Abnormal Detection And State Assessment Of Wind Turbine Main Transmission System Considering Uncertain Factors

The main transmission system is a critical subsystem responsible for converting and transmitting energy in wind turbines.However,with the increasing complexity structure of the turbine structures and harsh operating environments,failures in the main transmission system are becoming more frequent,leading to exorbitant operation and maintenance costs.As a result,accurately identifying the unit’s operating state and ensuring its stable operation is an urgent problem that requires immediate attention.While there are numerous methods for detecting abnormalities and evaluating state,the presence of uncertain factors in the operation service process can lead to false positives and omissions in state monitoring.Consequently,assessment results may not accurately reflect the actual status of the turbines.Therefore,comprehensively understanding state changes in the main transmission system and early detection of any abnormal trends play a crucial role in ensuring stable unit operation and minimizing losses in the wind field.With the support of the National Natural Science Foundation of China project "SCADA data-based Dynamic behavior and state assessment of main transmission system of wind turbine under multiple working condition ",this paper focuses on the issue of low matching degree between the detected and assessed state of wind turbines and their actual state,which is caused by various uncertain factors during the service process.Taking the main transmission system as the research object,an anomaly detection and state assessment method considering uncertain factors is proposed to improve the accuracy of anomaly detection and state assessment of the main transmission system.The effectiveness of the proposed method is verified through examples analysis and method comparison.The research presented in this paper has strengthened the ability of unit anomaly detection and state assessment to withstand the influence of uncertain factors,thereby improving the accuracy of state assessment and anomaly detection.As a result,the operational costs of wind turbines have been reduced,and the efficiency of wind farms has been enhanced.The specific research contents are as follows:(1)Considering the impact of uncertain factors on the reliability and accuracy of monitoring data in wind turbines,the research object of this paper analyzes the types,causes and effects of uncertain factors,and conducts research on how to reduce the influence of uncertain factors on information interference.Construct a deep autoencoder based on loss optimization(Loss-Deep Auto Encoder,L-DAE for short),by compressing the monitoring data into the smallest sufficient statistical variable and then reconstructing it,combined with data cleaning to reduce the interference information caused by uncertain factors,so as to improve the reliability of the monitoring data and the accuracy of reflecting the state of the unit.The experiments show that the L-DAE model proposed in this paper has a strong ability to resist the influence of uncertain factors,and can perform feature extraction and reconstruction on the original data well,which can be used for subsequent anomaly detection and status evaluation.This provides a solid data foundation for subsequent abnormal detection and state assessment.(2)A method for detecting abnormal states in wind turbine main transmission systems considering uncertain factors is proposed.On the basis of selecting appropriate input and output monitoring parameters,a deep autoencoder based on loss function optimization-bidirectional gated recurrent network(Loss-Deep Auto EncoderBidirectional Gated Recurrent Unit,L-DAE-Bi GRU)parameter prediction model is constructed.An abnormal detection threshold is established using the 3σ criterion.This threshold identifies anomalies by comparing data residuals under the detected state with the abnormal detection threshold.The experimental results and method comparisons demonstrate that the proposed method accurately identifies anomalies in the main transmission system of wind turbines.It also exhibits good anti-interference ability against uncertain factors.Compared to traditional abnormal detection methods,it has higher accuracy and less false alarms and can provide alarms 8 to 26 hours earlier than the alarm system of unit.(3)In order to gain a clearer understanding of the state changes of the wind turbine main transmission system,this research aims to explore how to reduce the uncertainty and ambiguity in the state assessment by proposing new method.The method uses the residual between the predicted and actual values of a deep autoencoder based on loss function optimization-bidirectional gated recurrent neural network model as the state assessment index.Additionally,a fuzzy comprehensive evaluation approach is employed to achieve both quantitative and qualitative analysis of the main transmission system’s operational state,thereby reducing the problem of unclear recognition of the system’s state due to uncertain factors.The case study demonstrates that the proposed state assessment method can accurately reflect the state changes of the main transmission system,identify abnormal deterioration trend of the unit in advance,and exhibit better matching performance compared to traditional state assessment methods.