BP神经网络技术在城市建筑热环境研究中的应用

Application of BP Neural Network Technology in Thermal Environment Study for Urban Buildings

本文针对现代城市中越来越严重的热岛现象与能源问题,首先分析了北京市近60年的温度资料,可知60年来城区内的年平均温度升高了2.28℃,温度增幅为0.38℃/10 a。而后综合考虑城市建筑热环境的各种影响因素,利用BP神经网络技术建立了城市尺度下针对建筑热环境(温度)的预测模型,并对以往的数学模型和计算方法进行了改进。在改进后的预测模型中,通过枚举法选择隐含层最佳神经元个数,用贝叶斯正规化算法进行了网络训练,结果表明:与BP神经网络基本的L-M优化算法相比,该算法有较高的泛化能力和准确性,更适合于这一问题的研究。

In this paper,focusing on the more and more serious urban heat island phenomenon and energy problem in modern cities,the basic temperature data of Beijing in the past 60 years was firstly analyzed. It was known that the mean annual temperature in the internal city had increased by 2.28℃ in the past 60 years, and the increase rate was 0. 38 ℃/10a. Secondly, various influencing factors of urban thermal environment were comprehensively considered,and a prediction model was established by BP neural network in the city-scale, also improvements for past mathematical models and calculation methods were given out. With the improved prediction model, optimal neuron number of hidden layer was selected through enumeration method and network training with Bayesian normalization algorithm was carried out. Compared with the basic BP L-M optimistic algorithm, the algorithm presented in this paper possessed higher degree of generalization and accuracy.