Power engineering cost prediction method based on hybrid natural gradient and light gradient boosting
SONG Kun
SHI Jing
ZHENG Yingnan
ZHANG Ruyu
LIU Bonan
Abstract：[Objective]Accurate prediction of construction costs in power engineering is crucial for resource allocation and decision optimization.Traditional cost estimation methods rely on manual experience,which are often influenced by the complexity and uncertainty of engineering projects and thus lead to a large prediction bias.In recent years,machine learning techniques have gained extensive attention,which provide new solutions for cost prediction in power engineering.However,existing models often lack uncertainty estimation for the prediction results and have problems of low prediction accuracy,low training efficiency,and being prone to overfitting.This paper proposed a power engineering cost prediction method based on a hybrid model of natural gradient boosting(NGBoost)and light gradient boosting(LGBoost),aimed at improving prediction accuracy while providing uncertainty estimation for the predicted results.[Methods]In this study,the NGBoost model,which could estimate the probability distribution of predicted values,was introduced into the field of power engineering cost prediction.However,considering the low efficiency and overfitting problems of NGBoost,the histogram optimization algorithm of LGBoost was adopted and integrated into NGBoost.This resulted in the development of a hybrid model combining NGBoost and LGBoost,which not only improved prediction accuracy but also enabled the quantification of uncertainty in the prediction results.[Results]To validate the effectiveness of the proposed model,this study used the BIM database of real engineering costs,spanning from 2002 to 2022,which included 2 000 pieces of power engineering data.Experimental results show that the proposed hybrid model outperforms others in terms of correlation coefficient,root mean square error,and mean bias error.Additionally,the prediction results for the test set reach a probability of 94.3％at the 95％confidence level.Compared to NGBoost,the hybrid model not only enhances prediction accuracy but also effectively avoids overfitting and demonstrates better training efficiency.[Conclusion]The hybrid NGBoost and LGBoost model presented in this paper improves prediction accuracy and is able to provide uncertainty estimation,fulfilling the diverse needs of power engineering cost prediction.Experimental validation confirms the advantages of the model in prediction accuracy,generalization capability,and training efficiency,which makes the model particularly suitable for cost estimation of complex power engineering projects.The highlight of this study is the development of a novel hybrid model that combines the high training efficiency of LGBoost and the uncertainty estimation capability of NGBoost,addressing the limitations of traditional models such as low training efficiency and overfitting and having the capability to quantify the uncertainty of prediction results.The proposed model can offer valuable support for optimizing resource allocation and improving decision-making efficiency.
Keywords：power engineeringcost predictionnatural gradient boosting modellight gradient boosting modelhybrid modelhistogram optimization algorithmprediction resultuncertainty
Publication Date：2025-03-31
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 183-189 )