A Data-Driven Design Approach for Eco-Districts with Early-Stage Real-Time Predictive Energy Modelling

The rapid urbanisation and growing awareness of sustainable development have led to the emergence of eco-districts as a potential solution for creating sustainable urban environments. However, existing eco-district evaluation systems and design approaches face challenges in providing actionable guidance and integrating data-driven methods, particularly in the early design stages. This research aims to develop a comprehensive eco-district data-driven design (DDD) approach that integrates a real-time district energy surrogate model and eco-district design evaluation indicators to achieve immediate evaluation and optimisation of district design at early stages. The research establishes a novel eco-district evaluation system tailored for the DDD approach by analysing existing systems, generating a customised indicator classification framework, and selecting the most pertinent criteria through a structured process. A district-level energy surrogate model is developed using deep neural networks (DNNs) to provide rapid performance feedback, even with limited design details. This model demonstrates high accuracy, achieving R2 values of 0.978 and 0.988 for predictions of cooling and heating energy consumption, respectively. The research then formulates a generic DDD methodology for eco-district planning, employing a grid-based genetic algorithm (GA) for multi-objective optimisation, integrating real-time assessment with interactive energy modelling. Case studies are conducted to demonstrate the practical viability of the model and the DDD approach in real-world contexts. Comparative experiments are performed to benchmark the outcomes of the GA against those created by human design experts, assessing the GA's strengths and limitations. It reveals that GA-generated designs with DDD approach outperform human-designed solutions, achieving a 9.97% reduction in total heating and cooling energy use intensity (EUI) and a 38.6% higher average aggregate fitness value. Moreover, the DDD approach significantly reduces design time by 97.8% compared to human designers. The research contributes to the advancement of data-driven approaches in eco-district planning and design, offering an integrated framework that combines eco-district evaluation, district-level energy modelling, and optimisation methods. This innovative approach provides new tools and perspectives for achieving urban sustainability goals, paving the way for more sustainable and energy-efficient urban developments.