建设需求量预测分析中的人工神经网络和多元回归方法

Construction demand forecasting by Artificial Neural Networks and Multiple Regression

利用人工神经网络(ANN)和多元回归(MR)预测方法分别基于中国统计年鉴和香港房屋署的相关数据对中国内地和香港地区的建设需求量进行预测,并对两种预测手段得到的预测结果的可信度和离散程度进行对比分析.基于ANN和MR两种预测手段的不同特性,从预测结果中可以看出,就香港地区的预测情况而言,ANN方法产生的结果比回归模型更加精确;从内地的预测结果来看,ANN和MR的预测精度几乎一致.对于存在较大波动性的数据而言,ANN模型建立的非线性关系可以更精确地描述预测结果,反之,两种预测模型的应用均可得出良好结果.同时,经预测得知,两地的建筑需求量都存在上升趋势,有关部门应采取相应措施提前做好规划工作.

Based on the relevant data from China Statistical Yearbook and the Hong Kong Housing Department,Artificial Neural Networks （ANN） and Multiple Regression （MR） were adopted to forecast construction requirement for China mainland and Hong Kong,the credibility and the degree of dispersion of forecasting results were analyzed.According to different characteristics of ANN and MR,for Hong Kong case,ANN method generates a more accurate result than regression model,but for the mainland,both ANN and MR perform well.The data which has great volatility is described more accuracy by non-linear relationship model which is generated by ANN method.Otherwise,the forecasting results with same credibility are generated by both two methods.Meanwhile,according to forecasting results,building demands have a raising tendency,as a result,the relevant departments should take appropriate measures to plan the work ahead of time.