The problem of residual life prediction of high-reliability device was studied.Considering the lack of life data and the difficulty in establishing a physical model,we used the inverse Gaussian regression model to build up the degradation process model of device in combination with the monotonous degradation data.Then we obtained the degradation model by using the method of parameter estimation for forecasting the residual life of device.The problem of data fusion may emerge when there are data of multiple sets in the same batch to estimate inverse Gaussian model parameter.We used the method based on Evidential Reasoning (ER) to fuse the multi-source data,and put forward the concept of attribute weights,in order to estimate the inverse Gaussian model parameters more accurately.Finally,simulation experiment,proved that the presented method can obtain more reliable parameter estimation results.