Five models are used to test the model fault on the validation data set. The validation data set has 5000 data. According to the status code of SCADA data, the oil temperature of the wind turbineis too high at 4421 points. The LSTM network model is used to verify the oil temperature prediction curve as shown in Figure 1, and the corresponding EWMA threshold control chart is shown in Figure 2.
According to figure 1 and Figure 2, it can be concluded that the residual of LSTM model between the real value and the predicted value is very small when there is no fault, and it has a good fitting ability to the gearbox oil temperature; when there is a fault, the residual increases rapidly, and reaches the threshold value calculated by the formula at the 3594 point and continues to rise; Compared with the data points recorded by the status code, this model is 821 points ahead of time. According to the data recorded once in 30 s, this model predicts the gearbox fault 6.84 hours ahead of time.
Five models are used to predict the fault data, and the threshold, the number of data points and the prediction time are calculated.
Compared with the other four models, LSTM model has the smallest threshold and can predict the gearbox fault first, which is conducive to taking measures earlier and avoiding more serious faults.