The aging of every part of transformer will affect its service life. The description of its health will help us to predict the failure rate and residual life of transformer during operation. In this paper, using the idea of transformer fault bathtub curve, Weibull fitting is carried out to collect transformer fault rate, and the fault rate curve is obtained. Considering the operation environment and load factors of transformer, the residual life prediction model of health index is constructed by using the content of furfural. By optimizing the weight parameters of BP neural network through chaotic sequence, a transformer life prediction model with multi-parameter correlation is constructed, and cross-validation mechanism is introduced to improve the generalization ability of the network. By comparing the example training with the test, the method in this paper has higher prediction accuracy and can be applied to the life prediction of transformer accurately.