ARIMA-BP复合模型在建筑能耗预测中的应用研究

The application of ARIMA-BP neural network composite model in building energy consumption forecast

建筑的能耗受到如季节、建筑的构造结构等多种因素的影响,目前对一栋建筑楼实现能耗预测往往采用单一模型,往往无法得到相对准确的结果.为了更好地描述建筑能耗规律,以南方某地为研究区域提出一种基于ARIMA和BP神经网络的复合模型,模型的实例数据来源为南方某地某市政办公楼近两年的能耗月数据.首先,通过ARIMA建模得到能耗值的拟合误差序列,再用BP模型修正误差值得到最终预测值.结果表明:复合预测模型的平均相对误差为0.278 3%,而单一模型则高达2.657 8%,复合模型的预测效果远优于单一模型,为准确实现建筑节能提出了一种新思路.

The energy consumption of buildings is influenced by many factors such as season and the structure of buildings. At present,as a single model is often used to predict the energy consumption of a building,it is unable to get the accurate results. In order to better describe the energy consumption of buildings,using somewhere in the south as the study area,we present a composite model of ARIMA and BP based on neural network. The source data of the model instance come from the energy consumption data of a municipal office building for two years. By building ARIMA model,we get fitting error series energy consumption value at first. Then,we use BP model to modify error value to get the final prediction value. The results show that the average relative error of the composite prediction model was 0.2783%,and the single model is as high as 2.6578%. The prediction effect of composite model is much better than that of the single model. The study has presented a new idea for building energy conservation.