In order to enhance the ability of neural network to diagnose transformer faults and avoid local optimal and premature generation of bee colony algorithm, an improved adaptive search strategy bee colony optimization algorithm is proposed. The method adjusts the global and local search ability of the bee colony algorithm by adaptively adjusting the population update step size to avoid the local optimal condition, and introduces the Levy mutation factor to improve the performance of the local search. The improved bee colony algorithm is used to optimize the BP neural network weights and thresholds, and the iterative algorithm is repeated until the target accuracy requirement is reached. The method is tested based on transformer data. Simulation results show that the improved method has faster convergence speed and higher accuracy of discriminating faults. This work is supported byScience and Technology Research Project of Henan Province (No. 172102210274) and Key Scientific Research Project of Colleges and Universities in Henan Province (No. 18A470013).