SHEDAD: SNN-Enhanced District Heating Anomaly Detection for Urban Substations

TL;DR

SHEDAD addresses local, privacy-preserving anomaly detection in district heating networks by building a relative topology from substation supply-temperature profiles using a multi-adaptive $k$-NN graph and DTW distances under Sakoe-Chiba constraints. Edges are merged via Fleiss' kappa and clustered with Agglomerative Ward, followed by MST-based intra-cluster anomaly scoring using MAD and modified $z$-scores to identify supply-temperature and substation-performance anomalies. On a confidential DH dataset with 248 substations, SHEDAD identifies 30 supply-temperature anomalies and 14 performance anomalies, achieving approximately $65\%$ sensitivity and $97\%$ specificity, enabling targeted maintenance while preserving network confidentiality. The work demonstrates improved clustering validity (lower intra-cluster variance/distance) and provides a practical, privacy-aware pathway to reduce energy usage through focused interventions and iterative retraining for future predictive maintenance.

Abstract

District Heating (DH) systems are essential for energy-efficient urban heating. However, despite the advancements in automated fault detection and diagnosis (FDD), DH still faces challenges in operational faults that impact efficiency. This study introduces the Shared Nearest Neighbor Enhanced District Heating Anomaly Detection (SHEDAD) approach, designed to approximate the DH network topology and allow for local anomaly detection without disclosing sensitive information, such as substation locations. The approach leverages a multi-adaptive k-Nearest Neighbor (k-NN) graph to improve the initial neighborhood creation. Moreover, it introduces a merging technique that reduces noise and eliminates trivial edges. We use the Median Absolute Deviation (MAD) and modified z-scores to flag anomalous substations. The results reveal that SHEDAD outperforms traditional clustering methods, achieving significantly lower intra-cluster variance and distance. Additionally, SHEDAD effectively isolates and identifies two distinct categories of anomalies: supply temperatures and substation performance. We identified 30 anomalous substations and reached a sensitivity of approximately 65\% and specificity of approximately 97\%. By focusing on this subset of poor-performing substations in the network, SHEDAD enables more targeted and effective maintenance interventions, which can reduce energy usage while optimizing network performance.

Figures (5)

• Figure 1: Illustration of a DH network with a single heat source and various pipe diameters and flow rates. Substation supply temperatures are affected due to heat loss during transportation. Each color represents a neighboring group of substations with similar supply temperature profiles.
• Figure 2: Diagram of our proposed method for the relative approximation of a DH network.
• Figure 3: Performance evaluation for various time-series clustering methods, with mean intra-cluster variance in (a) and mean DTW intra-cluster distance in (b).
• Figure 4: Clusters of substations. Each line denotes the supply temperature profile of a substation, some of which include noise in the measurement data.
• Figure 5: Subset of supply temperature anomalies. Each substation is shown with its supply temperature (a) and return temperature (b), demonstrating substantial deviations and unusual patterns (cluster 20, 23, 27, and 30), while other substations (cluster 24 and 25) may be unusual though having normal operation.