基于多阶段特征提取的变电工程典型造价预测方法  

Forecasting Method of Typical Costs of Substation Projects Based on Multi-stage Feature Extraction

作　　者：乔慧婷 吴良峥 张继钢 万正东 QIAO Hui-ting;WU Liang-zheng;ZHANG Ji-gang;WAN Zheng-dong(China Southern Power Grid Energy Development Research Institute Co.,Ltd.,Guangzhou 510530,China)

机构地区：[1]南方电网能源发展研究院有限责任公司,广东广州510530

出　　处：《计算机技术与发展》2025年第4期100-106,共7页Computer Technology and Development

基　　金：中国南方电网有限责任公司科技项目(ZBKJXM20220003)。

摘　　要：在电力建设快速发展的背景下,准确制定典型造价对于有效控制工程成本具有重要意义。然而现有典型造价修订方法面临着编制精准性不足的挑战。因此,提出了一种基于多阶段特征提取的变电工程典型造价预测方法。该方法首先通过线性支持向量机-递归特征消除(LSVM-RFE)算法对特征进行重要性排序,选择出关键特征;接着,利用主成分分析(PCA)算法对LSVM-RFE模块筛选后的剩余特征进行降维处理,提取出主要信息并降低模型复杂度;最后,引入CatBoost模型对PCA降维后的数据进行预测。在某电力公司的真实变电工程典型造价数据集上进行的实践验证表明,该方法在多个误差评价指标上均优于其他对比模型,且通过消融实验验证了该方法中的多特征提取模块对整体预测性能的提升,为电网公司修订典型造价提供了一种科学、有效的新途径。In the context of rapid development in power construction,accurately formulating typical cost is of significant importance for effective project cost control.However,current methods for revising typical cost face challenges in precision.Therefore,a new method for predicting typical costs of substation projects is proposed.Firstly,the Linear Support Vector Machine-Recursive Feature Elimination(LSVM-RFE)is utilized to rank the importance of features and select key features.Secondly,the PCA is employed to perform dimensionality reduction on the remaining features selected by the LSVM-RFE module,extracting major information and reducing model complexity.Finally,the CatBoost model is introduced to predict the data after dimensionality reduction by PCA.Practical validation on a real substation project cost dataset from a power company demonstrates that the proposed method outperforms other comparison models in various error evaluation metrics.Additionally,the ablation experiment has verified the enhancement of the overall predictive performance by each feature processing module in the method proposed,providing a scientific and effective new approach for power grid companies to revise typical costs.

关 键 词：变电工程 典型造价 预测 递归特征消除 主成分分析 CatBoost 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]