Data-driven modeling and supervisory control system optimization for plug-in hybrid electric vehicles

Overview

• This paper focuses on developing a learning-based intelligent energy management system for plug-in hybrid electric vehicles (PHEVs) to achieve efficient energy utilization.
• The main challenge addressed is the lack of system reliability in real-world applications, which has prevented widespread acceptance by original equipment manufacturers (OEMs).
• The paper proposes a control framework that combines horizon-extended reinforcement learning (RL)-based energy management with the equivalent consumption minimization strategy (ECMS) to enhance practical applicability.
• The framework also improves the flawed method of equivalent factor evaluation found in existing research.
• Comprehensive simulation and hardware-in-the-loop validation are carried out to demonstrate the advantages of the proposed control framework in fuel economy over adaptive-ECMS and rule-based strategies.

Plain English Explanation

Plug-in hybrid electric vehicles (PHEVs) can help improve energy efficiency, but their energy management systems have faced reliability issues that have prevented widespread adoption by car manufacturers. This paper presents a new control system that uses a combination of reinforcement learning and a strategy called equivalent consumption minimization (ECMS) to better manage the energy use in PHEVs.

The key innovation is that this new control system not only achieves similar optimization as previous approaches, but it also significantly improves the system's ability to handle disturbances and unpredictable conditions. This makes the energy management system more reliable and practical for real-world use by car companies.

The paper also fixes a problem in how previous research evaluated the ECMS equivalent factor, which is an important part of the energy management strategy. Through simulations and testing on actual vehicle hardware, the authors demonstrate that their proposed control framework outperforms other common energy management approaches in terms of fuel economy.

Overall, this research provides an effective control framework for using reinforcement learning to manage energy in PHEVs, addressing key practical challenges that have held back the widespread adoption of this energy-efficient vehicle technology.

Technical Explanation

The paper begins by establishing a high-fidelity PHEV simulation model based on both physical and data-driven approaches. This provides a robust training environment for evaluating energy management strategies.

The researchers then propose a novel control framework that combines horizon-extended reinforcement learning (RL)-based energy management with the equivalent consumption minimization strategy (ECMS). The RL component learns an optimal energy management policy, while ECMS is used to enhance the practical applicability of the control system.

Crucially, the paper also improves upon the flawed method of equivalent factor evaluation found in prior ECMS-based research. The equivalent factor is a key parameter that impacts the energy management strategy, and the authors show that basing it solely on the instantaneous driving cycle and powertrain states is problematic.

To validate their approach, the researchers conduct comprehensive simulations as well as hardware-in-the-loop testing. The results demonstrate that their proposed control framework outperforms both adaptive-ECMS and rule-based energy management strategies in terms of fuel economy.

Compared to conventional RL architectures that directly control powertrain components, the authors' method achieves similar optimality while significantly enhancing the disturbance resistance of the energy management system. This makes it a more effective control framework for real-world PHEV applications from the perspective of OEMs.

Critical Analysis

The paper provides a well-designed control framework that addresses key practical challenges in applying reinforcement learning to PHEV energy management. The authors' innovations around the ECMS equivalent factor evaluation and the combination of RL with ECMS are particularly noteworthy.

However, the paper does not delve into the potential limitations of their approach. For example, the training and validation were conducted in simulation and hardware-in-the-loop environments, so there may be additional real-world factors that could impact the performance and reliability of the control system.

Additionally, the paper does not discuss the computational complexity and resource requirements of the proposed framework, which could be an important consideration for OEMs implementing the technology on production vehicles. Comparing the centralized and decentralized approaches to PHEV energy management could also provide useful insights.

Overall, the research represents a valuable contribution to the field of learning-based energy management for electric vehicles. However, further investigation of the practical limitations and potential trade-offs would help provide a more comprehensive understanding of the framework's real-world applicability.

Conclusion

This paper presents an innovative control framework that combines reinforcement learning-based energy management with the equivalent consumption minimization strategy (ECMS) to address the reliability challenges of learning-based intelligent energy systems for plug-in hybrid electric vehicles (PHEVs).

The key contributions include improving the flawed equivalent factor evaluation method in ECMS and demonstrating that the proposed framework can outperform adaptive-ECMS and rule-based strategies in terms of fuel economy through comprehensive simulations and hardware-in-the-loop testing.

By enhancing the disturbance resistance of the energy management system, this research provides an effective control framework for applying reinforcement learning to real-world PHEV applications, which is crucial for widespread adoption by original equipment manufacturers. Further investigation of practical limitations and trade-offs would help solidify the framework's viability for large-scale deployment.