Research On Cost Prediction Of High-rise Residential Projects Based On Stacking Integration

The project cost management runs through the whole process of the construction project.As the core of the project cost management,the accuracy of the reasonable determination and effective control will directly affect the economic benefits of the construction project.In 2020,the Reform Plan of Project Cost was issued,which has emphasized enhancing the accumulation of historical project data,and using big data,artificial intelligence and other information technologies to provide basis for budget preparation.This plan reflects the trend of project cost reform and points out the direction of data application.In the information age,the trend of informatization of the construction industry develops speedily.How to make use of data efficiently in the early stage of construction projects to predict the project cost quickly and accurately is the key to investment control and cost planning,which is worthy of further studying.In recent years,BP neural network,multiple regression analysis,grey prediction,fuzzy mathematics and other methods have been widely used in the research of project cost prediction,however the constructed prediction models still have disadvantages.For example,although BP neural network model has certain nonlinear mapping ability,it is difficult to explain the model prediction reasoning process with a poor generalization ability due to its own “black box” characteristics.This paper analyzes the single prediction model advantages and disadvantages,introduces Stacking method,integrates multiple models with the integrated learning theory,and effectively improves the prediction accuracy of the project cost.Firstly,through literature research,the influencing factors of high-rise residential project cost are obtained.In the light of the initial index system,the thermal diagram method is used to remove the highly relevant indexes,ensure the scientificity and independence of cost characteristic indexes of high-rise residential projects,and construct the high-rise residential project cost prediction index system.Secondly,in the Python environment,the simulation analysis of historical data of many high-rise residential projects is carried out.The cost prediction model based on ridge regression,random forest and XGBoost algorithm is completed by using the characteristic index system of high-rise residential project.The parameters of each model are optimized by using ridge trace and grid search method,which both tests the prediction accuracy of single model,and improves the model prediction performance.Finally,a project cost prediction model based on Stacking fusion is constructed.According to the model performance evaluation index,the predicted values of the three single models and the fusion model are compared and analyzed with the real values of the samples,and the prediction of the integrated learning model with ridge regression as the meta-learner is obtained.The accuracy is the highest,and the model results after fusion are more stable.In this paper establishes a cost prediction model of high-rise residential projects based on Stacking algorithm,and the average absolute error is within ± 5%,which can meet the requirements of project cost management and control in the early stage,and has practical guiding significance for construction cost prediction.