Periodic Reporting for period 1 - ATEM (Accelerating transport electrification by machine learning)
Berichtszeitraum: 2021-08-01 bis 2023-07-31
Problem/Issue Being Addressed:
This project seeks to address several interconnected challenges within the realm of transportation and environmental sustainability. These challenges include heavy reliance on fossil fuels, environmental pollution, energy inefficiency, and underutilization of emerging technologies in the road transport sector. The advent of Electric Vehicles (EVs) and Autonomous Electric Vehicles (AEVs) brings forth a promising solution, but optimal usage is constrained by limitations in battery management, route optimization, and overall vehicle efficiency.
Importance for Society:
Tackling these challenges holds significant societal importance. Transitioning from fossil fuels to electric transport contributes to the mitigation of climate change, improves air quality, and promotes energy sustainability. Moreover, optimizing the use of EVs and AEVs through advanced AI technologies can enhance transportation efficiency, user experience, and overall traffic management. The use of AI can contribute to smarter, safer, and more energy-efficient transportation systems, thereby positively impacting the quality of life and sustainability on a global scale.
Overall Objectives:
The overall objectives of this project can be summarized as follows: To enhance the lifespan and energy efficiency of batteries in EVs and AEVs. This will be achieved through the advanced design of battery management systems (BMS) that optimally control operations within the physical limits of the battery system, thereby ensuring safety and efficiency. To build a personalized AEV recommendation model leveraging AI technologies. The model will recommend optimal routes to users with the lowest energy consumption, further reducing energy cost and improving the user experience. To improve the efficiency of AEVs and overall user experience by integrating the electrochemical battery model (using a data-driven method), AEV speed control (using deep reinforcement learning), and the AEV recommendation model in real-time. This unified approach will contribute to safer, faster, and more environmentally friendly transport. By successfully achieving these objectives, the project will make significant strides towards more sustainable and efficient transportation, harnessing the potential of AI technologies to revolutionize the field of transport electrification.
This project seeks to address several interconnected challenges within the realm of transportation and environmental sustainability. These challenges include heavy reliance on fossil fuels, environmental pollution, energy inefficiency, and underutilization of emerging technologies in the road transport sector. The advent of Electric Vehicles (EVs) and Autonomous Electric Vehicles (AEVs) brings forth a promising solution, but optimal usage is constrained by limitations in battery management, route optimization, and overall vehicle efficiency.
Importance for Society:
Tackling these challenges holds significant societal importance. Transitioning from fossil fuels to electric transport contributes to the mitigation of climate change, improves air quality, and promotes energy sustainability. Moreover, optimizing the use of EVs and AEVs through advanced AI technologies can enhance transportation efficiency, user experience, and overall traffic management. The use of AI can contribute to smarter, safer, and more energy-efficient transportation systems, thereby positively impacting the quality of life and sustainability on a global scale.
Overall Objectives:
The overall objectives of this project can be summarized as follows: To enhance the lifespan and energy efficiency of batteries in EVs and AEVs. This will be achieved through the advanced design of battery management systems (BMS) that optimally control operations within the physical limits of the battery system, thereby ensuring safety and efficiency. To build a personalized AEV recommendation model leveraging AI technologies. The model will recommend optimal routes to users with the lowest energy consumption, further reducing energy cost and improving the user experience. To improve the efficiency of AEVs and overall user experience by integrating the electrochemical battery model (using a data-driven method), AEV speed control (using deep reinforcement learning), and the AEV recommendation model in real-time. This unified approach will contribute to safer, faster, and more environmentally friendly transport. By successfully achieving these objectives, the project will make significant strides towards more sustainable and efficient transportation, harnessing the potential of AI technologies to revolutionize the field of transport electrification.
Work Performed:
In WP1, we developed a novel method termed 'Model Controlled Prediction' that can improve the efficiency and accuracy of electrochemical models [1]. In WP2, we crafted a method known as 'Deep Adaptive Control' which entailed the design of DRL-based adaptive vehicle trajectory control algorithms for different risk levels [7]. We further studied the enhancement of state representation in Multi-Agent RL for Platoon-Following Models, contributing significant insights to this field [8]. In WP3, we developed a personalized recommendation system. This system leveraged users' historical records to predict travel destinations, provided optimal route recommendations to users based on energy consumption and personal preferences [4].
Exploitation and Dissemination of Results:
I hosted a special session as the co-chair at the IEEE ITSC (https://2023.ieee-itsc.org/special-sessions/). This session served as a platform to showcase our project's accomplishments to the broader community. Additionally, I hold the position of Associate Editor for IEEE TIV and JICV, and serve as a Young Editor for the Innovation (Cell Press’ flagship general journal, Impact factor = 32.1) and IEEE/CAA JAS. I have been actively promoting our project's outcomes in these journals' activities, including paper submissions, social media dissemination via the journals' Twitter accounts, and during seminars.
[1] Li, S., Liu, Y.*, & Qu, X. (2022). Model controlled prediction: A reciprocal alternative of model predictive control. IEEE/CAA JAS, 9(6), 1107-1110.
[2] Liu, Y., et al. (2022). How machine learning informs ride-hailing services: A survey. COMMTR, 2, 100075.
[3] Liu, Y., et al. (2021). DeepTSP: Deep traffic state prediction model based on large-scale empirical data. COMMTR, 1, 100012.
[4] Wu, F., Lyu, C., & Liu, Y.* (2022). A personalized recommendation system for multi-modal transportation systems. Multimodal Transportation, 1(2), 100016.
[5] Jia, R., Chamoun, R., Wallenbring, A., Advand, M., Yu, S., Liu, Y., & Gao, K. (2023). A spatio-temporal deep learning model for short-term bike-sharing demand prediction. ERA, 31(2), 1031-1047.
[6] Lin, P., He, C., Zhong, L., Pei, M., Zhou, C., & Liu, Y. (2023). Bus timetable optimization model in response to the diverse and uncertain requirements of passengers for travel comfort. ERA, 31(4), 2315-2336.
Under review
[7] He, Y., Liu, Y.*, & Qu, X., Deep adaptive control: Deep reinforcement learning-based adaptive vehicle trajectory control algorithms for different risk levels. Submitted to IEEE TIV.
[8] Lin, H., Liu, Y.*, et al., Enhancing State Representation in Multi-Agent Reinforcement Learning for Platoon-Following Models. Submitted to IEEE TIV.
In WP1, we developed a novel method termed 'Model Controlled Prediction' that can improve the efficiency and accuracy of electrochemical models [1]. In WP2, we crafted a method known as 'Deep Adaptive Control' which entailed the design of DRL-based adaptive vehicle trajectory control algorithms for different risk levels [7]. We further studied the enhancement of state representation in Multi-Agent RL for Platoon-Following Models, contributing significant insights to this field [8]. In WP3, we developed a personalized recommendation system. This system leveraged users' historical records to predict travel destinations, provided optimal route recommendations to users based on energy consumption and personal preferences [4].
Exploitation and Dissemination of Results:
I hosted a special session as the co-chair at the IEEE ITSC (https://2023.ieee-itsc.org/special-sessions/). This session served as a platform to showcase our project's accomplishments to the broader community. Additionally, I hold the position of Associate Editor for IEEE TIV and JICV, and serve as a Young Editor for the Innovation (Cell Press’ flagship general journal, Impact factor = 32.1) and IEEE/CAA JAS. I have been actively promoting our project's outcomes in these journals' activities, including paper submissions, social media dissemination via the journals' Twitter accounts, and during seminars.
[1] Li, S., Liu, Y.*, & Qu, X. (2022). Model controlled prediction: A reciprocal alternative of model predictive control. IEEE/CAA JAS, 9(6), 1107-1110.
[2] Liu, Y., et al. (2022). How machine learning informs ride-hailing services: A survey. COMMTR, 2, 100075.
[3] Liu, Y., et al. (2021). DeepTSP: Deep traffic state prediction model based on large-scale empirical data. COMMTR, 1, 100012.
[4] Wu, F., Lyu, C., & Liu, Y.* (2022). A personalized recommendation system for multi-modal transportation systems. Multimodal Transportation, 1(2), 100016.
[5] Jia, R., Chamoun, R., Wallenbring, A., Advand, M., Yu, S., Liu, Y., & Gao, K. (2023). A spatio-temporal deep learning model for short-term bike-sharing demand prediction. ERA, 31(2), 1031-1047.
[6] Lin, P., He, C., Zhong, L., Pei, M., Zhou, C., & Liu, Y. (2023). Bus timetable optimization model in response to the diverse and uncertain requirements of passengers for travel comfort. ERA, 31(4), 2315-2336.
Under review
[7] He, Y., Liu, Y.*, & Qu, X., Deep adaptive control: Deep reinforcement learning-based adaptive vehicle trajectory control algorithms for different risk levels. Submitted to IEEE TIV.
[8] Lin, H., Liu, Y.*, et al., Enhancing State Representation in Multi-Agent Reinforcement Learning for Platoon-Following Models. Submitted to IEEE TIV.
Progress Beyond the State of the Art:
This project represents a significant advancement beyond the current state of the art in the domain of transport electrification and autonomous vehicle management. Our data-driven approach and the implementation of machine learning for route optimization provide new, innovative solutions to the challenges facing the EV industry today. The machine learning algorithms developed have demonstrated practical utility and superiority over existing methods in prestigious AI challenges, securing two world runner-up positions in the KDD CUP 2022.
Expected Results and Potential Impacts:
By the end of the project, we anticipate even more breakthroughs that could reshape the way we view and interact with autonomous electric vehicles. The long-term expectation is the successful integration of our developed models into industrial partners' smart vehicle products. Such implementation can dramatically improve the safety and economy of contemporary smart connected vehicles, as reflected in our ongoing collaboration with the Swedish Energy Agency (https://research.chalmers.se/en/project/11003). We also intend to continue submitting research outcomes to top-tier scientific journals, ensuring that our findings are disseminated widely within the academic community and beyond.
The socio-economic impact of this project is multi-faceted. On one hand, our work supports the broader transition towards sustainable and intelligent transportation, contributing to climate action efforts and energy conservation. On the other hand, improvements in battery life and energy utilization efficiency have the potential to reduce costs for both manufacturers and consumers, making EVs a more affordable and attractive option. In terms of wider societal implications, the project has generated significant interest from renowned research institutions such as the University of Wisconsin-Madison in the United States. We have also co-applied for a staff exchange within the applied AI program in Vinnova (https://www.vinnova.se/en/p/applied-ai-for-connected-and-autonomous-transportation-systems/). This interest reflects the project's contribution to developing cutting-edge knowledge and technologies that can be disseminated and implemented globally, ultimately leading to safer, more efficient, and more sustainable transportation systems worldwide.
This project represents a significant advancement beyond the current state of the art in the domain of transport electrification and autonomous vehicle management. Our data-driven approach and the implementation of machine learning for route optimization provide new, innovative solutions to the challenges facing the EV industry today. The machine learning algorithms developed have demonstrated practical utility and superiority over existing methods in prestigious AI challenges, securing two world runner-up positions in the KDD CUP 2022.
Expected Results and Potential Impacts:
By the end of the project, we anticipate even more breakthroughs that could reshape the way we view and interact with autonomous electric vehicles. The long-term expectation is the successful integration of our developed models into industrial partners' smart vehicle products. Such implementation can dramatically improve the safety and economy of contemporary smart connected vehicles, as reflected in our ongoing collaboration with the Swedish Energy Agency (https://research.chalmers.se/en/project/11003). We also intend to continue submitting research outcomes to top-tier scientific journals, ensuring that our findings are disseminated widely within the academic community and beyond.
The socio-economic impact of this project is multi-faceted. On one hand, our work supports the broader transition towards sustainable and intelligent transportation, contributing to climate action efforts and energy conservation. On the other hand, improvements in battery life and energy utilization efficiency have the potential to reduce costs for both manufacturers and consumers, making EVs a more affordable and attractive option. In terms of wider societal implications, the project has generated significant interest from renowned research institutions such as the University of Wisconsin-Madison in the United States. We have also co-applied for a staff exchange within the applied AI program in Vinnova (https://www.vinnova.se/en/p/applied-ai-for-connected-and-autonomous-transportation-systems/). This interest reflects the project's contribution to developing cutting-edge knowledge and technologies that can be disseminated and implemented globally, ultimately leading to safer, more efficient, and more sustainable transportation systems worldwide.