基于神经网络的建筑室内温度预测比较研究

Contrast analysis of indoor temperature predict of buildings based on neural network

自然通风建筑室内温度受气象因素、建筑材料等多种非线性因素的影响,应用机理建模难度较大且计算复杂,很难得到精确的数学模型。利用神经网络不依赖模型和收敛速度快的优势和特性,可以很好地解决该问题。本文对经典的BP(Back Propagation)和RBF(Radial-based Function Method)神经网络建模方法进行对比研究,并对室内温度进行短期的预测。结果表明,在两种神经网络未经过优化的情况下,RBF神经网络在室内温度预测的应用要优于BP神经网络,基于RBF神经网络的室内温度预测方法在工程上有广阔的应用前景。

The indoor temperature of natural ventilation buildings is affected by many factors including meteorological factors,building materials etc. The application of mechanism modeling is difficult and complex which is difficult to obtain accurate mathematical model. Therefore in order to solve this problem it′s necessary to take advantage of the neural network characteristics of no-dependency on model and fast convergence speed. This paper presents comparative study of the classical modeling methods of BP（Back Propagation） and RBF（Radial-based Function Method） neural network method, then predict the short-term indoor temperature.Through the comparison between BP and RBF neural network we conclude that RBF neural network is better than BP neural network in accuracy, training efficiency and ability of generalization in this field.