Prediction for Future Yaw Rate Values of Vehicles Using Long Short-Term Memory Network.

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Control Systems

Background:

• Accurate sensor signal processing is vital for safety-critical systems in electric mobility and autonomous vehicles.
• Vehicle yaw rate is a key descriptor of vehicle dynamics, essential for intervention strategies.
• Predicting yaw rate aids in enhancing vehicle stability and control.

Purpose of the Study:

• To propose a Long Short-Term Memory (LSTM) network model for predicting future vehicle yaw rate values.
• To evaluate the model's accuracy using experimental data from diverse driving scenarios.
• To demonstrate the model's capability in real-time yaw rate prediction for enhanced vehicle safety.

Main Methods:

• Development of a neural network model utilizing Long Short-Term Memory (LSTM) architecture.
• Training, validation, and testing of the LSTM model with experimental sensor data from three distinct driving scenarios.
• Utilizing historical vehicle sensor signals (last 0.3 seconds) to predict future yaw rate (0.2 seconds ahead).

Main Results:

• The proposed LSTM model achieved high accuracy in predicting future yaw rate values.
• R2 values for the model ranged from 0.8938 to 0.9719 across different driving scenarios.
• In a mixed driving scenario, the model demonstrated a strong performance with an R2 value of 0.9624.

Conclusions:

• The LSTM-based neural network model effectively predicts vehicle yaw rate with high accuracy.
• The model's ability to forecast yaw rate using past sensor data is valuable for advanced driver-assistance systems (ADAS) and autonomous driving.
• This predictive capability contributes to improved safety and control strategies in electric and autonomous vehicles.