A Coordinated Strategy for Residential Energy Consumption and Photovoltaic Generation Based on Machine Learning and Multi-Objective Optimization: A Case Study of Lhasa

Lhasa combines abundant solar resources with a high heating demand and low winter solar altitude, making residential morphology simultaneously important to operational energy use and roof–façade photovoltaic (PV) generation. Based on a residential-area inventory covering more than 80 residential areas and 15 typical slab-type blocks, this study developed locally constrained parametric prototypes and generated 2186 valid physics-based samples. Thirteen morphological predictors were retained for surrogate modelling, followed by sunlight-constrained bi-objective optimization at floor area ratio (FAR) = 1.5, 1.8, and 2.2. On the independent test set, the selected energy and PV surrogate models achieved R2 values of 0.999 and 0.989, respectively. The Pareto fronts at FAR = 1.5 and 1.8 included near-zero annual net-energy solutions, whereas FAR = 2.2 retained a minimum deficit of 0.682 × 106 kWh. Low-energy solutions generally favoured a lower site coverage, continuous elongated slabs, and a greater mean height, whereas high-generation solutions favoured larger building footprints, a lower height, and controlled shading. As FAR increased, façade PV made a larger contribution, but this was insufficient to offset the decline in the rooftop supply and the rise in total energy demand. Energy-balance-oriented solutions generally concentrated the building orientation within 5–15° and the building depth within 14.0–15.5 m, with the density and height requiring a joint adjustment across FAR scenarios. The results provide an interpretable basis for the early-stage morphology screening and planning control of slab-type residential development in Lhasa.