Back to results
US 11,876,464 B2Granted
Systems, methods and devices for neural network control for IPM motor drives
AssigneeThe Board of Trustees of The University of Alabama
InventorsShuhui Li, Michael H. Fairbank
Life of this patent
Estimating the term…
Abstract
Described herein is a method and system for controlling an interior-mounted permanent magnet (IPM) alternating-current (AC) electrical machine utilizing a space vector pulse-width modulated (SVPWM) converter operably connected between an electrical power source and the IPM AC electrical machine comprising three neural networks (NNs), including a controller NN operably connected to the SVPWM converter, a parameter estimator NN, and a flux-weakening and MTPA NN.
Claims18
A method for controlling an interior-mounted permanent magnet (IPM) alternating-current (AC) electrical machine by:
- providing a space vector pulse-width modulated (SVPWM) converter operably connected between an electrical power source and the IPM AC electrical machine;
- providing a neural network vector control system operably connected to the SVPWM converter, the neural network vector control system comprising a controller neural network configured and trained to implement an approximate dynamic programming (ADP) algorithm, wherein the neural network vector control system is trained at times when the IPM AC electrical machine is not operating;
- providing a parameter estimator neural network, wherein the parameter estimator neural network receives as inputs actual d and q-axis currents and actual d and q-axis voltages representative of current and voltage output from the SVPWM converter, wherein the parameter estimator neural network captures and estimates parameters for the IPM AC electrical machine, wherein an adjustable IPM motor model is applied with the parameter estimator neural network to provide estimated d and q-axis currents, wherein the estimated d and q-axis currents are compared with the actual d and q-axis currents and fed-back as inputs to the estimator neural network, and wherein the parameter estimator neural network is trained to implement an ADP algorithm and the parameter estimator neural network system is trained at times when the IPM AC electrical machine is not operating;
- providing a flux weakening and maximum torque per ampere (MTPA) neural network comprising an input layer, multiple hidden layers and an output layer, wherein the flux-weakening and MTPA neural network receives as an input a desired torque command and receives as inputs from the parameter estimator neural network at the input layer a reference torque Ld, Lq, ωe, and ψpm, wherein Ld and Lq are stator d- and q-axis inductances, ωe is the IPM AC electrical machine's electrical rotational speed, and ψpm is a flux linkage of a permanent rotor magnet of the IPM AC electrical machine, and wherein the rotor position of the IPM AC electrical machine is estimated from co, using the parameter estimator neural network and the rotor position is used to convert the current and voltage outputs from the SVPWM converter to their d and q-axis currents and d and q-axis voltages representative of current and voltage output from the SVPWM converter,
- wherein the flux-weakening and MTPA neural network outputs the d and q-axis currents to the controller neural network at the output layer as reference d and q-axis stator currents,
- wherein the flux-weakening and MTPA neural network is trained to minimize a cost function, and
- wherein the cost function can be used to measure how close the reference d and q-axis stator currents from the flux-weakening and MTPA neural network match the ideal optimal reference d and q-axis stator currents for different reference torque and motor parameters.
4 more elements
A system for controlling an interior-mounted permanent magnet (IPM) alternating-current (AC) electrical machine comprising:
- a space vector pulse-width modulated (SVPWM) converter operably connected between an electrical power source and the IPM AC electrical machine;
- a neural network vector control system operably connected to the SVPWM converter, the neural network vector control system comprising a processor and a controller neural network configured to implement an approximate dynamic programming (ADP) algorithm for training the neural network vector control system when the IPM AC electrical machine is not operating;
- a parameter estimator neural network operably connected to an output of the SVPWM converter through a dq-axis converter, wherein the parameter estimator neural network comprises a processor and wherein the parameter estimator neural network is configured to receive as inputs d and q-axis currents and d and q-axis voltages representative of current and voltage output from the SVPWM converter, capture and estimate parameters for the IPM AC electrical machine including at least an electrical rotational speed of the IPM AC electrical machine and a rotor position of the IPM AC electrical machine, and implement an ADP training algorithm, wherein an adjustable IPM motor model is applied with the parameter estimator neural network to provide the estimated d and q-axis currents, wherein the estimated d and q-axis currents are compared with actual d and q-axis currents and fed-backed as inputs to the estimator neural network, wherein the parameter estimator neural network system is trained at times when the IPM AC electrical machine is not operating;
- a flux weakening and maximum torque per ampere (MTPA) neural network operably connected to the neural network vector control system and the parameter estimator neural network, wherein the flux-weakening and MTPA neural network comprises at least a processor, an input layer, multiple hidden layers and an output layer, and is configured to receive as an input a desired torque command and receive as inputs from the parameter estimator neural network at the input layer a reference torque, Ld, Lq, ωe, and ψpm, wherein Ld and Lq are stator d- and q-axis inductances, ωe is the IPM AC electrical machine's electrical rotational speed, and ψpm is a flux linkage of a permanent rotor magnet of the IPM AC electrical machine, and wherein the rotor position of the IPM AC electrical machine is estimated from ωe using the parameter estimator neural network and the rotor position is used to convert the current and voltage outputs from the SVPWM converter to their d and q-axis currents and d and q-axis voltages representative of current and voltage output from the SVPWM converter,
- wherein the flux-weakening and MTPA neural network is configured to output the d and q-axis currents to the controller neural network at the output layer as reference d and q-axis stator currents,
- wherein the flux-weakening and MTPA neural network is trained to minimize a cost function, and
- wherein the cost function can be used to measure how close the reference d and q-axis stator currents from the flux-weakening and MTPA neural network match the ideal optimal reference d and q-axis stator currents for different reference torque and motor parameters.
4 more elements
A system for controlling an interior-mounted permanent magnet (IPM) alternating-current (AC) electrical machine comprising:
- a space vector pulse-width modulated (SVPWM) converter operably connected between an electrical power source and the IPM AC electrical machine;
- a neural network vector control system operably connected to the SVPWM converter, the neural network vector control system comprising a processor and a controller neural network configured to implement an algorithm for training the neural network vector control system when the IPM AC electrical machine is not operating;
- a parameter estimator neural network operably connected to an output of the SVPWM converter through a converter, wherein the parameter estimator neural network comprises a processor and wherein the parameter estimator neural network is configured to receive as inputs d and q-axis currents and d and q-axis voltages representative of current and voltage output from the SVPWM converter, capture and estimate parameters for the IPM AC electrical machine including at least an electrical rotational speed of the IPM AC electrical machine and a rotor position of the IPM AC electrical machine, and implement an ADP training algorithm;
- a flux weakening and maximum torque per ampere (MTPA) neural network operably connected to the neural network vector control system and the parameter estimator neural network, wherein the flux-weakening and MTPA neural network comprises at least a processor, an input layer, multiple hidden layers and an output layer, and is configured to receive as an input a desired torque command and receive as inputs from the parameter estimator neural network at the input layer a reference torque, Ld, Lq, ωe, and ψpm wherein Ld and Lq are stator d- and q-axis inductances, ωe is the IPM AC electrical machine's electrical rotational speed, and ψpm is a flux linkage of a permanent rotor magnet of the IPM AC electrical machine, and wherein the rotor position of the IPM AC electrical machine is estimated from ωe using the parameter estimator neural network,
- wherein the flux-weakening and MTPA neural network is configured to output the d and q-axis currents to the controller neural network at the output layer as reference d and q-axis stator currents,
- wherein the flux-weakening and MTPA neural network is trained to minimize a cost function, and
- wherein the cost function can be used to measure how close the reference d and q-axis stator currents from the flux-weakening and MTPA neural network match the ideal optimal reference d and q-axis stator currents for different reference torque and motor parameters.
4 more elements
Dependencies are inferred from the claim text.