Energy Management Strategy Based on a Novel Speed Prediction Method.

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Accurate vehicle speed prediction is crucial for optimizing energy management strategies in vehicles.
• Existing methods may struggle with the complex, nonlinear dynamics of vehicle movement.

Purpose of the Study:

• To introduce VSNet, a novel deep learning architecture for precise vehicle speed prediction.
• To evaluate the performance of VSNet in predicting future vehicle speeds and its impact on energy management.

Main Methods:

• Developed VSNet by integrating Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM) networks in a unique serial-parallel structure.
• Utilized a 'fake image' input composed of 15 vehicle signals from the past 15 seconds to predict speed over the next 5 seconds.
• Incorporated CNNs with varying convolutional kernel sizes to capture complex, nonlinear relationships.

Main Results:

• VSNet demonstrated strong prediction accuracy, with Root Mean Square Error (RMSE) ranging from 0.519 to 2.681 and R-squared (R2) values between 0.929 and 0.997 for 5-second speed predictions.
• Simulations combining VSNet with Model Predictive Control (MPC) for energy management showed a minimal 4.74% increase in fuel consumption compared to dynamic programming (DP) and a 2.82% decrease compared to less accurate prediction methods.

Conclusions:

• The proposed VSNet architecture effectively predicts vehicle speed by leveraging a hybrid deep learning approach.
• VSNet integration into energy management strategies offers a promising balance between performance and efficiency, outperforming less accurate prediction techniques.