Using intelligent multi-objective optimization and artificial neural networking to achieve maximum solar radiation with minimum volume in the archetype urban block

Abstract

Nowadays, with the rapid development of cities, the demand for efficient and renewable energy, particularly solar radiation, is growing. Thus, urban planners and designers should employ solar energy analysis as a requirement to propose efficient solutions. This research aims to determine the maximum annual solar radiation with the least volume in various patterns of urban blocks and convert solar radiation modeling and analysis from simulation to emulation for types of urban blocks. As a result, simulation approaches such as Multi-Objective Optimization and Artificial Neural Networking were presented to characterize the influence of different urban scales on solar access and reach the optimal dimensions of the block. In Kermanshah, different patterns of urban blocks were investigated for the study on the urban block scale. The height, direction, number of plots, and scales impact access solar radiation energy and optimize the block’s dimensions. A study of 34 different urban blocks found that the ideal size is roughly 100∗100 m, while the best height is between 10 and 15 meters. Furthermore, the findings of predicting the data training models revealed that three outputs, the rate of solar radiation absorption, volume, and area of the urban block, can be predicted with an R2 score of 98% using the research input variables. Thus, radiation absorption estimations utilizing AI algorithms can ultimately replace traditional approaches and time-consuming large-scale energy estimates. The ANN methodology outperforms other machine learning techniques in predicting the objective function, which contains the amount of solar radiation absorption, volume, and surface area, according to the results of this research. Also, this research demonstrates how a simulation technique based on NSGA-II and ANN algorithms may drastically decrease computation time and costs while predicting optimum states prior to getting a real sample.