IGBT Lifetime Prediction Method And Its Application Based On Variable-scale Gaussian Process Model

As an important driving area of sustainable development strategy,the development of new energy industry has been widely valued and promoted.IGBT is the core component of new energy equipment such as new energy vehicles,wind power,photovoltaics,etc.Accurate prediction of IGBT power module remaining useful life(RUL)is conducive to timely and effective maintenance and replacement of IGBT power module,ensuring the safe and stable operation of the equipment as a whole and promoting the sustainable development of society.Before and after the failure of the IGBT module,the collector-emitter saturation voltage drop(Vce)jumps significantly.This feature can accurately determine the RUL of IGBT.However,Vce time series measurement value have obvious random non-stationary features,and many current methods do not work well when predicting time series with such features.Therefore,in this paper,the method of predicting IGBT RUL based on Vce time series value is studied.A variable-scale Gaussian process(VSGP)model prediction method based on optimization algorithm to obtain optimal parameters is proposed.Firstly,the basic structure of IGBT and several common IGBT structure types are introduced.The working principle of IGBT and the aging failure mechanism of IGBT module are explained.The reason why Vce is selected as the failure characteristic of IGBT module is analyzed in detail.The failure standard of IGBT module is determined.Secondly,the accelerated aging experiment of the PT-IGBT is introduced.The Vce change percentage time series datum of PT-IGBT1,PT-IGBT2 and PT-IGBT3 are obtained and analyzed.The obtained time series datum is denoised using the median filtering method.Thirdly,the false nearest neighbor method and the mutual information method are used to determine the optimal embedding dimension and the optimal time delay for phase space reconstruction,respectively.The phase space reconstruction of the Vce change percentage time series of PT-IGBT1,PT-IGBT2 and PT-IGBT3 is carried out.The phase trajectories recursive graphs of three time series are drawn,and the recurrence quantification analysis to the time series is carried out.The VSGP model and the method for obtaining optimal parameters of VSGP model based on an optimization algorithm are introduced in detail.Finally,the optimal parameters of the VSGP model are obtained using the ant lion optimization algorithm and the starting data.The RUL of PT-IGBT2 and PT-IGBT3 is predicted using the VSGP model with optimal parameters,and the prediction effect is compared with the comparison models.The confidence intervals of the VSGP and GP prediction results are compared.The effect of sample sampling interval change on the prediction effect of VSGP model is tested.The threshold of RUL is set.The variable-scale function of the VSGP model makes it possible to handle randomness and non-stationarity in time series prediction well.The results show that VSGP model is better than the traditional prediction model in terms of prediction accuracy.It is a prediction model with high prediction accuracy and adaptability to training samples with small-scale and different sampling intervals,and can output the confidence interval of prediction results.