FPGA Heterogeneous Computing And Application Of Predictive Controller Based On Support Vector Machine

Owing to its ability to handle multi-variable/multi-objective problems and deal with system constrains explicitly and actively,model predictive control(MPC)has been widely used,which includes model prediction,online optimization and feedback correction.Since the solution process of MPC is calculated repeatedly in every limited sampling instant,traditional controller hardware implementations have no longer been able to meet the requirements of high real-time,embedded system,low-cost constraints and high-level integration proposed by fast dynamic system.Therefore,the field programmable gate array(FPGA)implementation has become a hot research topic.The complexity of system in the new application field is gradually increasing,and the requirements for integration level,scheduling capability,memory resource and communication methods of control system are rapidly increasing.The traditional single FPGA implementation cannot meet the demand gradually.In order to improve the flexibility of the system and enhance the system performance,this thesis proposes the FPGA heterogeneous computing implementation scheme of MPC,making full use of the advantages of Acorn RISC Machine(ARM)and FPGA.Through the combination of software and hardware design,a high-performance MPC controller is obtained,which meets the requirements of speed,area and system capability.The design contents include data access design,memory structure optimization,system port design and parallel design.To verify the quickness and the effectiveness of the MPC controller,the path following control of autonomous vehicle real-time experiment is carried out.Experimental results verify the quickness and effectiveness of the MPC controller based on FPGA heterogenous computing.This paper mainly includes the following three parts:1.Considering that MPC is a model-based algorithm,it is difficult to establish a controller-oriented mechanism model,since the controlled objects in new application areas,such as vehicle systems,have the characteristics of strong nonlinearity,mechanism coupling,etc.Hence,this thesis proposes Support Vector Machine(SVM)based MPC method.The SVM algorithm is applied to train the controlled object model in order to obtain the SVM model with simple structure and comprehensive information of the controlled object.Then,the predictive controller is designed based on SVM model and the original dual optimization algorithm is adopted to solve the quadratic programming problem of controller.Finally,the vehicle path following problem is applied to the application object,building "SVM based MPC controller-autonomous vehicle system" closed loop system and completing the autonomous vehicle path following experiment off-line.The effectiveness of SVM based MPC controller is verified.2.Owing to its characteristics that MPC algorithm need to be solved online and considering the real-time,miniaturization,low cost and other requirements of the automotive electronic control system,this thesis proposes an FPGA heterogeneous computing acceleration scheme,which focus on improving the online calculation's ability of the predictive controller.An MPC controller combining software and hardware is designed based on ZYNQ(ARM+FPGA).First,according to the development process of heterogeneous computing,the software design on the ARM processor of controller is carried out.Analyze the algorithm structure of predictive controller and implement the algorithm by C/C++ code.Then,according to the data precision requirements of each module,design the fixed-point data structure and complete the fixed-point C/C++ code of control algorithm.The open-loop experiment of software module on a chip verification is designed by SDSoC,which proves the accuracy of the algorithm.Second,based on the structure of algorithm,the original dual algorithm is transplanted to FPGA.Through data access design,memory structure optimization and system port design,the communication efficiency between hardware and software is improved.Then,optimize the pipeline expansion of matrix operation module to improve calculation speed of FPGA.Finally,the open-loop verification of controller with software and hardware on chip is designed and prove the accuracy of the MPC controller.3.In order to verify the effectiveness and real-time performance of MPC controller based on FPGA heterogenous computing,the hardware in the loop experiment combined with MicroAutoBox for the path following of autonomous vehicle is carried out.First,the autonomous vehicle's path following experiment platform is built based on MicroAutoBox and FPGA,in which MicroAutoBox works as real-time simulation system for autonomous vehicle and FPGA works as the hardware platform for MPC controller.In the experiment platform,ethernet is used to connect FPGA and MicroAutoBox.The autonomous vehicle path following experiment is completed on the real-time experiment platform and then compare computing controllers' performance between ARM scheme and FPGA heterogenous scheme.The calculation time of MPC controller based on FPGA heterogenous computing acceleration scheme is 3.74 ms which is increased by 2.88 times from the calculation time of pure software scheme,and the autonomous vehicle sampling time is 10 ms.The experimental results verify the real-time and effectiveness of FPGA heterogenous computing scheme of MPC controller based on support vector machine.The research work of this thesis is supported by the National Nature Science Foundation of China(No.6179560010),the National Nature Science Foundation of China(No.61703176)and the Funds for Joint Project of Jilin Province and Jilin University under Grant(No.SXGJSF2017-2-1-1).