Image Generation Technique of Indoor Temperature Distribution Using Transposed Convolutional Neural Network

Abstract

It is important to maintain a comfortable indoor thermal environment because people spend most of their time in buildings. Computational fluid dynamics (CFD) applications can not only simulate the indoor environment under various boundary conditions but also visualize analysis results as contour plots to help humans understand them easily. However, calculations are required for each boundary condition, and a significant amount of time is required to obtain the results. In recent years, artificial intelligence has been utilized in various fields, including image processing technology. Generating temperature distribution images using artificial intelligence instead of CFD analysis is expected to reduce the number of CFD cases and time required for preliminary studies. In this study, a transposed convolutional neural network was used to generate a model for the fast prediction of indoor temperature distribution, and case studies were conducted using a combination of training data. Consequently, it was confirmed that by appropriately selecting 108 out of 300 data points arranged in a grid, a model that predicts temperature distribution with generally good accuracy was generated for the untrained data.