| 摘要: |
| 目的 风电机组带来清洁能源的同时,也面临着一系列技术挑战,其中最为突出的是机组的稳定性和可靠
性,针对风电机组数据的异常检测问题,提出一种科学且有效的改进方法———基于残差的孤立森林。 方法 鉴于传
统孤立森林方法在正常运行时波动的干扰和参数敏感性方面存在的问题,提出基于残差的孤立森林方法进行异常
检测。 首先采用 RANSAC 回归模型预测风速与功率之间的关系,并计算实际功率与预测功率之间的残差;进一步,
将这些残差作为输入特征,在孤立森林模型中实施异常检测。 此方法着重关注模型难以解释的部分,减少了正常
运行状态下的自然波动对异常检测的干扰,增强了对潜在异常点的精确识别。 结果 通过对异常值进行可视化验证
结果表明:改进方法在识别功率曲线上的局部异常方面表现更加突出。 结论 这种基于残差的孤立森林方法能够在
多数操作条件下更精确地识别出异常,为风电场的运维管理提供强有力的数据支持;同时也为其他领域的异常检
测提供了新的思路和参考,具有广泛的应用前景。 |
| 关键词: 孤立森林算法 RANSAC 回归 异常检测 风电机组 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Wind Turbine Data Anomaly Detection Based on Improved Isolated Forest Method |
|
ZHONG Qiuhui LI Yong
|
|
School of Statistics Chengdu University of Information Technology Chengdu 610225 China
|
| Abstract: |
| Objective Wind turbines bring clean energy but also face a series of technical challenges. Among them the
most prominent issues are the stability and reliability of the turbines. Addressing the issue of anomaly detection in wind
turbine data a scientific and effective improvement method is proposed—residual-based isolation forest. Methods Given
the disturbances from normal operational fluctuations and sensitivity to parameters in traditional isolation forest methods a
residual-based isolation forest approach is proposed for anomaly detection. Initially a RANSAC regression model predicts
the relationship between wind speed and power calculating the residuals between actual and predicted power. These
residuals are then used as input features in the isolation forest model for anomaly detection. This method focuses on parts
of the model that are difficult to interpret reducing interference from natural fluctuations during normal operation and
enhancing precise identification of potential anomaly points. Results Visualizations of anomaly detection results
demonstrate that the improved method excels in identifying local anomalies in power curves. Conclusion The residualbased isolation forest method can more accurately identify anomalies under various operational conditions providing robustdata support for the maintenance management of wind farms. Additionally this method offers new perspectives and
references for anomaly detection in other fields with broad application prospects |
| Key words: isolation forest algorithm RANSAC regression anomaly detection wind turbine |
