Underground spaces hold significant potential for enhancing urban resilience against disasters, a key dimension of sustainable urban development. However, due to persistent associations of underground environments with negative psychological perceptions, these spaces—despite their superior protective advantages—are often overlooked as viable refuge options during emergencies. Guided by the theoretical framework of environmental psychology, this research focuses on underground parking garages in Hangzhou, China as its primary research object. The target participants are residents of Hangzhou aged 18–58 years (encompassing diverse occupations such as students, office workers, and service industry employees), who represent potential users of such spaces as refuges. To explore human behavioral patterns, psychological responses, and needs related to underground refuge spaces, we employed a two-phase methodology: first, a questionnaire survey to capture broader behavioral tendencies and subjective perceptions; complementing this, Virtual Reality (VR) experiments—a more immersive method—utilizing semantic analysis and the Likert scale to assess psychological indicators influenced by underground environments. The experimental data were analyzed via mean analysis, correlation analysis, and multiple linear regression analysis to identify the key environmental factors that influence psychological responses, as well as their optimal design parameters. These analyses reveal significant correlations between various environmental factors and psychological indicators. This research synthesizes individuals’ psychological tendencies in underground environments and proposes quantitative physical design guidelines to meet fundamental psychological needs. The findings provide theoretical and practical support for the design of underground space for refuge and the development of sustainable urban emergency shelter systems, thereby contributing to resilient and sustainable urban development.

This research proposes a conditional diffusion-based workflow for early-stage floor plan design in university research buildings, addressing complex functional organization, strict boundary constraints, and quantitative area control. The method performs denoising directly in two-dimensional grid space and coordinates building outlines and functional area proportions through dual-condition injection using boundary masks and functional area matrices. A two-stage generation mechanism first constructs horizontal circulation and then generates the complete layout, while a statistic-network-guided explicit constraint improves global area consistency. Based on 600 standard-floor samples and an independent test set of 10 real projects, the method is evaluated through model comparison, ablation, and double-blind experiments. The results show that the proposed model achieves the best overall performance, with an FID of 50.3, a building boundary IoU of 99.9%, and horizontal circulation connectivity of 89.8%. The ablation results confirm that the two-stage mechanism and explicit statistical constraint substantially improve generation success and reduce area error. The expert evaluation indicates that AI-generated floor plans approach real cases in functional spatial form and design inspiration, although spatial organization rationality still requires improvement. The generated layouts can be converted into layered DXF files, supporting subsequent editing and human–AI collaborative design.

Against the background of green campus development and building decarbonization, photovoltaic (PV) systems in university research buildings involve not only energy performance but also campus character, architectural form, and spatial experience. To address the limited attention paid to such multidimensional adaptability, this study derives candidate indicators through a literature review and practical project analysis, refines them through two rounds of Delphi expert consultation, and determines their relative weights using the analytic hierarchy process (AHP). The resulting evaluation framework comprises five criteria and eighteen indicators covering planning pattern coordination, architectural form integration, building physical environmental performance, landscape character integration, and technical performance and innovation. Architectural form integration received the highest criterion weight (0.240), while power generation efficiency had the highest comprehensive indicator weight (0.142). Applied to the research building clusters at Zhejiang University Zijingang Campus, the framework yielded a comprehensive adaptability score of 87.3/100, corresponding to the excellent level. By translating multidimensional considerations into a unified quantitative result, the framework provides a structured basis for scheme assessment and adaptability classification at the design stage.