Tongji team develops multi-mode energy management strategy for FCVs based on neural network driving pattern recognition
Researchers at Tongji University (China), with colleagues at Zhengzhou Yutong Bus and HeFei University of Technology, have developed a multi-mode energy management strategy for electric vehicles with a fuel cell range extender based on driving pattern identification using a learning vector quantization (LVQ) neural network algorithm.
This multi-mode strategy can automatically switch to a genetic-algorithm-optimized strategy under specific driving conditions. The Simulation results on a dynamometer show that the proposed strategy can obtain better economic performance than the single-mode strategies under dynamic driving conditions. An open-access paper on their work is published in the Journal of Power Sources.
|Powertrain system architecture of FCEV. (a)multi-mode energy management strategy and (b) power train system topology. Song et al. Click to enlarge.|
With this type of fuel cell vehicle, the energy management strategy distributes power demand between the fuel cell system and a battery system; the addition of a battery pack makes quick start-up and energy recovery from regenerative braking possible.
Research done so far on energy management falls mainly into rule-based and optimization-based approaches, the researchers said. Rule-based systems cannot guarantee optimal power distribution due to a lack of road information, leading to a focus on optimization based approaches generally based on a cost function using linear programming, dynamic programming and genetic algorithms.
In such strategies, the system is optimized offline to minimize fuel consumption according to a determined driving cycle, which requires a huge amount of computation and previous information, and which, therefore, limits the use of such kinds of strategies in real-time controllers and in the environment. Consequently, the globally optimal solution is generally used as a benchmark for other strategies.
For the purpose of simplifying the complexity of calculations for real-time applications, several algorithm applications, such as stochastic dynamic programming, Pontryagin’s minimum principle (PMP), convex programming, or equivalent consumption minimization strategy (ECMS) are suggested as solutions and can usually approximate the globally optimal solution obtained using dynamic programming. Genetic algorithms are also often used for parameter optimization in energy management strategy to enhance economic performance, especially in rule-based strategies.
… It has been found, however, that these optimized parameters are often obtained under specific conditions, so they are not appropriate for the complex road conditions under which most vehicles travel. Therefore, under real driving conditions without predefined driving cycles, an adaptive multi-mode energy management strategy should be proposed to optimize vehicle performance.
… A neural network can correctly map the pattern from the feature space to the class space, and has a strong learning and adaptive ability, which is widely used in the field of pattern recognition. … In this research, the driving condition information extractor extracts 16 types of characteristic parameters and the driving environment recognizer includes a road type identifier, driving style identifier, driving trend recognizer, and driving mode recognizer. Among these parameters, 11 kinds of typical driving conditions were selected and constructed by the learning vector quantization (LVQ) neural network algorithm in the road type identifier. The latter three kinds of recognizers (environment, driving trend, and driving mode) were realized by the fuzzy controller.—Song et al.
The Tongji study focused on a a low-speed fuel cell electric vehicle (FCEV) in large venues. The power sources consist of a fuel cell range extender and a battery system for the sake of quick dynamic response and durability. The fuel cell system can drive the motor and simultaneously charge the power battery.
Ke Song, Feiqiang Li, Xiao Hu, Lin He, Wenxu Niu, Sihao Lu, Tong Zhang (2018) “Multi-mode energy management strategy for fuel cell electric vehicles based on driving pattern identification using learning vector quantization neural network algorithm” Journal of Power Sources, Volume 389, Pages 230-239 doi: 10.1016/j.jpowsour.2018.04.024