software

The software realizing the algorithm of NMPC for permanent magnet synchronous motor which uses population optimization methods. The principle of population optimization methods is the population of so-called agents, which move in the space of possible solutions according to a defined strategy. The behavior of every agent can be evaluated separately in a parallel thread. A designed algorithm works in four sequential stages: - Initialization – In this stage, a pseudo-random number generator is used to generate the position of 2^p agents. This position represents increments of stator voltage across the prediction horizon. - Prediction – The model of the controlled plant is evaluated according to the agent's position. - Cost function evaluation – the weighting coefficients represent the effort put on the control of a given value. - Agent movement – agents move by a defined strategy. In a designed algorithm, the agent with the lowest value of cost function attracts the remaining ones. ) Once again, parallelism is used here. Usage of this approach reduces the number of necessary sequential comparisons from 2^p-1 to p. The algorithm performs the stages 2)-4) iteratively for a specified number of iterations. We found this number by experiment. After this number of iterations, the found solutions respect the Karush-Kuhn-Tucker conditions within the requested precision. The algorithm finds the solution in the time requested for proper control of the motor in real-time. Predictive control algorithms require the model of the controlled plant as precise as possible. Unfortunately, the motor parameters are not static and change with operating point . This was to reason to expand the control algorithm with an adaptability block. As the block is separated from the control algorithm itself, it does not affect the time necessary for the proper solution of an optimization problem. On the other hand, it improves the results of control by not affecting the weighting coefficients but only the parameters of a model.

Nonlinear MPC, agents, population optimization method, adaptive control, PMS motor

15. 9. 2021

Software je umístěn na sharepointu NewControl skupiny.

K využití výsledku jiným subjektem je vždy nutné nabytí licence

Poskytovatel licence na výsledek požaduje licenční poplatek

https://newcontrol.ceitec.cz/vysledky/nmpc_pms_pom_adapt