Research On High Efficient Predictive Contorl Strategies And Implementation

Model Predictive Control(MPC)has been widely used in the industrial field for decades because of its low requirement for models,good robustness and ability to deal with constraints.One of the core ideas of MPC is rolling optimization,which requires solving an optimization problem online in each control cycle.Practically,in order to get better control results,we need set a smaller sampling interval and a longer control horizon.Then,the reduced sampling interval and rapidly increasing amount of online computation limit the application of MPC in some fast systems.So,the motivation of this paper is to make MPC controllers to solve optimization problems online in a shorter time.In the aspect of algorithm,we propose the Double Speed Frame based Fast Predictive Control Algorithm(DSF-MPC)for high frequency sampling systems.In the aspect of hardware acceleration,we propose the Multi-point Radiation based Parallel Branch and Bound Algorithm(MPRP-BB)and implement it in GPU.The main work of this paper can be summarized as follows:1)Propose the DSF-MPC algorithm.Explain the core idea of the algorithm and the scheme which decomposes the solving process of real-time control into two time scales in detail.Provide the detailed derivation process and the flow diagram of DSF-MPC.Analysis the nominal stability and the algorithm complexity.Discuss the strategy of further improvement of DSF-MPC under the box constraints.Establish the models of the inverted pendulum system and the armature DC motor system.Apply the DSF-MPC algorithm to the both systems in MATLAB environment.Show the control results of DSF-MPC,traditional MPC and PID respectively.After detailed comparison and analysis,verify the effectiveness and high-speed ability of DSF-MPC.2)Give the strategy to parallel the Discrete-time Simplified Dual Neural Network,which is used as a tool to solve QP problems.Split the parallel algorithm to meet the thread structure of GPU.Program on GPU to achieve QP solver.To further improve the efficiency,we do many optimizations on memory management and underlying algorithms,including using shared memory and reduction.3)Propose the Multi-Point Radiation based Parallel Branch and Bound Algorithm(MPRP-BB)through combining the feature of MIQP problems,which can also be implemented in large-scale parallel.Explain its principle,implement method and process in detail.Split the MPRP-BB algorithm to meet the thread structure of GPU.Program on GPU to achieve MIQP solver.4)Establish the model of the unit commitment problem in power dispatch.After linearization,transform the Unit Commitment(UC)problem into a standard MIQP problem.Moreover,inspired by the idea of the rolling optimization in MPC,propose a rolling optimization strategy to solve the UC problem and then solve it by using the MIQP solver built in the paper.This has reference significance for the flexible power dispatch in smart grid.