Urban Buildings Thermal Environment Research based on BP Neural Network

摘要

Aiming at more and more serious urban heat island intensity and energy consumption issues in our modem society, the urban thermal environmental problems have increasingly became the focus of all the people. Taking Beijing city as the research object, this article improved the previous mathematical models by BP neural network technology and proved the feasibility of this approach here. By numerical fitting calculation, almost 60 years of temperature data were analyzed. The urban annual average temperature increased by 2.28~C totally during the last 60 years, with the warming rate 0.38°C/10-years. Also the warming rate in winter was obviously higher than it in summer. Because of the multivariate, distributed parameters and nonlinear features of the urban buildings thermal environment, various influencing factors of this research system were comprehensively considered and analyzed from the perspective of urban thermal balance, and a prediction model was established by BP neural network in the city-scale in this article, improving the previous ones. In the new mathematical model, the adaptive regulation algorithm was used to select the best number of the hidden layer neurons, and the Bayesian regularization algorithm was used for network training. The result showed that under the same network size and parameters setting, this improved algorithm had better generalization capacity and accuracy than the basic BP algorithms or other improved BP algorithms. So we concluded this method was suitable to predict the urban buildings thermal environment temperature and for the further research of this field.