Optimising IP based SCADA IEC 60870-5-104 protocol performance through adaptive machine learning driven traffic shaping in classical computing environments

Protocol/Multi-Protocol Label Switching (IP/MPLS) utility networks. While IEC 104 is widely adopted for realtime telecontrol messaging between control centres and substations, its performance can degrade under adverse network conditions such as congestion, latency, and packet loss. Most existing research prioritises cybersecurity and anomaly detection for IEC 104, leaving a notable gap in adaptive performance optimisation techniques suited for legacy, CPU bound SCADA systems. To address this gap, the paper proposes an adaptive traffic shaping framework that integrates machine learning–driven traffic analysis with deterministic IP/MPLS Quality of Service (QoS) mechanisms. The framework introduces an intelligent middleware layer that leverages protocol semantic awareness such as Application Service Data Unit (ASDU) types and Causes of

Transmission to classify IEC 104 traffic by operational criticality. Supervised learning models forecast potential congestion using network performance metrics, while unsupervised clustering dynamically groups IEC 104 flows by urgency and behavioural characteristics. These insights guide hierarchical prioritisation strategy aligned with utility telecommunications best practices, ensuring that protection and control messages are prioritised over routine monitoring and non-SCADA services.

Deterministic enforcement is achieved through Differentiated Services Code Point (DSCP) marking, Multiprotocol Label Switching (MPLS) Experimental (EXP) mapping, class based queuing, and strict priority scheduling within the IP/MPLS core. A simulation testbed replicating realistic utility network conditions including multimedia congestion, alarm storms, and network failures demonstrates the framework’s ability to sustain bounded latency and minimise packet loss for high critical traffic. Results indicate that machine learning (ML) assisted prioritisation, when combined with established QoS mechanisms, significantly enhances IEC 104 performance without requiring specialised hardware or disruptive architectural modifications.

Overall, the study presents a novel, integrated approach that combines predictive analytics with deterministic transport behaviour, offering a practical and scalable pathway to improving

SCADA protocol performance in modernised yet resource constrained utility environments.