Study On The Energy Management Strategy Of Hybrid Electric Vehicle Based On Driving Condition Recognition

Countries around the world are getting more extreme in the creation and execution of fuel consumption and emission restrictions in light of the global energy crisis and environmental pollution.Plug-in hybrid electric vehicles(PHEVs)are a significant technical mean for energy conservation and emission reduction in the automobile industry because they may reduce emissions and fuel consumption by fully optimizing the efficiency of the power system,energy recovery,and idle cancellation.The energy allocation of the PHEV multi-energy power system,which is directly related to vehicle fuel economy,is determined by the energy management strategy.Currently,energy management strategies are mostly based on optimization and calibration of specific driving conditions.In order to improve the strategy’s adaptability to different driving conditions,this paper studies the energy management strategy based on driving condition recognition,including urban driving condition construction,vehicle model construction and simulation verification,global optimization and instantaneous optimization energy management strategy research,and energy management strategy research based on driving condition adaptation.The specific work is as follows:First,the method of constructing vehicle driving condition was studied.The data obtained from autonomous driving were preprocessed,and then divided into several kinematic segments.To describe each section,15 feature parameters were used.Principal component analysis was used to reduce the dimensionality of features from 15 to 4dimensions in order to lessen information redundancy among them.Then,based on the reduced feature space,kinematic segments were clustered into four categories using the Kmeans clustering algorithm.And the silhouette coefficient approach was used to assess the validity of the clustering,According to the proportion of time in each category,the driving conditions were synthesized.The relative error rate of the majority of the driving conditions was within 5%,demonstrating the effectiveness of the construction process,when compared to the characteristics of the driving condition database.The research towards driving condition provides a research-based driving condition and data foundation for subsequent strategy development.Secondly,a platform for simulating PHEV vehicles was created.The properties of a single-axis parallel PHEV configuration were examined,then model based on the parameters of the entire vehicle and its parts was built in the Matlab/Simulink environment.Subsequently,a rule-based strategy was designed in conjunction with the vehicle model to verify the effectiveness of the model and evaluate the control effect of the rule-based energy management strategy.Thirdly,optimization-based energy management strategies were studied.To evaluate and reference subsequent strategies,a dynamic programming energy management strategy was designed,and the SOC trajectory and torque allocation under this strategy were studied.To implement the application of online policies,an instantaneous optimal equivalent fuel consumption minimization strategy was constructed,and simulations were conducted under different driving cycle conditions to analyze the impact of the equivalent factor on SOC trajectory and torque distribution strategy.Subsequently,the factors affecting the equivalent factor were studied,and the relationship between the category of driving conditions,initial SOC,driving distance,and equivalent factor was analyzed.Based on the results of this study,the equivalent factor was optimized using the gray wolf algorithm,and the ideal equivalent factor MAP based on the initial SOC and driving distance was established for each typical driving condition.This established the foundation for the real-time updating of the equivalent factor of subsequent strategies.Finally,The energy management strategies based on driving condition recognition were studied.A bidirectional long-short term memory(Bi-LSTM)network was used to create a driving condition recognition model based on the clustering of the driving condition feature parameters.The Bayesian theory was presented to optimize the network hyperparameters with the minimal error rate as the optimization goal in order to minimize manual tweaking of the hyperparameters.Offline validation of the driving condition recognition model using the driving condition testing set yielded great recognition accuracy for various typical driving conditions.Two linear SOC reference trajectories were developed,and the energy consumption was constrained by the SOC trajectory following strategy.An adaptive ECMS strategy model based on Bi-LSTM driving condition recognition was created by coupling the driving condition recognition model,vehicle model,equivalent fuel consumption minimization model,SOC trajectory following strategy,and optimal equivalent factor MAP.Simulations were conducted under different testing driving conditions to verify the effectiveness of the proposed strategy.Results indicate that the adaptive ECMS strategy based on Bi-LSTM driving condition recognition has good adaptability to various driving conditions.The energy distribution is reasonable for different typical driving conditions,and the battery SOC trajectory follows well.The proportion of operating points in the engine’s low efficiency zone decreased significantly.The fuel economy has improved by 6.09% to8.17% compared to proposed rule-based strategy under different testing driving cycle conditions.The proposed strategy reduces the sensitivity to driving conditions and provides a reference for exploring the energy-saving potential of PHEVs.