Accurate identification of hysteresis parameters in the Jiles-Atherton (J-A) model of current transformers is crucial for power system measurement and protection. However, in practice, problems such as measurement noise and insufficient data acquisition often degrade parameter identification accuracy. To address these challenges, a J-A model hysteresis parameter identification strategy for current transformers based on a Bayesian neural network (BNN) and the H5N1 optimization algorithm is proposed. BNN is used for data preprocessing, including denoising and prediction, thereby improving data quality. The H5N1 optimization algorithm is used for identifying hysteresis parameters of the J-A model. Meanwhile, multiple metaheuristic algorithms are selected for comparative validation. Simulation results show that the combination of BNN based data preprocessing and H5N1 optimization algorithm can significantly improve the accuracy and stability of hysteresis parameter identification compared to approaches without data preprocessing, providing a more efficient and accurate method for parameter identification of current transformer J-A model. For example, under denoising data, the identification accuracy increases by 22.90%, with an error of 1.2386; under predicted data conditions, the accuracy is improved by 89.33%, with an error of 0.7267.