Current research lacks an optimized control level prediction method to ensure optimal motor power output during train operation. This paper, grounded in the operation of asynchronous motors in high-speed trains, designs a dynamic level control model integrated into the train's traction/braking unit closed-loop control system. Leveraging system identification, the level prediction model integrates linear models within the complete information space with unknown nonlinear dynamic models. To mitigate the impact of missing data on model accuracy, this paper introduces an enhanced bidirectional Long Short-Term Memory (Bi-LSTM) neural network model to interpolate missing data and employs a combination of Particle Swarm Optimization (PSO) and Improved Grey Wolf Optimization (IGWO) algorithms for selecting hyperparameters related to the neural network. To enhance the accuracy of predicting unknown nonlinear systems, this paper proposes a Bi-LSTM/LSTM model switching prediction algorithm based on autocorrelation coefficients. Finally, the effectiveness of the prediction method is validated using actual data from the CRH380B train.