Abstract: This paper addresses the issue of fault diagnosis and life prediction of power transformers, specifically focusing on oil-encroached transformers. We propose a comprehensive approach combining a bi-directional long- and short-term memory network (Bi-LSTM) and a statistical method based on the Weibull distribution to perform state assessment and life prediction. Firstly, we systematically summarize data-driven equipment fault state assessment and life prediction methods from two dimensions: machine learning-based approaches and statistical data-based approaches. Secondly, we determine characteristic state transfer sequences based on the transformer operation mechanism and construct a deep neural network-based oil-intrusive transformer fault assessment model utilizing the Bi-LSTM architecture. Next, we fit the oil-intrusive transformer life using the Weibull distribution function, establish a two-parameter Weibull distribution transformer life model, and demonstrate oil-intrusive transformer life prediction by utilizing data from a large-scale oil-immersed power transformer owned by a power grid company during the period of 2007 to 2016. Finally, we conduct simulation experiments to validate the effectiveness of our proposed method.