that uses a Graph Attention Network (GAT) to model spatial relations and an LSTM to capture temporal dynamics for attack-type classification. The approach achieves high detection accuracy of $98.8\%$ and reduces data transmission to $15\%$ of all samples, outperforming GAT-only and LSTM-only baselines. This work demonstrates a scalable, privacy-preserving HVAC security solution that leverages spatial-temporal graph learning and encrypted cloud processing for robust cyber-physical attack detection.

Abstract

Heating, Ventilation, and Air Conditioning (HVAC) systems are essential for maintaining indoor environmental quality, but their interconnected nature and reliance on sensor networks make them vulnerable to cyber-physical attacks. Such attacks can interrupt system operations and risk leaking sensitive personal information through measurement data. In this paper, we propose a novel attack detection framework for HVAC systems, integrating an Event-Triggering Unit (ETU) for local monitoring and a cloud-based classification system using the Graph Attention Network (GAT) and the Long Short-Term Memory (LSTM) network. The ETU performs a binary classification to identify potential anomalies and selectively triggers encrypted data transmission to the cloud, significantly reducing communication cost. The cloud-side GAT module models the spatial relationships among HVAC components, while the LSTM module captures temporal dependencies across encrypted state sequences to classify the attack type. Our approach is evaluated on datasets that simulate diverse attack scenarios. Compared to GAT-only (94.2% accuracy) and LSTM-only (91.5%) ablations, our full GAT-LSTM model achieves 98.8% overall detection accuracy and reduces data transmission to 15%. These results demonstrate that the proposed framework achieves high detection accuracy while preserving data privacy by using the spatial-temporal characteristics of HVAC systems and minimizing transmission costs through event-triggered communication.

Figures (5)

• Figure 1: The overview of the proposed framework.
• Figure 2: Schematic diagram of experimental RTU.
• Figure 3: Comparison of communication overhead between the traditional method and the proposed ETU-based framework.
• Figure 4: Confusion matrix for attack type classification.
• Figure 5: ETU–Cloud detection workflow: ETU anomaly probability (blue $\circ$), ETU trigger (orange ×), cloud receives data (green $\triangle$), cloud classification confidence (red $\square$), classification done (purple $\diamond$), and alarm returned (brown $\triangledown$).