Temperature Monitoring of Heavy Wind Power Unit Inverters Based on FD-AT-LSTM

In order to monitor the condition of wind turbine inverters, a condition monitoring method for large wind turbine inverters based on finite differential regression vector and long short-term memory neural network with attention mechanism was proposed. The minimum mean square error variable step adaptive filtering method was used to perform adaptive filtering on the input and output to eliminate the random noise of the data. The Akachi information criterion was used to determine the dynamic delay order of input and output, and the finite differential regression vector was constructed and the inverter condition monitoring model was established with the finite differential regression vector as the input. And then the monitoring index was calculated according to the model residual and the monitoring index threshold was determined by kernel density estimation. Results show that compared with other machine learning methods, the proposed finite difference neural network model has great improvements in the model evaluation index.This method can be applied to the frequency converter monitoring and early warning in large wind turbines, and has certain industrial application prospects.