Abstract：With the increase in electricity demand，the rising maintenance costs of substation projects have put

power grid companies under huge market pressure. Traditional cost methods cannot meet the needs of modern

markets，so the construction of intelligent cost calculation models is the key to realize the modernization and

refinement of capital management and control of power grid enterprises. Focuses on the cost estimation of primary

equipment maintenance projects in substations，based on the compilation and analysis of historical project

settlement data，the key influencing factors of total project costs were identified，and then，three deterministic

prediction base models were established：a deep belief network，a BP neural network optimized by genetic

algorithm，and a grey prediction model. Furthermore，considering the impact of uncertain factors in practical

engineering scenarios on cost estimation，a probabilistic interval estimation model based on quantile regression was proposed by integrating the aforementioned base models. Finally，simulation tests using actual project data from Guangdong Province were conducted to verify the accuracy and reliability of the proposed models，which can

provide strong guidance and reference for engineering practice.

      Key words：substation equipment maintenance；prediction model；deep belief network（DBN）；BP neural

network（BPNN）；grey prediction；quantile regression

     Format Citation:黎立，庞圣养，钟荣豪,等. 基于分位数回归和机器学习的变电站检修工程造价区间预测模型[J].电气传动,2026,56(06):41-48. LI Li，PANG Shengyang，ZHONG Ronghao,et al. An interval prediction model for maintenance engineering cost of substations based on quantile regression and machine learning [J].Electric Drive,2026,56(06):41-48