AI-driven Classification of Heat Pump Usage in UK Homes

With environmental sustainability, low carbon, and renewable heat sources, Heat
pumps (HP) are the core pathway for decarbonising heat in the UK. This paper aims to investigate the performance and usage of HP and classify the HP demands by discretising 24 hours of space heating (SH) consumption and SH periodicity of HP operation across four major demand shape categories. Classification is achieved by applying an unsupervised machine learning approach, the Self-Organising Map (SOM), to clustering time-series features derived from the BEIS-measured dataset. The proposed methodology investigated selected 185 dwellings with Air Source Heat Pumps (ASHP) and utilised BEIS measured dataset. The data input matrix of 185[Dwellings] x 46[Features] is fed to the SOM machine learning model for clustering the HP demand shape. The SOM model performed well in clustering the SH demand shape of 185 dwellings to three unique demand shapes under the given set of computed time-series features. This approach will enable efficient tools for HP demands monitoring and HP operation management in buildings and industry.