Abstract：To address the issues of low accuracy and precision in transformer fault diagnosis，a transformer

fault diagnosis method based on an improved particle swarm optimization-long short-term memory（IPSO-LSTM）

network was proposed，integrating data feature selection and an improved particle swarm optimization algorithm.

Firstly，the raw dataset was preprocessed using the synthetic minority oversampling technique（SMOTE）to

increase the data volume. Secondly，the feature dimensions were expanded to 20 using the feature ratio method，and the random forest（RF）algorithm was employed to evaluate feature importance and perform feature selection，reducing the risk of overfitting. Subsequently，adaptive inertia weights were introduced to improve the PSO algorithm，which was then utilized to optimize the hyperparameters of the LSTM network. Finally，the featureselected data was input into the model for transformer fault diagnosis. Results demonstrate that the proposed diagnostic model achieves an accuracy of 91.6%. Compared with LSTM，HBA-LSTM，and PSO-LSTM diagnostic models，the accuracy improves by 10.12%，5.95%，and 3.57%，respectively，validating that the IPSO-LSTM diagnostic model provides superior diagnostic accuracy and holds practical significance in the field of transformer fault diagnosis.

      Key words：transformer fault diagnosis；feature selection；random forest（RF）；long short-term memory

（LSTM）network；particle swarm optimization（PSO）algorithm