A Study On Model Identification Of Automobile Lateral Dynamics And Steering Predictive Control

Vehicle dynamics model is the fundamental for system analysis and active control.Therefore,vehicle dynamics modeling has received a renewed attention in recent years.Because of vehicle's parameter uncertainties and under-modeling,traditional modeling methods,including physical modeling based on multi-body dynamics or first principle modeling based on physical law,are not suitable for rapid modeling or still too different from actual vehicle response.In order to improve this situation,a modeling method based on input and output?I/O?data is studied in this paper,which can provide a good tradeoff between model precision and modeling cost.The study on vehicle dynamics modeling by identification has both important academic meaning and practical value.Considering multi-input and multi-output?MIMO?characteristics of vehicle system,the application scope of identified vehicle model is extended from linear domain to non-linear domain,from open-loop off-line identification to closed-loop on-line identification,and the model is used to parameter estimation,output prediction and vehicle controller design.Firstly,under constant vehicle speed and small tire slip angle conditions,a linear time-invariant?LTI?vehicle model can be identified by classical subspace method using I/O data.In order to correspond to the widely-used single-track model,the model order is determined as 2nd.Validation results in time-domain and frequency-domain show that the identified model based on canonical variate analysis?CVA?method is of highest precision in comparison with multivariable output error state space method?MOESP?and numerical algorithms for subspace state space system identification method?N4SID?.And the CVA-step model not only has certain precision in time-domain,but also it accords well with the frequency characteristics of the empirical transfer function?ETF?by national standard.Furthermore,a tire cornering stiffness estimation method based on constant system poles is proposed,and the estimation method is proven effective by numerical examples.In consideration of vehicle roll motion,the model structure increases to ninth order,and the validation results in time-domain show that the identified CVA-step model can be accurately enough to simulate vehicle lateral acceleration and roll angle.In order to improve the identified LTI vehicle model,two non-linear time-varying vehicle model structures under closed-loop operation are proposed,including extended linear model S₁ and incremental non-linear model S₂.In this case,the assumptions of constant vehicle speed and the changeless of tire cornering stiffness are not necessary,which is more reasonable.By recursive optimized version of predictor-based subspace identification method(RPBSID_opt),the above vehicle models can be updated in real-time.The model validation results show that S₁ and S₂ models can improve the model precision in linear domain and describe the vehicle behavior accurately.Even in the non-linear condition when the lateral acceleration is over 0.6g,the proposed vehicle model can still provide a consistently and unbiased estimation.The identified model S₂ has a better predictive accuracy than model S₁ and well-established extended kalman filter?EKF?vehicle state estimator,and it has a better prospect of application.At last,combining the predictive outputs with optimal criterion,an enhanced recursive subspace predictive control method is proposed.The method is not limited to LTI system in strict sense,and the complicated diophantine equations are avoided,which improves the efficiency of the algorithm.For the vehicle active front steering system,two controllers,named recursive subspace predictive controller?RSPC?and recursive subspace predictive controller with integrator?RSCPI?,are designed in order to improve the steering stability of vehicle system.The results of numerical examples show that when tire lateral forces are smaller than the limit of adhesion,both controllers are effective to improve the vehicle stability.When the tire lateral forces reach the limit of adhesion,the RSPCI controller still has a satisfied control effect,and it can make the vehicle to track the desire states well.