摘要
室内照明设计初期往往需要多次方案调节迭代,并快速获得仿真结果。为辅助照明设计,应对传统室内照度仿真计算工具计算量大、耗时长的问题,本文提出了一种基于条件生成式对抗网络(Conditional Generative Adversarial Network,CGAN)的室内照度分布快速预测方法。首先利用Honeybee对2000个住宅建筑样本进行高精度的工作面照度模拟,构建了训练数据集。在此基础上,研究以Pix2pix算法为原型,设计了相关深度学习网络构架,经过200代训练,实现了从平面灯具布点图到室内照度分布图的映射。结果表明,模型在测试集上的结构相似性指数(Structural Similarity Index,SSIM)平均值为0.974,平均绝对误差(Mean Absolute Error,MAE)为0.014,预测精度较高。与传统数值仿真方法相比,该方法实现了毫秒级别的性能反馈,提升了设计效率。该研究为室内照明设计的快速评估和优化提供了一种高效的工具,对促进绿色建筑发展和提高室内环境品质具有重要意义。
In the early stages of indoor lighting design,multiple iterations and rapid acquisition simulation results are often required.To support lighting design and address the problems of high computational load and long processing time associated with traditional indoor illuminance simulation tools,this paper proposes a fast prediction method for indoor illuminance distribution based on Conditional Generative Adversarial Networks(CGAN).First,high-precision workplane illuminance simulation were conducted on 2000 residential building samples using Honeybee to construct a training dataset.Furthermore,based on the Pix2pix algorithm,a corresponding deep learning network architecture was designed.After 200 epochs of training,the model successfully established a mapping from 2D lighting layout diagrams to indoor illuminance distribution maps.The result show that the model achieved a high prediction accuracy,with an average Structural Similarity Index(SSIM)of 0.974 and a Mean Absolute Error(MAE)of 0.014 on the test set.Compared with traditional numerical simulation methods,this approach provides millisecond-level performance feedback,greatly enhancing design efficiency.This study offers an efficient tool for rapid evaluation and optimization in indoor lighting design,contributing significantly to the advancement of green buildings and the enhancement of indoor environmental quality.
作者
张力化
白金妮
韩臻
刘刚
ZHANG Lihua;BAI Jinni;Han Zhen;LIU Gang(School of Architecture,Tianjin University,Tianjin 300072,China;Tianjin Key Laboratory of Architectural Physical Environment and Ecological Technologies,Tianjin University,Tianjin 300072,China)
出处
《照明工程学报》
2025年第4期139-147,共9页
China Illuminating Engineering Journal
关键词
条件生成式对抗网络
室内照度预测
代理模型
conditional generative adversarial networks
indoor illuminance prediction
surrogate model
