Detection method of wind speed anomaly fluctuation based on SSA−LSTM

Abstract

Aiming at the problem of high leakage rate and false alarm rate of traditional statistical methods for abnormal fluctuation in sensor monitoring data caused by dampers opening and closing, a SSA-LSTM wind speed abnormal fluctuation detection method based on the combination of Singular Spectrum Analysis (SSA) and Long and Short-Term Memory Neural Network (LSTM) was proposed by mining the data features in the time-series data in the wind speed sensors. Firstly, SSA was used to pre-process the wind speed sensor monitoring data, and the wind speed data was decomposed into trend component, periodic component and noise component. The data noise generated by turbulent pulsation was removed via reorganizing the trend component and noise component. The LSTM parameters was then optimized, and the optimized LSTM model was used to predict the pre-processed data and obtain the reconstructed wind speed. Finally, the anomaly fraction of the monitored wind speed and reconstructed wind speed was calculated by using the logarithmic probability density function. Anomaly detection for monitoring wind speed was performed by calculating the threshold set value of training set data samples. The experimental results shown that, the removing effect for the data noise generated by turbulence pulsation via SSA was better. Removing the noise component without affecting the data fluctuation was helpful in improving the wind speed reconstruction effect and the anomaly detection accuracy. LSTM can correctly reconstruct the small amplitude wave due to turbulence pulsation without anomalous fluctuation and fits well with the actual data. The reconstruction of abnormal fluctuation segment based on historical fluctuation trend when there was abnormal fluctuation can effectively improve the accuracy of anomaly detection. Through comparative analysis, the reconstruction effect of proposed method in this paper was better than ARIMA, BP and CNN models, with an anomaly detection accuracy of 99.2% and an F1-Score of 0.97, which verified the reliability of the proposed method. The method proposed in the paper has important application value in detecting the abnormal fluctuation of wind speed caused by the opening and closing of dampers.

Keywords

• abnormal fluctutions
• dampers opening and closing
• anomaly detection
• ssa
• lstm
• time-series