Under extreme high-temperature conditions, heat accumulation in rooftop photovoltaic (PV) arrays can substantially reduce PV conversion efficiency and output power. To improve the thermal and power-generation performance of rooftop PV systems, this study developed a computational fluid dynamics (CFD) model for a rooftop PV array and validated it using field measurements from Chongqing, China. The relative root mean square errors (rRMSEs) between simulated and measured backsheet temperatures at the three measurement points were 6.31%, 7.52%, and 8.45%, respectively, indicating acceptable model accuracy. The effects of mounting height, tilt angle, and front-to-rear row spacing on PV backsheet temperature, conversion efficiency, and output power were then investigated. A central composite design (CCD) within response surface methodology (RSM) was used to establish regression models linking the design variables to the objective functions. Finally, an NSGA-III-based multi-objective optimization framework combined with TOPSIS was used to identify the optimal configuration. For the rooftop PV array studied under extreme summer conditions in Chongqing, the TOPSIS-selected compromise solution corresponded to a mounting height of 0.90 m, a tilt angle of 15.63°, and a front-to-rear row spacing of 2.96 m. Compared with the original configuration, the optimized passive installation geometry reduced the peak PV backsheet temperature by 2.3 °C without active cooling, water consumption, or additional energy input. Under the same meteorological and irradiance conditions, this temperature reduction increased conversion efficiency by 0.5% and output power by 0.4%. Detailed inter-row short-wave shading and electrical mismatch were not explicitly modeled. Therefore, the row-spacing effect mainly reflects changes in ventilation and module temperature under the same irradiance input. The proposed framework provides a practical reference for installing and optimizing rooftop PV arrays in hot-climate regions.

Using a data-driven framework, we examined when recorded older-adult metro activity occurs and whether detailed spatial measures predict station activity better than simple contextual variables. The data comprise 22,908,241 recorded entry and exit events from age-65+ fare-exempt cards at 263 Chongqing metro stations in March 2025, aggregated by station and clock hour. We characterized entry–exit directionality and modelled total and population-adjusted activity using older-population catchments, points of interest (POIs), metro topology, pedestrian networks, and route-level terrain. Predictive comparisons used geographically blocked cross-validation, with all preprocessing and tuning confined to the training data. Entry and exit counts showed clear temporal and spatial differences, although net inflow prediction transferred poorly to held-out areas (out-of-fold R² = −0.099 to 0.042). Population aged 65+ and transfer status supplied most of the transferable signal (out-of-fold R² = 0.472). POIs added little (ΔR² = 0.007), while richer spatial representations did not improve held-out prediction. Thirty repeated geographic partitions confirmed a small average advantage for the POI-augmented model but also showed sensitivity to the partition. These findings support parsimonious station activity modelling. A complementary screening analysis defined a reproducible 45-station pool for targeted access audits.