Real-Time Algorithms and Applications of Nonlinear Model Predictive Control
Conference Room 22
June 9, 2019, 13:00 - 17:30
Department of Systems Science, Graduate School of Informatics, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, Kyoto 606-8501, Japan
In nonlinear model predictive control (NMPC), a nonlinear optimal control problem over a finite future is solved at each sampling time, and the initial value of the optimal control is used as the actual control input to the system, which results in a kind of state feedback.
NMPC can deal with a wide variety of control problems as long as the nonlinear optimal control problem can be solved in real time, i.e., within the sampling period of the feedback control system.
Real-time algorithms for NMPC and their applications have been active areas of research for more than a decade.
The goal of this workshop is to provide an overview of this field and to introduce challenging directions of research such as applications to fast systems or complicated problems and parallel computation.
- Students and researchers in systems and control who are looking for challenging research topics with practical applications
- Engineers and researchers in various engineering fields looking for tools for dealing with complicated control problems
|13:00 - 14:00||Toshiyuki Ohtsuka (Kyoto University)|
|Session 1: Overview of Nonlinear Model Predictive Control: Formulations, Algorithms, and Applications|
|14:00 - 15:00||Haoyang Deng (Kyoto University)|
|Session 2: Parallel Computing for Nonlinear Model Predictive Control: A Highly Parallelizable Algorithm and Its Code Generation Toolkit|
|15:00 - 15:30||Break|
|15:30 - 16:30||Md Abdus Samad Kamal (Gunma University and Adjunct faculty of Monash University)|
|Session 3: Efficient Driving of Automated Vehicles at Intersections Using Nonlinear Model Predictive Control|
|16:30 - 17:30||Tielong Shen (Sophia University)|
|Session 4: Model Predictive Control of Automotive Powertrains|
Overview of Nonlinear Model Predictive Control: Formulations, Algorithms, and Applications
This talk gives an overview of NMPC. In particular, we focus on such practical issues as numerical algorithms, applications, and software tools of NMPC. Real-time optimization is a key component of NMPC, and various numerical algorithms dedicated for NMPC have been developed in recent years. Progresses in numerical algorithms and computing power have been widening application areas of NMPC from slow and simple systems to fast and complicated systems. Application examples in this talk include a robotic manipulator, a hovercraft, automobiles, drones, and a climbing humanoid robot. To deal with complicated mathematical models, symbolic computation and automatic code generation are essential in implementation of NMPC. Some software tools to automate programming of numerical algorithms for NMPC are also introduced.
Parallel Computing for Nonlinear Model Predictive Control: A Highly Parallelizable Algorithm and Its Code Generation Toolkit
Due to the abilities in handling MIMO nonlinear dynamics and complicated constraints, NMPC has been applied to numerous areas. However, its computational demand makes it difficult to be applied to large-scale and high-sampling-rate problems. Existing algorithms based on serial computing has reached the limit of performance improvement due to the stagnation of single-core performance. One solution is to make use of the computational power of multi-core processors, which can be seen everywhere nowadays. However, the difficulty of parallel computing for NMPC is the lack of parallel algorithms and corresponding software. In this talk, a highly parallelizable algorithm dedicated for NMPC and its open-source software requiring no parallel programming skill will be introduced. Also, several numerical and real-world examples will be shown.
Md Abdus Samad Kamal
Efficient Driving of Automated Vehicles at Intersections Using Nonlinear Model Predictive Control
Anticipative driving of vehicles, considering the microscopic dynamics of the preceding traffic and signal timings at the intersection, can bring great improvement in the traffic flows, fuel consumption and emission of CO2 gas. For automated driving vehicles and traffic management systems, such anticipative decision can be realized using NMPC since dynamics of individual vehicles and their interactions are highly nonlinear. In this workshop, we would like to present NMPC solutions for two different traffic scenarios.
Firstly, we present a coordination scheme for fully connected and automated vehicles at an intersection without using any traffic lights. All vehicles approaching the intersection are globally coordinated using an NMPC framework in order to achieve smooth traffic flows at the intersection. The optimal trajectories of the vehicles are computed based on avoidance of their cross-collision risks around the intersection under relevant constraints and preferences. The proposed NMPC framework is evaluated through microscopic traffic simulation in a typical test intersection consisting of both multi-lanes and single-lane approaches with turning movements of vehicles.
Secondly, we present a mixed manual-automated traffic scenario at an isolated intersection where automated vehicles need to be controlled efficiently considering the dynamics of manually driven vehicles and traffic signal timings. The optimal acceleration of each automated vehicle is calculated in the NMPC framework on the basis of the preceding vehicle’s behavior and traffic signal timings. To keep the optimization problem simple and computationally tractable, in the objective function we have incorporated a continuous potential function to replicate influences of discrete change of traffic signal. We have evaluated both the proposed NMPC methods using microscopic traffic simulator.
Model Predictive Control of Automotive Powertrains
This lecture focuses on application of model predictive control to automotive powertrains. After a brief review of model predictive control algorithms including C/GMRES. The following three case studies for the model predictive control of combustion engine and hybrid electric vehicle are introduced; the torque control of gasoline engine, real-time optimization of energy consumption of hybrid electric powertrain, and optimization of hybrid electric powertrain with V2X information.