Dr. Hiroshi Chung and Professor Toshiyuki Otsuka of Kyoto University have succeeded in developing a high-speed algorithm that executes real-time optimal control (model predictive control) by parallel computing.
The problem of finding the optimum movement under constraints is called the optimum control problem.It can be applied to control that skillfully moves all objects, such as automatic operation, stabilization of the power system, and efficiency of chemical reactions.In particular, model predictive control, in which control is performed while resolving the optimal control problem in real time according to the situation from moment to moment, is being actively researched as a method with a wide range of applications.However, complicated problems are complicated and time-consuming, and it is difficult to realize model predictive control.
Model predictive control requires optimization over a finite future period to be performed at each time.However, the movements of the long future are sensitive to the initial state, and it is difficult to optimize the movements of the future as a whole.The method proposed this time decomposes future movements into several fragments, and optimizes each fragment at the same time by parallel calculation.At that time, we devised a method of decomposition so that the influence of adjacent fragments could be properly considered, and succeeded in significantly speeding up the calculation while achieving overall optimization.In general, increasing the number of fragments increases the calculation time, but in the case of the proposed method, multiple fragments can be optimized at the same time, so the rate of increase in calculation time is less than one-fourth of that of the conventional method.
With this research result, it is possible to make the best use of the performance of multi-core processors that have been developed in recent years, and to improve the calculation efficiency by using multiple inexpensive processors.It is expected that the implementation cost of model predictive control will be reduced.Furthermore, it will be possible to apply it to new fields such as automatic driving, robot control, and power network control.In the future, he plans to develop software for industrial applications.
Paper information:[Automatica] A parallel Newton-type method for nonlinear model predictive control
