基于残差网络的风电机组基础健康监测数据修复研究
RESEARCH ON HEALTH MONITORING DATA RESTORATION OF WIND TURBINE FOUNDATION BASED ON RESIDUAL NETWORK
魏焕卫 1宋志鑫 2雷树立 2惠俊梅 3郑晓1
作者信息
- 1. 山东建筑大学土木工程学院,济南 250101;山东建筑大学建筑结构加固改造与地下空间工程教育部重点实验室,济南 250101;山东建筑大学地铁保护研究所,济南 250101
- 2. 山东建筑大学土木工程学院,济南 250101;山东建筑大学建筑结构加固改造与地下空间工程教育部重点实验室,济南 250101
- 3. 山东大卫国际建筑设计有限公司,济南 250100
- 折叠
摘要
为精准有效地修复连续性异常数据,提出一种基于残差块优化卷积神经网络的残差网络数据修复模型.以乳山风电场的风电机组基础健康监测数据为例对模型进行工程验证.同时选取具有修复功能的多种模型对实际异常数据进行修复验证,并对所有模型的性能以及自身的修复精度进行对比分析.结果表明:ResNet模型避免了FCN以及CNN模型存在的缺陷,提高了数据修复的精度;ResNet模型适用于缺失或异常比例低于30%的数据修复;ResNet模型修复实例的结果符合数据变化趋势,能较好吻合监测数据的原始曲线.
Abstract
To accurately and effectively repair the abnormal data of the continuous monitoring dataset,a residual network data repairing(ResNet)model was proposed based on residual block optimized convolutional neural network in this paper.The engineering validation of the proposed ResNet model was carried out by using the health monitoring data of the foundation of the onshore wind turbine at Rushan Wind Farm.Some models with repair functions were also selected to repair the abnormal data of the monitoring dataset.According to the results of the repaired monitoring dataset,the comparative analysis was conducted on the repair performance and accuracy of these models.The results reveals that comparing with FCN and CNN model,ResNet model exhibits a rather high accuracy of abnormal data repair;the ResNet model is more suitable for repairing datasets where the ratio of missing or abnormal data is less than 30%;the repaired results based on the ResNet model can fit better to the original curve of the continuous monitoring data,which shows a good agreement with the monitored trend.
关键词
风电机组/深度学习/健康监测/数据修复Key words
wind turbines/deep learning/health monitoring/data restoration引用本文复制引用
基金项目
国家自然科学基金(41272281)
山东省自然科学基金(ZR2019MEE021)
出版年
2024
NETL
NSTL
万方数据