神经网络在寒冷地区建筑能耗预测中的应用

Application of neural network in building energy consumption prediction in cold area

影响建筑能耗的因素很多,并且各自之间存在着非线性和相互耦合的特性,给建筑能耗的准确预测带来了一定困难,而神经网络是提高建筑能耗预测准确程度的重要方法。文章以寒冷地区建筑为例,提出了一种基于神经网络模型的能耗预测方法,从建筑的房间使用功能、室内热扰强度及建筑围护结构热工特性等方面选取16项影响能耗的参数作为神经网络预测模型的输入变量,与网络每层特性相结合,得到神经网络预测模型的结构为16-4-1。用该模型训练学习40组能耗模拟数据,从中选取25%的数据输入到已训练完成的神经网络预测模型中进行验证。结果表明:与DeST软件模拟结果相比,模型预测值的平均误差约为1.4%,从理论上验证了神经网络在建筑能耗预测方面的可行性;选取10栋实际建筑进行全年能耗预测,模型预测结果与能耗实际监测值吻合得较好。

There are many factors that affect building energy consumption,and there are non-linear and mutual coupling characteristics among various factors,which brings certain difficulties to the accurate prediction of building energy consumption,and neural network is an effective method to improve the accuracy of building energy consumption prediction.Taking the building in a cold area as an example,the author proposes a method of energy consumption prediction based on a neural network model,and selects 16 parameters that affect energy consumption as aspects of room use function of building,indoor thermal disturbance intensity,and thermal characteristics of building envelopes,and the characteristics of each layer of the network are combined to obtain the neural network prediction The structure of the model is 16-4-1.The input variables of the neural network prediction model are used to train and learn 40 sets of energy consumption simulation data,and 25%of the data are selected and input into the trained neural network prediction model for verification.The research results show that compared with the results of DeST,the average errors of the prediction value of the neural network energy consumption model is 1.4%,which theoretically verify the feasibility of the neural network in building energy consumption prediction.Through the annual energy consumption forecast of 10 actual buildings,the model prediction results are in good agreement with the actual monitoring value of energy consumption,which provides a new method for building energy consumption prediction in cold regions.