建筑的智能化方案设计是智能建造的重要内容。但现有人工设计与自动优化设计方法往往难以学习和利用既有设计资料与经验，难以满足智能化和高效的建筑方案设计需求。具备数据学习与推理能力的深度学习方法有望推动建筑智能化方案设计的发展。因此，本文面向剪力墙建筑的抗震与隔震（振）设计需求，研究基于深度学习的智能化方案设计方法，主要工作内容如下：（1）基于图像合成生成对抗网络，提出输入建筑方案进而端到端生成剪力墙结构方案的智能化方案设计方法。开展约250份建筑-结构图纸的数据集构建、特征初步提取和数据特征分组研究，基于此数据集开展智能设计模型的训练，提出设计合理性评价方法，并针对24个平面方案设计与2个整体结构方案设计开展案例分析研究，验证了本方法的有效性。（2）针对设计条件文本-图像多模态数据融合学习并生成方案设计的需求，构建文本-图像特征融合生成对抗网络架构。开展神经网络模型、超参数和设计性能分析与对比，得到性能较好的生成对抗网络模型架构。通过平面与整体结构设计案例分析，验证了智能化设计的结构方案可同时基本满足输入图像与文本要求。（3）针对结构设计需满足力学性能的要求，构建物理增强生成对抗网络模型。构建力学性能评估模型，提出深度学习与高效力学性能计算协同工作方法，完成生成器-判别器-物理评估器网络架构的建立，分析并优化网络架构性能。开展近20种设计条件下的方案设计案例研究，证明了物理增强的有效性。（4）针对剪力墙结构基础隔震数据难获取问题，构建物理-规则引导的自监督生成对抗网络模型。构建隔震力学性能评估器与规则评价器，提出伪标签数据-物理-规则分阶段协同引导生成网络学习的方法，开展协同引导有效性分析。案例分析验证了无数据驱动的智能化基础隔震方案设计的合理性。（5）针对复杂多源振动干扰下建筑设备主动隔振设计需求，提出基于深度判别网络识别振动类型、设计智能隔振系统中振动控制参数的方法。收集约1200条振动干扰数据，并采用小波变换对振动信号特征进行初步提取，基于卷积网络对振动特征进行深度提取并识别分类，进而研究基于振动识别的控制参数自适应设计方法。不同隔振系统和振动干扰的案例研究验证了应用该方法后的隔振有效性。

The intelligent scheme design of buildings is an essential component of intelligent construction. However, the existing manual design and automated optimization design methods can hardly learn from and utilize the existing design data and knowledge, causing them challenging to meet the demands of intelligent and efficient building scheme design. In contrast, deep learning methods with data learning and inferencing capabilities are expected to promote intelligent scheme design of buildings. Therefore, this study investigates deep learning-based intellectual scheme design methods, oriented to the demands of seismic design and seismic (vibration) isolation design of shear wall buildings. The main work is as follows.(1) Based on generative adversarial networks of image synthesis, an end-to-end intelligent scheme design method is proposed to input architectural designs and then generate the corresponding shear wall structural scheme designs. The dataset establishment, preliminary design feature extraction, and data feature sub-division of approximately 250 pairs of architecture-structure drawings are conducted, and then the intelligent design model is trained based on the dataset. Moreover, the evaluation method for design rationality is proposed. Case studies of 24 planer and 2 overall structural designs are conducted to validate the efficiency of the proposed intelligent design method. (2) Oriented to the requirements of design condition texts and design drawings fusion learning and then generating scheme design, a novel generative adversarial network architecture of image-text feature fusion is proposed. The neural network model, hyperparameters, and design performance analysis are carried out to optimize the generative adversarial network's performance. The planar and overall structure design case studies verified that the intelligent design structure scheme could basically meet the requirements of input image and text simultaneously.(3) A physics-enhanced generative adversarial network model for structural scheme design is developed to meet the requirement of structural mechanics performance. First, a deep neural network-based mechanical performance evaluator is proposed, and then a mechanical analysis-deep learning coupling work mechanism is constructed. Subsequently, the neural network architecture of the generator-discriminator-physical evaluator is developed and then optimized through design performance analysis. Furthermore, case studies of scheme design under nearly 20 various design conditions are conducted to demonstrate the effectiveness of physical enhancement.(4) A physics-rule-guided self-supervised generative adversarial network model for scheme design of base isolation of shear wall structure is constructed to address the problem of challenging access to seismic isolation design data. First, a mechanical performance evaluator of base isolation structure and a rule-based evaluator are built. Then the method of "pseudo-labeled data"-physics-rule co-guided generative network learning is proposed, and the effectiveness analysis of the multi-phase co-guided mechanism is performed. Finally, the case studies verify the rationality of the intelligent scheme designs of structural base isolation without data drive.(5) To address the demand of vibration isolation design of equipment in building under complex multi-source vibration disturbances, the design method of vibration control parameters of the intelligent vibration isolation system is proposed, aided by deep discriminative networks-based vibration identification. About 1200 vibration disturbance data are collected; the preliminary time-frequency features of vibrations are extracted using wavelet transform; deep convolutional neural networks are then adopted to extract deep vibration features and learn identification; based on the vibration identification, adaptive design of vibration control parameters is conducted. The adaptive design method is analyzed and verified, through case studies of vibration isolation of different controlled systems under various vibration disturbances.