| 梁喜旺,付冬梅,杨焘.基于流形降维和梯度提升树的大气腐蚀速率预测模型[J].装备环境工程,2018,15(6):41-47. LIANG Xi-wang,FU Dong-mei,YANG Tao.Predicting Method of Atmospheric Corrosion Rate Based on Manifold Dimension Reduction and Gradient Boosting Decision Trees[J].Equipment Environmental Engineering,2018,15(6):41-47. |
| 基于流形降维和梯度提升树的大气腐蚀速率预测模型 |
| Predicting Method of Atmospheric Corrosion Rate Based on Manifold Dimension Reduction and Gradient Boosting Decision Trees |
| 投稿时间：2018-03-26 修订日期：2018-06-25 |
| DOI：10.7643/ issn.1672-9242.2018.06.008 |
| 中文关键词: 金属化学成分 腐蚀速率 流形方法 梯度提升决策树 |
| 英文关键词:chemical components of steels corrosion rate manifold methods GBDT |
| 基金项目:国家重点研发计划（2017YFB0702104）；博士后科学基金（2017M620615）；中央高校基本科研业务费（FRF-TP-16-082A1） |
|
|
|
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 为了挖掘大气腐蚀速率与金属化学成分和暴露时间两个影响因素的定量关系，针对数据集特点，提出一种局部保持投影（Locality Preserving Projection）和梯度提升树（Gradient Boosting Decision Trees）结合的大气腐蚀速率预测模型（LPP-GBDT）。方法 采用LPP算法对金属化学成分进行降维处理，得到金属化学成分低维特征，然后引入时间因素，并利用GBDT进行建立预测模型。以青岛海洋大气环境下积累的16年内的腐蚀速率数据进行模型性能验证，结果 LPP-GBDT模型测试集平均绝对误差为1.73 μm/a，平均绝对百分误差为6.30%。正交化LPP-GBDT模型测试集平均绝对误差为1.21 μm/a，平均绝对百分误差为4.42%。结论 与多个典型预测模型相比，LPP-GBDT模型基于暴露时间和化学成分因素实现了大气腐蚀速率较为准确的预测，对特定环境下金属选材具有一定的参考价值。 |
| 英文摘要: |
| Objective To analyze quantitatively the relationship between atmospheric corrosion rate and the two factors including outdoor exposure time and chemical components of steel materials, an atmospheric corrosion rate predicting model was proposed in combination with locality preserving projection (LPP) and gradient boosting decision trees (GBDT). Methods First, LPP was applied to have dimension reduction process on chemical components of steels to get low-dimensional features. Then GBDT were used to build a predicting model. Corrosion rate data of marine atmospheric environment within 16 years in Qingdao were used to validate the proposed model. Results Testing MAE and MAPE of LPP-GBDT model were 1.73 μm/a and 6.30% respectively. Testing MAE and MAPE of LPP-GBDT with orthogonalization were 1.21 μm/a and 4.42% respectively. Conclusion Compared with other common predicting methods, the proposed model has preferable prediction accuracy and offers some reference to steels selection in specific environment. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
