Dual-Input and Multi-Channel Convolutional Neural Network Model for Vehicle Speed Prediction

Area of Science:

• * Intelligent Transportation Systems
• * Machine Learning for Vehicle Dynamics
• * Automotive Control Systems

Background:

• * Accurate vehicle speed prediction is crucial for intelligent vehicle control strategies.
• * The time-varying and nonlinear characteristics of vehicle speed pose significant prediction challenges.
• * Existing prediction methods struggle with the complexity of real-world driving conditions.

Purpose of the Study:

• * To propose a novel deep convolutional neural network (CNN) architecture for enhanced vehicle speed prediction.
• * To develop a dual-input CNN (DICNN) capable of integrating vehicle signals and driver intention.
• * To evaluate the performance of DICNN against traditional and single-input CNN models and assess its impact on fuel consumption.

Main Methods:

• * Development of a dual-input CNN (DICNN) architecture with four channels utilizing 10 vehicle signals and driver's intention.
• * Comparative analysis of DICNN against Markov chain combined with Monte Carlo (MCMC) simulation, Support Vector Machine (SVM), and Single-Input CNN (SICNN).
• * Integration of DICNN into Equivalent Fuel Consumption Minimization Strategies (ECMS) for energy management and simulation-based verification.

Main Results:

• * DICNN achieved superior prediction accuracy, evidenced by optimal Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Error (ME), and R-squared (R²) values compared to MCMC, SVM, and SICNN.
• * ECMS combined with DICNN demonstrated improved fuel efficiency, with only a 4.89% increase compared to dynamic programming and a 5.40% decrease compared to low-accuracy prediction methods.

Conclusions:

• * The proposed DICNN model offers a significant advancement in vehicle speed prediction accuracy by effectively extracting temporal information and driver intent.
• * DICNN-based ECMS provides an efficient energy management strategy, balancing performance and fuel economy.
• * This research contributes to the development of more sophisticated and efficient intelligent vehicle control systems.