Machine Learning Practice of Radio-oriented Module

Teams
Yang Lu, Mao Tianquan, Pu Hongjin,

Lu Liyuan, Cao Mengrui.

Tutors
Zhong Ximing, Li Kun

Abstract

Through case search and literature review, we understand and learn the concept of modules, sort out and learn some experience, and at the same time introduce composite factors (solar radiation, the ratio of usable area to building area, etc.) into the architectural shape layout in machine learning, to analyze the module unit in detail.


Research

This research project is a series of outcomes from the AI architecture course launched by Archiford and Xi’an University of Architecture and Technology. The course aims to use deep learning methods to design and advance architectural design solutions.

The problem of this project is to analyze how to differentiate the spaces with high – medium – low light intensity, or with/without sunlight from the light gain of modules in a certain volume, and to classify the spaces.

Through case studies and literature reading, we learned the concept of modules, sorted out and learned some experiences, and introduced compound factors (sunlight radiation, usable area to floor area ratio, etc.) into the building form layout in machine learning to analyze modular units specifically.


We used four expressions as summer radiation minimum value, winter radiation maximum value, body shape factor minimum value and satisfaction shape factor minimum value and satisfaction rating as X to describe the shape. The 27 Os and 1 as Y were brought into the code. And a 5-laver neural network was built.

The data set is between 0.1 which is more in line with the machine recognition logic. After learning, the machine considered less than 0.5 as O and greater than or equal to 0.5 as 1. greater than or equal to 0.5 as 1. After checking, the machine predicted more accurately. Export the weights file and bring it into GH. You can input the performance X and get the predicted value of Y to generate the design.

Leave a Comment