Abstract: By developing the traditional gray non-equidistance GM(1,1) model, a new non-equidistance GM(1,1) model was proposed. In the data generation processing, the original data for modeling and forecasting was not directly used, but smooth processing was done by taking the logarithm transformation of the original data, and then the background value was weighed. The weight of the background weight was the relative distance, which was not the traditional absolute distance. This method can effectively avoid changing the nature of the original data. The example showed that the new model has better modeling accuracy and stronger adaptability than the Arrhenius model and the traditional gray non-equidistance GM(1,1) model for non-equidistance accelerated stress test life prediction.