A data-driven approach for discovering heat load patterns in district heating

Understanding the heat usage of customers is crucial for effective district heating operations and management. Unfortunately, existing knowledge about customers and their heat load behaviors is quite scarce. Most previous studies are limited to small-scale analyses that are not representative enough to understand the behavior of the overall network. In this work, we propose a data-driven approach that enables large-scale automatic analysis of heat load patterns in district heating networks without requiring prior knowledge. Our method clusters the customer profiles into different groups, extracts their representative patterns, and detects unusual customers whose profiles deviate significantly from the rest of their group. Using our approach, we present the first large-scale, comprehensive analysis of the heat load patterns by conducting a case study on many buildings in six different customer categories connected to two district heating networks in the south of Sweden. The 1222 buildings had a total floor space of 3.4 million square meters and used 1540 TJ heat during 2016. The results show that the proposed method has a high potential to be deployed and used in practice to analyze and understand customers’ heat-use habits.

💡 Research Summary

This paper presents a fully data‑driven methodology for automatically discovering heat‑load patterns and identifying anomalous customers in large district‑heating (DH) networks, without requiring any prior domain knowledge. The authors first define a “heat‑load profile” as the average hourly heat consumption of a building, organized into a weekly (168‑hour) matrix for each of four calendar seasons (winter, early‑spring/late‑autumn, late‑spring/early‑autumn, and summer). This representation captures daily cycles, weekday‑weekend differences, and seasonal variations in a compact yet information‑rich form.

To cluster these high‑dimensional time‑series profiles, the study adopts the k‑shape algorithm, a centroid‑based clustering method that uses the Shape‑Based Distance (SBD) metric. SBD aligns time series based on shape rather than raw Euclidean distance, offering robustness to small temporal shifts and noise while remaining computationally efficient compared with Dynamic Time Warping. The algorithm iteratively assigns profiles to clusters and updates cluster centroids by extracting a representative “shape” that preserves the characteristic pattern of all members. Each centroid is termed a “heat‑load pattern,” providing a visual and quantitative summary of the typical behavior of a group of buildings.

Anomalous customers are detected in two complementary ways. First, a distance‑based outlier test flags any building whose SBD to its cluster centroid exceeds a predefined threshold, labeling it a “pattern‑deviation” customer. Second, the authors incorporate domain‑specific expectations for each customer category (e.g., schools, hospitals, residential blocks) and identify “control‑incompatible” customers whose observed patterns diverge from the expected control strategy for that category. This dual approach captures both statistical outliers and operationally problematic cases such as mis‑configured substations or unsuitable temperature set‑points.

The methodology is validated on a comprehensive case study involving two DH networks in southern Sweden. The dataset comprises 1,222 buildings across six customer categories (residential, commercial, educational, medical, public services, and mixed‑use), covering a total floor area of 3.4 million m² and an annual heat demand of 1,540 TJ for the year 2016. For each building, hourly heat‑meter readings were aggregated into the seasonal weekly matrices described above.

Clustering results reveal that each customer category naturally splits into 3–5 meaningful clusters, with average intra‑cluster SBD values above 0.85, indicating high cohesion. Representative patterns illustrate, for example, the sharp winter peak typical of schools, the relatively flat but non‑zero summer load of hospitals, and the pronounced weekday‑weekend contrast in office buildings. Anomalous detection flags approximately 4 % of the buildings as either pattern‑deviation or control‑incompatible. Investigation of these cases shows that many are linked to substation control faults, inappropriate temperature set‑points, or unexpected occupancy patterns, confirming the practical relevance of the approach.

The paper’s contributions are threefold: (1) it delivers a scalable, fully automated pipeline for heat‑load pattern discovery applicable to entire DH networks; (2) it demonstrates that k‑shape clustering, with its shape‑preserving centroid extraction, is well suited for seasonal weekly heat‑load time series; (3) it provides a systematic framework for flagging customers that may degrade overall network efficiency, thereby supporting the transition toward “fourth‑generation” DH systems that operate at low distribution temperatures and high flexibility.

Beyond the presented results, the authors discuss future extensions such as online clustering for real‑time monitoring, integration of additional exogenous variables (weather, occupancy, building envelope characteristics), and the use of the identified patterns to inform demand‑side management strategies and optimal control design. By enabling data‑driven insight at the network level, this work lays a solid foundation for more intelligent, sustainable, and carbon‑neutral district‑heating operations.