Модель прогнозирования материальных ресурсов и сметной стоимости на ранних этапах жизненного цикла объектов строительства

Модель прогнозирования материальных ресурсов и сметной стоимости на ранних этапах жизненного цикла объектов строительства
М. В. Гуреев, Aleksandr Nicolaevich Makarov
2024-12-02

automated machine learningconstruction cost estimationearly life-cycle predictionensemble modelsmaterial resource forecasting
Introduction. Digitalization of construction involves the possibility of forecasting material resources with a given degree of accuracy at the early stages of the life cycle of the construction object, which will make it possible to form cost indicators, as well as the volume of material resources and equipment — one of the key elements of management and planning of various stages of the life cycle of the object. The possibility of creating a forecasting tool based on artificial intelligence (AI) and machine learning (ML) technologies for these parameters seems to be a promising development direction, allowing to achieve a high level of accuracy in budget planning and project duration at the pre-project stage of an investment and construction project. Materials and methods. Design documentation for 37 multi-apartment residential buildings with allocated and normalized parameters: technical and economic indicators, material and technical resources, cost and others. To prepare and train AI models based on Python, the methods of the scikit-learn library were selected to compare the following mathematical models: decision trees, regressions and algorithms based on boosting. Results. The training and research were conducted using the automated machine learning (AutoML) method. Based on a comparison of the coefficient of determination R 2 and the standard deviation (RMSE), ensembles of models were selected that form a forecast for the volume of material resources and equipment, as well as for the estimated cost with an error range of ± 8 %. The input values of the models were 11 quantitative and qualitative parameters describing the characteristics of the planned object, the formation of which is possible at the early stages of the life cycle of the object without the development of design documentation. Conclusions. The results of the study demonstrate the possibility of obtaining actual design data already at the pre-design stage with accuracy corresponding to the stage of development of working documentation for the construction object. Thus, the accuracy of forecasts of the total estimated cost is significantly increased, and it also becomes possible to predict with a given accuracy the volumes of materials and equipment at the early stages of the life cycle of the construction object in order to optimize the entire construction process.
1
Forecasts use 11 quantitative and qualitative parameters describing planned buildings, available before design documentation is developed.
2
Model ensembles selected using R² and RMSE comparisons forecast material resources, equipment, and estimated cost with an error range of ±8%.
3
The models were trained on design documentation from 37 multi-apartment residential buildings using decision trees, regression methods, boosting algorithms, and AutoML.
4
The results demonstrate the feasibility of obtaining relevant design and cost data at the pre-design stage.
5
The study develops AI and machine-learning forecasting models for material resources, equipment volumes, and estimated construction costs at early project stages.

Multi-apartment residential construction projects at the early, pre-design stages of their life cycle

Forecasting material and technical resource volumes, equipment requirements, and estimated construction cost from early-stage project parameters

Publication Details
Publication Date
2024-12-02
Journal
Publisher
ISSN
Cited by
3
Access Type
Author Information
Authors
М. В. Гуреев
Aleksandr Nicolaevich Makarov
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%