Research On Tensor Decomposition Predictive Control Strategy Of Vehicular Intelligent Stable Platform

In the engineering fields such as geological prospecting,aerospace,instrument manufacturing and others,the application scenarios of high-precision equipment are becoming increasingly extensive,thus the demands of vehicle-mounted or airborne-mounted carrier platform have been arising.Due to the impact of external excitation such as road bumps and crosswinds,mobile exploration conditions are severe,the significance of specialized stable vibration isolation system for on-board equipment has become more and more important.At present,the mainstream proposal of vehicle vibration isolation system is mainly active and semi-active suspension technology,which can improve the ride comfort and safety of vehicle.The semi-active suspension based on the continuous adjustable damping technology can adaptively control the suspension parameters according to different input vibration with reliable structure and high efficiency.The disadvantage is that the design of a vehicle suspension must balance the handling stability and safety performance,and there are optimization contradictions between the stability and ride comfort under different excitation frequency bands,thus the vibration isolation optimization for carrying equipment is limited.While the vehicular stable platform serves as a secondary suspension system which can focus on improving the vibration isolation performance and kinematic control precision.The vehicular stable platform system this paper proposed conducts research on nonlinear dynamics modeling and intelligent control theory.Based on the tensor product model transformation and high-order singular value decomposition method,the gain-scheduled road preview controller is designed to achieve adaptive stable control of vibration and angular vibration according to different road conditions.It has the advantages of fast online calculating,small memory occupation,and high engineering applicability.Specific research contents are as follows:(1)Modeling of vehicular stable platform nonlinear system.In view of the nonlinear characteristics of the vehicular stable platform model,the polytopic linear variable parameter method is applied to transform the nonlinear determination problem into convex combination form,and then the gain-scheduled controller can be used to solve nonlinear dynamic problem.(2)Tensor decomposition for high-order system dimension reduction.In order to otimize the high-order model discreted from nonlinear system,utilize the tensor product model transformation method to decompose the high-order tensors of the state space.Truncated high-order singular value decomposition is used to approximate the eigenvectors corresponding to the orthogonal subspace.Thus,the complexity of tensor space elements is greatly reduced while the approximation accuracy of reduced dimension system is guaranteed,which reduces the occupancy of storage space and subsequent calculation difficulty.(3)Explicit model predictive control strategy research.Model predictive control algorithm is been applied in multi-input multi-output system control for its high controllable and antiinterference performance.Yet solving nonlinear control problems with MPC approach is still limited by the increased online computing complexity.In order to optimize the calculation burden of nonlinear model predictive controller,the explicit model predictive control method is proposed.Based on the idea of parameter scheduling,EMPC solves the parameter related control gain offline,stores and arranges the gain into an explicit high dimensional matrix.For real time EMPC controller implementation,the control gain is obtained by solving the explicit function.Thus,the online calculation is simplified to a function evaluation.Combine the EMPC algorithm with tensor decomposition to realize dimension reduction,so that the orthogonality subspace can be obtained through tensor product transformation and the corresponding feature vectors can be approximated.Thus,the feasibility and efficiency of real-time calculation of gain-scheduled controller for nonlinear system are improved,which makes it more practical in engineering.(4)Intelligent preview module design.The design requirement of gain scheduling controller is that all the time-varying parameters in the gain-scheduled system can be measured.In the intelligent road preview stable platform system,the road excitation signal is an uncertain interference input to the vehicle suspension,which needs to be identified as a measurable variable parameter.The empirical mode decomposition road signal feature classification algorithm based on support vector machine is proposed.In this module,the road stochastic signal is preprocessed into intrinsic mode function which satisfy the EMD constraints,then the time-domain features of the components are extracted to compose the trainer sample.The discrete road samples are trained in SVM classifier to realize realtime identification and classification of road level under different vehicle speeds.Thus,the real-time road level is identified as a variable parameter for gain-scheduled road preview stable platform system.(5)Closed loop experiment verification.Design and process of customized CDC damper to complete the assembling of vehicular stable platform.The feasibility and effectiveness of the platform controller are verified,which provides data support for the targeted debugging of semi-active control algorithm of the platform.The research content of this paper covers the whole process of vibration isolation system design,diversified modeling,control algorithm theoretical research and experimental closedloop verification.The model of vehicular stable vibration isolation platform system was established,and the nonlinear characteristics were linearized by gain-scheduled method.Based on empirical mode decomposition and support vector machine feature classifier,a realtime road preview identification module is designed.It can realize real-time identification and classification of road level as preview information under different vehicle speeds,so the gainscheduled control system is established.And we proposed a method based on tensor product transform and explicit model predictive control strategy.The heavy online computation of predictive controller is replaced by off-line computation to reduce the hardware computing requirement,and the dimension of the off-line system is optimized by tensor product model transformation,which greatly reduces the model complexity and memory occupation,which makes it more practical in engineering.