Development of a hybrid machine-learning and optimization tool for performance-based solar shading design
Authors:
Maryam Daneshi,
Reza Taghavi Fard,
Zahra Sadat Zomorodian,
Mohammad Tahsildoost
Abstract:
Solar shading design should be done for the desired Indoor Environmental Quality (IEQ) in the early design stages. This field can be very challenging and time-consuming also requires experts, sophisticated software, and a large amount of money. The primary purpose of this research is to design a simple tool to study various models of solar shadings and make decisions easier and faster in the early…
▽ More
Solar shading design should be done for the desired Indoor Environmental Quality (IEQ) in the early design stages. This field can be very challenging and time-consuming also requires experts, sophisticated software, and a large amount of money. The primary purpose of this research is to design a simple tool to study various models of solar shadings and make decisions easier and faster in the early stages. Database generation methods, artificial intelligence, and optimization have been used to achieve this goal. This tool includes two main parts of 1. predicting the performance of the user-selected model along with proposing effective parameters and 2. proposing optimal pre-prepared models to the user. In this regard, initially, a side-lit shoebox model with variable parameters was modeled parametrically, and five common solar shading models with their variables were applied to the space. For each solar shadings and the state without shading, metrics related to daylight and glare, view, and initial costs were simulated. The database generated in this research includes 87912 alternatives and six calculated metrics introduced to optimized machine learning models, including neural network, random Forrest, support vector regression, and k nearest neighbor. According to the results, the most accurate and fastest estimation model was Random Forrest, with an r2_score of 0.967 to 1. Then, sensitivity analysis was performed to identify the most influential parameters for each shading model and the state without it. This analysis distinguished the most effective parameters, including window orientation, WWR, room width, length, and shading depth. Finally, by optimizing the estimation function of machine learning models with the NSGA II algorithm, about 7300 optimal models were identified. The developed tool can evaluate various design alternatives in less than a few seconds for each.
△ Less
Submitted 9 January, 2022;
originally announced January 2022.
