Research on an Intelligent Optimization Algorithm for Indoor Space Design Based on Deep Learning

View at publisher Summarise with VIRAT
Authors: Wu Binxi
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Traditional indoor space design optimization algorithms suffer from insufficient multimodal information fusion, low solution feasibility, and low optimization efficiency. This paper proposes a Multimodal Feature Fusion Indoor Space Optimization Network (MFF-ISON). This algorithm innovatively introduces a spatial semantic attention mechanism to achieve precise cross-modal fusion of floor plans, user requirement text, and environmental parameters. Furthermore, a constraint-aware loss function is designed to incorporate spatial physical and ergonomic constraints into the optimization process, effectively reducing the conflict rate of solutions. To validate the algorithm's performance, an experimental dataset consisting of 5,000 sets of floor plan data and user requirements was constructed. Experiments were conducted using the PyTorch framework on an Intel Core i9-13900K CPU and an NVIDIA RTX 4090 GPU. Experimental results comparing a traditional genetic algorithm, a single CNN optimization algorithm, and a basic GAN algorithm show that the MFF-ISON algorithm achieves an average space utilization rate of 89.2%, a 15.6 percentage point improvement over the traditional genetic algorithm. The average comfort score is 92.3 out of 100, 12.7 points higher than the basic GAN algorithm. The algorithm runs for an average of 4.8 minutes, 42.3% less than the single CNN algorithm. The conflict rate is only 2.1%, 6.8 percentage points lower than the lowest value among the compared algorithms. These experimental results demonstrate that the MFF-ISON algorithm significantly improves the practicality and satisfaction of interior space design solutions while maintaining optimization efficiency, providing effective technical support for intelligent optimization of interior space design.

Fields of Study & Subjects

Architecture and Computational Design Advanced Neural Network Applications Architecture Design and Social History

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
1711–1715
Publication date
28 January 2026
Language
EN