Long Short-Term Memory-Model Predictive Control Speed Prediction-Based Double Deep Q-Network Energy Management for...

Area of Science:

• Hybrid Vehicle Technology
• Artificial Intelligence in Automotive Engineering
• Sustainable Transportation Systems

Background:

• Improving hybrid vehicle fuel economy and emission performance is crucial for current environmental and energy challenges.
• Existing energy management strategies require scientific and reasonable optimization for enhanced efficiency.

Purpose of the Study:

• To propose an advanced energy management model for hybrid vehicles.
• To enhance fuel economy and emission performance through intelligent speed prediction and control.

Main Methods:

• Utilizing Long Short-Term Memory (LSTM) neural networks for speed prediction.
• Optimizing LSTM parameters with the Double Deep Q-Network (DDQN) algorithm.
• Implementing a Model Predictive Control (MPC) framework integrated with LSTM speed prediction.
• Employing a fuzzy logic system for driving mode recognition and classification.

Main Results:

• The LSTM-MPC-DDQN strategy achieved optimal fuel consumption at a 5-second prediction horizon, saving 2.034 L compared to a rule-based strategy.
• Demonstrated low Root Mean Square Error (RMSE) across various prediction horizons.
• Reduced fuel consumption by 0.2729 L compared to the rule-based approach under the UDDS driving cycle.
• Showed minimal fuel consumption difference (0.0749 L) compared to the dynamic programming (DP) strategy.

Conclusions:

• The proposed LSTM-MPC-DDQN energy management strategy offers a significant improvement in hybrid vehicle fuel economy and emission performance.
• Speed prediction using LSTM, optimized by DDQN and integrated with MPC, is effective for real-time energy management.
• The fuzzy logic-based driving mode recognition system enhances the adaptability and effectiveness of the control strategy across different driving conditions.