Research On Driving Cycle Prediction And Intelligent Energy Management Strategy For Plug-in Hybrid Electric Bus

Under the background of global energy shortage and increasingly severe environmental pollution,the key technologies in the field of new energy vehicles are constantly developing.Vigorously promoting the development of plug-in hybrid electric vehicles is the only way to fully popularize pure electric vehicles.Plug in hybrid system is a kind of nonlinear,multivariable and time-varying complex system.Due to its special energy distribution structure,the control of engine,motor and other power sources is more complex than traditional hybrid system.Energy management strategy as one of the key technologies of hybrid system,how to optimize the control of each power source under the premise of meeting the vehicle's driving demand and driver's power demand,realize the efficient and reasonable work of each power source,further exploit its energy saving potential so as to improve the fuel economy of the vehicle is the key to the current research of plug-in hybrid system.At the same time,it is also an urgent demand for the development of new energy vehicle industry.This paper takes the dual motor coaxial hybrid plug-in hybrid electric bus as the research object,carried out the research on method of driving cycle construction,modeling of hybrid system,method of future driving cycle prediction,prediction energy management strategy,intelligent energy management strategy and hardware in the loop test,completed the working process of plug-in hybrid vehicle system modeling,simulation analysis and energy management strategy verification.Based on The National Key R&D Program of China "Dynamic Modeling Optimization and Dynamic Control Method of Plug-in/Extended Range Hybrid Electric System",the main research contents include the following six aspects:1)Driving cycle construction of plug-in hybrid electric bus.The representative bus line of a city is selected and the characteristics are analyzed,and the original driving data are collected.The wavelet transform method is used to filter and denoise the original data,and then the original driving data is divided into independent short journey segments.The characteristic parameters which can fully reflect the characteristics of driving data are selected.The first four principal components are chosen and the self-organizing map neural network algorithm is used to cluster all the short journey samples according to the similarity degree of the characteristic parameters.The driving cycle of a bus line is obtained by combining the clustering results,and the rationality of the constructed driving cycle is verified by comparing with the characteristic parameters of the original driving data.2)Vehicle longitudinal dynamics modeling and energy management strategy construction.Firstly,the configuration characteristics of the dual motor coaxial hybrid plug-in hybrid bus and the energy flow under different working modes are analyzed in detail,and the vehicle longitudinal dynamic model of the plug-in hybrid system including engine,driving motor,ISG motor and power battery pack is established according to the forward modeling idea in the simulation environment of MATLAB/Simulink.Based on the driving cycle construsted,the rule-based energy management strategy commonly used in engineering and the dynamic programming based energy management strategy model which can ensure the global optimization are established.The rationality of the two benchmark strategies is verified,and the reference evaluation standard is provided for the further research of energy management strategy.3)Establishment of the future vehicle speed prediction model based on the data-driven prediction method.Firstly,the speed prediction model based on Markov chain is established,and the initial transition probability matrix is solved by Bayesian estimation.Based on the constructed driving cycle,the vehicle speed is predicted several times in different prediction horizon.Then the RBF neural network vehicle speed prediction model based on driving intention sequence and historical vehicle speed sequence is established,and the rationality of the constructed neural network is verified On the basis of above verification,the future vehicle speed is predicted,then the difference of prediction accuracy between RBF neural network prediction model and Markov chain prediction model is compared and analyzed,and the effectiveness of introducing driving intention sequence in RBF neural network vehicle speed prediction model is verified.4)A hierarchical predictive energy management strategy is proposed.Combining Pontryagin minimization principle with model predictive control,a hierarchical predictive energy management strategy for plug-in hybrid power system based on PMP-MPC is proposed.The upper controller mainly receives the future vehicle speed sequence predicted by the vehicle speed prediction model,and uses PMP algorithm to solve the optimal control sequence in the prediction horizon according to the vehicle state.The lower controller outputs the control command to the vehicle model according to the calculation results of the upper controller.The PMP-MPC strategy is compared with DP-MPC energy management strategy which combines dynamic programming algorithm and model predictive control,and the advantages of PMP-MPC strategy in fuel economy and computational efficiency are verified.5)Intelligent energy management strategy based on deep reinforcement learning is proposed.Combining the structure of deep learning and the idea of reinforcement learning,the energy management strategy architecture of plug-in hybrid power system based on deep reinforcement learning is explored.Based on the two deep reinforcement learning algorithms of double deep Q-network and deep deterministic strategy gradient,the corresponding energy management strategies are developed respectively,and the two strategies are simulated.Compared with the PMP-MPC energy management strategy proposed,the results show that the intelligent energy management strategy based on deep reinforcement learning has a better ability to adapt to the uncertainty of environmental changes under random driving conditions,and the impact of the random changes of the environment on the vehicle economy is small.It provides a new idea and solution for the robustness of plug-in hybrid system in random driving environment.6)The hardware in the loop simulation experiments of predictive energy management strategy and intelligent energy management strategy are carried out.By building the hardware in the loop test platform,two types of energy management control strategies proposed in this paper are downloaded to the real controller respectively,and the control effect of the energy management strategies are verified in the real-time simulation environment.The hardware in the loop test results show that the two types of energy management strategies have good real-time performance and good control effect.