A hybrid Gaussian process-integrated deep learning model for retrofitted building energy optimization in smart city ecosystems

Traditional retrofitting methods need a considerable upgrade; therefore, they often require significant time and financial investment, and can disrupt the building operations and occupant activities during implementation. Consequently, developing fast-response retrofitting solutions to save energy on urban and large scales is critical for city planners and policymakers. This study integrates a Gaussian Process-based Deep Learning (GPDL) model to retrofit buildings on a metropolitan scale, aiming to accelerate the transition towards smart cities. Gaussian Process offers a probabilistic approach to assess uncertainty in data points, while deep learning captures complex data patterns. The hybrid approach enhances the accuracy and reliability of end use intensity (EUI) predictions, ultimately supporting the computation of the primary energy factor (PEF) for improved decision-making in energy management. The proposed GPDL model was applied to a case study consisting of 6076 buildings in Mullingar City, which is located in Westmeath County, Ireland. The case study evaluated various heating, ventilation, and air conditioning systems (HVAC) such as gas boilers, electric heaters, and heat pumps, as well as different wall structures with distinct insulation layers in retrofitted buildings. The results demonstrated a reduction of EUI by about 52.4 %, resulting in 39.2 % energy savings for the overall city, with the proposed GPDL. The developed GPDL presented high accuracy mean square error: 1.3005, root-mean-square error: 0.0036, and mean absolute percentage error: 0.0022, leading to rapid EUI predictions within seconds, compared to the state-of-the-art linear and spline regression models, which require hours for similar estimations. The findings also underscored that the choice of heating, water system, HVAC, fans, interior lighting, electric equipment, pump, and gas equipment system has a greater impact on EUI than wall structure, leading to valuable recommendations of natural gas or renewable energy sources due to lower primary energy factor for urban planners focused on developing smart cities. • Formulated energy intensity metrics and primary factors to optimize retrofitting efficiency. • Integrated Gaussian Process (GP) model with Deep Learning for smarter energy optimization. • Designed an architecture highlighting functional components of the GP-Deep learning framework. • Investigated energy dynamics of 6076 buildings in Mullingar City, Ireland, urban and suburban.