A Hybrid Framework for Dynamic Optimization of Racing Car Braking System Based on BiLSTM and GBDT

Aiming at the problems of insufficient dynamic adaptability due to racing brake force allocation, this paper proposes a hybrid framework that integrates physical adjustment and adaptive learning. Based on the collected sensor data combined with wavelet threshold denoising process, the signal-to-noise ratio reaches 64.13 dB, and the pressure decay rate is predicted using a five-fold cross-validated optimized Gradient Boosting Decision Tree (GBDT) model with an RMSE of up to 0.1592, and the proportional valve gears are classified via a Bidirectional Long Short-Term Memory (BiLSTM) to achieve a macro-averaged F1-score of 88 %, and a macroaveraged precision of 90 %. The experiments show that the framework validates the potential of fusing machine learning with mechanical architectures and provides new ideas for dynamic braking optimization.