Abstract: Transformer health status evaluation is an important prerequisite to improving power supply reliability and is also one of the basic tasks in achieving lean operation and maintenance control of equipment. The prerequisite of the existing method to achieve accurate transformer health evaluation is to obtain a complete and effective set of characteristic parameters. However, the irregularity of the actual data significantly reduces the accuracy of the existing health evaluation methods due to problems such as the unfixed sampling interval of the characteristic parameters, the error in acquiring some parameters, or the loss of transmission. For this reason, we explore the application of the Raindrop learning algorithm with high performance to deal with incomplete and irregular data sets for the first time in this field and introduce formal domain knowledge rules in the process of model training to enhance the effective information of sample space and guide the learner to optimize training. Finally, health evaluation under multiple scenarios is carried out based on the actual obtained 110kV transformer data. The results show that the proposed method is superior to the baseline learning method in all scenarios and can better adapt to the health evaluation needs of actual power equipment.