基于嵌入法与集成学习的线路工程造价预测
Prediction of Transmission Line Cost Based on Embedding Method and Ensemble Learning
叶煜明 1钱琪琪 1万正东 2张继钢2
作者信息
- 1. 中国南方电网有限责任公司,广东广州 510530
- 2. 南方电网能源发展研究院有限责任公司,广东广州 510530
- 折叠
摘要
架空线路工程造价的准确预测对于工程建设质量及造价管控具有十分重要的意义.针对传统架空线路工程造价预测中遇到的特征维度过高、单一预测模型难以拟合复杂造价数据等问题,提出了基于嵌入法数据降维与集成学习的线路工程造价预测算法.首先通过嵌入法与极端梯度提升(extreme gradient boosting,XGBoost)模型对特征进行排序,筛选出对造价影响显著的特征完成数据降维.然后对XGBoost、随机森林、支持向量机(support vector machine,SVM)等模型进行融合,构成双层集成学习模型并对线路工程造价进行预测.最后基于某电网公司近年线路工程造价数据进行实例分析,分别与XGBoost、随机森林、SVM、极限学习机(extreme learning machine,ELM)与反向传播(back propagation,BP)神经网络等模型进行对比.实验表明预测结果的平均绝对百分比误差低于 4%,优于其他单一模型,对线路工程造价控制研究具有较大价值.
Abstract
Accurate prediction of transmission line project cost is of great significance to construction quality and cost control.Since the feature dimension in the traditional transmission line project cost prediction is too high and a single prediction model is difficult to fit the complex cost data,a transmission line project cost prediction method is proposed based on embedding dimensionality reduction and ensemble learning.Firstly,the features are sorted with the embedding method and the XGBoost model to screen out the features that have a significant impact on the cost,achieving the data dimensionality reduction.Then the XGBoost,random forest,SVM and other models are integrated to form a two-layer ensemble learning model.Finally,a case study is carried out based on the data of real transmission line projects,and the proposed method is compared with the XGBoost,random forest,SVM,ELM,and BP neural network models.The rusults show that the mean absolute percentage error of the proposed method is within 4%,which is superior to other single model,and is of great value to the research of transmission line project cost control.
关键词
输电线路/造价预测/集成学习/数据降维/嵌入法Key words
transmission line/cost prediction/ensemble learning/dimensionality reduction/embedding method引用本文复制引用
基金项目
中国南方电网有限责任公司科技项目(ZBKJXM2022 0003)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据