Aiming at the problems of low diagnostic accuracy and easy to fall into local optimization of sparrow search algorithm (SSA) for transformer fault diagnosis, a transformer fault diagnosis model is proposed based on the optimized by sine-cosine and Cauchy mutation sparrow search algorithm (SCSSA). Firstly, based on the dissolved gas analysis (DGA) method in oil, five feature quantities are used as inputs. Secondly, the sparrow algorithm is improved by using the positive cosine strategy and Cauchy variation strategy, and then the performance of SCSSA algorithm, SSA algorithm and grey wolf optimizer (GWO) are compared on four kinds of test functions for performance comparison and verified the superiority of SCSSA algorithm. Finally, SCSSA algorithm is used to optimize the parameters in the BiLSTM network, so as to improve the performance of BiLSTM network in transformer fault diagnosis. The experimental results show that the proposed SCSSA-BiLSTM fault diagnosis model has an integrated diagnostic accuracy of 95.1 %, which is 7.3 %, 12.2 %, 14.6 %, and 19.5 % higher than the SSA-BiLSTM, GWO-BiLSTM, BiLSTM, and LSTM models, respectively, and the SCSSA-BiLSTM model has better robustness.