Tracking decarbonisation: Scalable and interpretable data-driven methods for district energy systems

The urgent push for decarbonisation demands innovative, transparent methods to analyse and track decarbonisation strategies. This study addresses the problem of modelling energy consumption patterns at both building and district scales, ensuring transparency and scalability. By integrating well-established measurement and verification (M&V) techniques with interpretable data-driven modelling strategies, the research proposes a modelling workflow to track energy performance on a dynamic basis. The methods makes use of readily available metering data for electricity, district heating, and natural gas across a district, collected within a digital platform. A multi-resolution modelling approach is employed, with data at monthly, daily, and hourly intervals, that pinpoints anomalies and is meant to support a continuous refinement of operational strategies and efficiency measures. The Highfield Campus at the University of Southampton serves as the case study, illustrating how scalable, interpretable data-driven models can identify performance deviations and inform both short-term facilities management and long-term decarbonisation strategies. Findings reveal that simple and interpretable regression models can identify substantial variations in energy consumption pattern over longer time frames (ranging from months to years), whereas high-resolution analyses enhance the comprehension of dynamic operational patterns (days to hours). Both objectives can be achieved while maintaining a level of continuity in the modelling process, progressing from basic to detailed models while retaining interpretability. Further research will refine these models through additional physics-based constraints and explore deeper integrations with digital energy management platforms, offering replicable insights for broader district and urban-scale applications. • Southampton University Highfield Campus monitored for 7 years, from 2017 to 2023. • Multi-resolution modelling approach, with data at monthly, daily, and hourly intervals. • Time Of Week and Temperature algorithm modification and simplification. • Hourly models fitted with Coefficient of Variation of Root Mean Square Error in the range 7.4–23.5 %. • Further modelling workflow automation possible while enhancing interpretability and scalability.