Design And Implementation Of Nonlinear Model Predictive Controller Based On FPGA

Model Predictive Control (MPC) is an advanced optimization algorithm and the lin-ear model is widely used to predict the future behavior of systems. While most of theactual systems have inherently non-linear characteristics. At the same time, the controlsystem is becoming more and more complex with higher requirements of performance.Linear model predictive control always can’t meet the requirements of control perfor-mance, and even afects the stability of the system. So Nonlinear Model Prediction hasattracted more and more research. Nonlinear model predictive control needs to solvenonlinear programming problems online, and the computational burden is much heavierwhich has limited its applications.To solve this problem, some researchers advanced novel optimization strategies andalgorithm to speed the online solution. Besides, some researchers designed hardwarestructure with pipeline or parallel to improve the computational speed. This paper adoptsParticle Swarm Optimization (PSO) algorithm to solve the nonlinear programming prob-lems, and high-level synthesis tools Catapult C is used to assist design of FPGA, and arelatively optimized solution is obtained after research of hardware structure. The workof this paper includes:1. In order to take full advantage of the parallel character of FPGA, this paper adoptsparticle swarm optimization algorithm to solve nonlinear programming problems, codingand verification of PSO algorithms is completed in MATLAB. Because M language is notsupported in Catapult C, fixed-point data types are used to increase computational speed,the PSO algorithm written by M language is coded in C with fixed-point data type.2. According to the characteristics of particle swarm optimization algorithm, opti-mized solution is obtained with trade-of between performance and area by parallel andpipeline processing in the Catapult C. At the same time, design flow of FPGA based onCatapult C is improved. For example, the efect of data loss caused by serial communi-cation is added in the of-line simulation, verification of fixed-point data type is executedin Visual Studio environment with data sampled in MATLAB. 3. In order to verify the efectiveness and real-time control of the designed con-troller, platform based on FPGA prototype board DE3and dSPACE real-time systemis built. And UART-RS232interface is implemented as communication prototype withhand-written Verilog. Then, a real-time simulation is conducted for stability control issuesof the in-wheel motor electric vehicle. Simulation results show that the designed nonlin-ear predictive controller has good properties of efectiveness and real-time control, whichproves the feasibility of using high-level synthesis tools,Catapult C, to design FPGA.4. For stability control issues of the in-wheel motor electric vehicle, this paperuses two diferent controller’s structure including stratification and unified integrationto achieve yaw stability control, similarly, the results are analyzed.This article implements full hardware FPGA design using high-level synthesis toolsCatapult C. The overall design process is improved and the difculty of development isinclined to some extent. However it still exists some problems in the process, such ascomplex design consumes much time both in Catapult C and Quartus II, which resultsin difculties in design and optimization, and timing closure are not always guaranteedwith RTL code got from Catapult C. In addition, there are some problems of particleswarm optimization, such as difculty in multi-variable optimization solving, sensitivityto initial conditions etc. Therefore, we consider its improvement or a new kind of nonlinearoptimization algorithm. For time-consuming problem, we consider a new bottom to updesign flow based on C-Core in the Catapult C. And timing analysis and optimization isconsidered to get the timing closure with constraints in Synthesis, Place and Route.