An Improved Deep Neural Network Model of Intelligent Vehicle Dynamics via Linear Decreasing Weight Particle Swarm and...

Area of Science:

• Vehicle dynamics and control
• Artificial intelligence in engineering
• Computational modeling

Background:

• Accurate prediction of vehicle longitudinal-lateral responses is crucial for intelligent vehicle systems.
• Traditional deep neural network (DNN) training can be susceptible to local optima, limiting model performance.
• Optimization algorithms offer potential solutions for enhancing DNN training stability and accuracy.

Purpose of the Study:

• To develop an improved DNN modeling method for predicting vehicle longitudinal-lateral responses.
• To enhance DNN training by preventing weight and bias matrices from converging to local optima.
• To evaluate the real-time prediction accuracy of the proposed optimized DNN models.

Main Methods:

• Vehicle dynamic simulations were conducted using a semirecursive multibody model for data acquisition.
• Two optimization algorithms, linear decreasing weight particle swarm optimization (LDWPSO) and invasive weed optimization (IWO), were integrated with DNN models.
• The LDWPSO-DNN and IWO-DNN models were trained and tested using simulated vehicle data.

Main Results:

• The LDWPSO-DNN and IWO-DNN models demonstrated accurate prediction of vehicle longitudinal-lateral responses.
• The optimized DNN models successfully avoided local optima during training.
• Real-time prediction capabilities were achieved without compromising accuracy.

Conclusions:

• The proposed LDWPSO-DNN and IWO-DNN models provide accurate and robust predictions of vehicle responses.
• These optimized DNN approaches effectively overcome the local optimum problem in model training.
• The improved DNN models are suitable for real-time simulation and preview control applications in intelligent vehicles.