| 王坤,王绍强,熊星,李昂,邱绪迪,王斌.一种检测水电机组振动异常的动态区间覆盖方法[J].水利学报,2025,56(5):599-610 |
| 一种检测水电机组振动异常的动态区间覆盖方法 |
| A dynamic interval coverage method for detecting abnormal vibration of hydroelectric units |
| 投稿时间：2024-05-24 |
| DOI：10.13243/j.cnki.slxb.20240313 |
| 中文关键词: 水电机组 异常检测 区间预测 滑动窗口 ROC曲线 |
| 英文关键词: hydroelectric units abnormal detection interval prediction sliding window receiver operating characteristic curve |
| 基金项目:国家自然科学基金重点项目（52339006） |
|
| 摘要点击次数: 174 |
| 全文下载次数: 100 |
| 中文摘要: |
| 水电机组异常检测对于机组稳定运行至关重要。针对基于点预测的异常检测方法未能充分考虑数据的不稳定性以及模型的局限性，本文提出了一种检测水电机组振动异常的动态区间覆盖方法。首先，采用直接区间估计法，引入分位数损失作为损失函数训练模型实现区间预测，从而构建水电机组健康模型。进而，构建一种融合滑动窗口与区间覆盖率的异常检测指标。然后，使用受试者工作特征曲线进行模型评价和确定最优阈值，提出一种通过置信水平、滑动窗口宽度和步长来灵活调整检测力度的异常检测方法。为验证所提方法的有效性，本文采用水电机组模拟异常和真实异常数据进行仿真实验，结果表明该方法在以上数据集真阳性率、假阳性率分别达到1 ． 0、0． 01 5与0． 979、0． 053，能够准确识别水电机组的振动异常，为水电机组的异常检测提供了一种新的技术手段。 |
| 英文摘要: |
| Abnormal detection in hydropower units is crucial for their stable operation. Existing anomaly detection methods based on point predictions often fail to adequately address data instability and model limitations. This paper proposes a dynamic interval coverage method for detecting vibration anomalies in hydroelectric units. Initially，an interval prediction model is constructed using direct interval estimation and quantile loss as the loss function to train the model. A dynamic interval coverage rate is then developed by integrating a sliding window with interval coverage rate of the model evaluation index. Subsequently，the receiver operating characteristic curve is utilized for model evaluation and optimal threshold determination. This method allows for flexible adjustment of detection intensity through confidence level，sliding window width，and step size. To validate the effectiveness of the proposed method，simulation experiments are conducted using both simulated and real anomaly data of hydroelectric units. The experimental results indicate that this method achieves true positive rates and false positive rates of 1.0 and 0.01 5，respectively，on simulated dataset，and 0.979 and 0.053 on another，accurately identifying vibration anomalies in hydropower units. This approach provides a new technological means for anomaly detection in hydropower units. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |