• Journal of Electronic Science and Technology
  • Vol. 22, Issue 4, 100285 (2024)
Yi-Feng Li1, Zhi-Ang Hu1, Jia-Wei Gao1, Yi-Sheng Zhang1..., Peng-Fei Li2 and Hai-Zhou Du2,*|Show fewer author(s)
Author Affiliations
  • 1Shanghai Electric Power Energy Technology Co., Ltd., Shanghai, 200233, China
  • 2School of Computer Science and Technology, Shanghai University of Electric Power, Shanghai, 201306, China
  • show less
    DOI: 10.1016/j.jnlest.2024.100285 Cite this Article
    Yi-Feng Li, Zhi-Ang Hu, Jia-Wei Gao, Yi-Sheng Zhang, Peng-Fei Li, Hai-Zhou Du. Efficient anomaly detection method for offshore wind turbines[J]. Journal of Electronic Science and Technology, 2024, 22(4): 100285 Copy Citation Text show less
    Two examples on the real-world offshore wind turbine operational dataset: (a) contrastive of L1L2 line voltage and (b) contrastive of wind speed (mechanical).
    Fig. 1. Two examples on the real-world offshore wind turbine operational dataset: (a) contrastive of L1L2 line voltage and (b) contrastive of wind speed (mechanical).
    Architecture of Hawkeye.
    Fig. 2. Architecture of Hawkeye.
    Difference between DSW embedding and conventional embedding approaches. DSW uses episodes of different features to predict different episodes of data with different features. While the conventional embedding approach uses all the features in the same time episode for embedding and does not take into account the correlation between the multivariate variables.
    Fig. 3. Difference between DSW embedding and conventional embedding approaches. DSW uses episodes of different features to predict different episodes of data with different features. While the conventional embedding approach uses all the features in the same time episode for embedding and does not take into account the correlation between the multivariate variables.
    Two-stage attention mechanism.
    Fig. 4. Two-stage attention mechanism.
    Comparison of (a) multi-headed self-attention mechanism and (b) our proposed router approach.
    Fig. 5. Comparison of (a) multi-headed self-attention mechanism and (b) our proposed router approach.
    Comparison of parameter file sizes.
    Fig. 6. Comparison of parameter file sizes.
    Anomaly detection results shown on two features: (a) L1L2 line voltage and (b) wind speed (mechanical).
    Fig. 7. Anomaly detection results shown on two features: (a) L1L2 line voltage and (b) wind speed (mechanical).
    DatasetDimensionTrainTestContamination (%)
    PSM27702721756830
    KDD99412964139880420
    OWTD6160359172797
    Table 1. Datasets description.
    Algorithm 1: Hawkeye process
    Data: Data ${\mathbf{x}}$ (n objects × D attributes)
    Result: Anomaly labels
    1: Function Predict(${\mathbf{x}}$)
    2:  Initialization
    3:  FillMissingData(${\mathbf{x}}$)
    4:  Standardization(${\mathbf{x}}$)
    5:  PredictionModule(x) → Predicteddata
    6:  residual ← $|{\mathbf{x}} - {{\hat {\bf{x}}}}|$
    7:  residual ← Standardization(residual)
    8:  return residual
    9: Function Automatic labeling (residual data, max_iter, tol)
    10:  Initialize cluster centers C with 3 randomly selected points from residual
    11:  for iter ← 1 to max_iter do
    12:   for each point residuali ∈ residual do
    13:    Compute distance di,k from residuali to each cluster center
    14:    Assign residuali to the nearest cluster center, update labelsi
    15:   end
    16:   for each cluster j∈{1, 2, ···, k} do
    17:    Update cluster center ${C_j}$ as the mean of points within the cluster
    18:   end
    19:   if change in cluster centers is less than tol then
    20:    break
    21:   end
    22: end
    23: return Anomaly labels
    Table 1. [in Chinese]
    MethodKDD99PSM
    PR${F_1}$PR${F_1}$
    LoF0.470.230.260.580.540.56
    OCSVM0.640.750.690.670.590.61
    Isolation Forest0.490.300.370.720.650.66
    USAD0.690.720.660.730.620.62
    Ours0.890.770.790.770.770.77
    Table 2. Performance on labeled public datasets.
    OursUSADIsolation ForestOCSVM
    Amount818280334568639
    Ratio (%)4.710.116.020.0
    Table 3. Anomaly detection results on an offshore wind turbine dataset.
    Yi-Feng Li, Zhi-Ang Hu, Jia-Wei Gao, Yi-Sheng Zhang, Peng-Fei Li, Hai-Zhou Du. Efficient anomaly detection method for offshore wind turbines[J]. Journal of Electronic Science and Technology, 2024, 22(4): 100285
    Download Citation