US 11,733,680 B2
Control of matrix converters using machine learning
Mahmoud El Chamie, Manchester, CT (US); and Vladimir Blasko, Avon, CT (US)
Assigned to Hamilton Sundstrand Corporation, Charlotte, NC (US)
Filed by Hamilton Sundstrand Corporation, Charlotte, NC (US)
Filed on Mar. 23, 2020, as Appl. No. 16/826,635.
Prior Publication US 2021/0294304 A1, Sep. 23, 2021
Int. Cl. G05B 19/4155 (2006.01); G06N 20/00 (2019.01); G05B 13/04 (2006.01); G06N 5/02 (2023.01); H02M 5/00 (2006.01)
CPC G05B 19/4155 (2013.01) [G05B 13/04 (2013.01); G06N 5/02 (2013.01); G06N 20/00 (2019.01); G05B 2219/42058 (2013.01); H02M 5/00 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A matrix converter system of an industrial plant system comprising:
a matrix converter having a switching matrix coupled between an input side and an output side;
a model predictive controller (MPC) configured to select a switching state of the switching matrix from a plurality of switching states, the MPC configured to:
receive an operating condition of the industrial plant system;
consult a Q-data structure to access reward values that are mapped to respective switching states of the switching matrix for an operating state that corresponds to the operating condition, wherein the Q-data structure has been trained in a real or simulation environment of the industrial plant system using Q-learning to map till convergence a reward value predicted for the respective switching states of the plurality of switching states and respective discrete operating states of a plurality of discrete operating states;
sort the reward values predicted for and mapped to the respective switching states for a discrete operating state of the plurality of discrete operating states that corresponds to the operating condition;
select a subset of the mappings as a function of a result of sorting the reward values associated with the switching states of the operating state;
evaluate by analyzing each switching state included in the subset; and
select an optimal switching state for the operating condition based on a result of evaluating the switching states of the subset, including using space vector modulation (SVM) to determine a discrete operating state of the plurality of discrete operating states that corresponds to the operating condition,
wherein the operating condition includes a multi-phase high-voltage side (HVS) voltage signal and a multi-phase low-voltage side reference voltage (LVSR) signal, and the SVM includes:
dividing the HVS voltage signal at any time instant into M even phase segments over a full cycle;
dividing phase of the LVSR voltage signal into N even sectors; and
dividing magnitude of each sector into P regions, wherein the plurality of discrete operating states includes M×N×P states.