基于不同机器学习方法的建筑工程造价预测研究

Research on Construction Engineering Cost Prediction Based on Different Machine Learning Methods

工程造价预测是建筑行业中重要的工作,准确地预测工程造价对于前期的投资决策具有重要的现实意义。为了提高造价预测的准确性,本文提出一种应用集成学习算法(XGBoost)和单一学习算法(BP、SSA-BP)构建预测模型的方法,采用统计学习的方法得到8种对工程造价有影响的变量因素:总建筑面积、高度、层数、层高、基础类型、结构形式、抗震烈度和外立面材料。以8种影响因素作为模型的输入变量,单位造价作为输出变量,通过对预测结果的指标评估得出单一学习算法中SSA-BP模型的预测性能最好。在此预测模型的基础上,使用SHAP分析方法,得到8种变量因素在模型中的重要性占比排序,并以此重新构建8种不同组合的输入变量来检验SSA-BP模型的可靠度。

Forecasting of engineering costs is a very important task in the construction industry.Accurate prediction of engineering cost is of great relevance for investment and decision making.In order to improve the accuracy of cost prediction,a new forecasting method is proposed in this paper.This approach applies integrated learning algorithms(XGBoost)and single learning algorithms(BP,SSA-BP)to construct predictive models.And the statistical learning method was used to obtain eight variable factors that have an impact on the engineering cost(total floor area,height,number of floors,storey height,basic type,structural form,seismic intensity and fa?ade material).In this study,eight influencing factors were used as input variables of the model and unit cost as output variables.A comparison of the three machine learning models shows that the SSA-BP model has the best predictive ability.Based on this prediction model,SHAP analysis was used to obtain the order of importance share of the eight variable factors in the model.The reliability of the SSA-BP model was tested by reconstructing eight different combinations of input variables using importance ranking.