Generation Design Of Residential Building Plan Layout Based On Deep Learning

In this paper,we try to put forward a method of generating house floorplan for building design in the construction planning industry.We use the new technology of deep learning to learn the design ideas of past architects in the design of residential buildings.To a certain extent,it can realize the automatic design of composite layout of residential plan,simplify the initial design process of designers,and explore the application prospect of CAAD in a deeper level.In this paper,we take a kind of residential building plan as samples,which are encoded as graph,a kind of computer data structure.Graphs include "node-edge",the rooms are taken as the nodes,and the connection relationships of each rooms are taken as the edge.We use the graph neural network model based on graph representation learning to train the samples and get the residential building floorplan score prediction model.We try to explore what kind of combinations are good designs.Secondly,we extract some main room combinations or function blocks related to high scores as subgraphs,and combine them into new building combinations to generate new and unique designs.The main work of this paper is as follows:First,we use graph representation learning method to encode floorplan samples into graph data structure.By extracting the important combination related to specific objective function,we find the potential topological features that constitute the basic components of residential building design;Second,We combine the discovered building blocks into new designs that meet the needs of new users and evaluate the feasibility of the generated candidate solutions.