Controller for controlling a technical system, and method for configuring the controller
A controller for a technical system is trained using a machine learning method. For this purpose, a chronological sequence of training data is detected for the machine learning method, the training data including both state data as well as control action data of the technical system. A change in the control action data over time is detected specifically and correlated with changes in the state data over time within different time windows, wherein a time window-specific correlation value is ascertained in each case. A resulting time window is then ascertained on the basis of the ascertained correlation values, and the training data which is found within the resulting time window is extracted in a time window-specific manner. The controller is trained by means of the machine learning method using the extracted training data and thereby configured to control the technical system.
1. A computer-implemented method for configuring a control device for a technical system using a machine learning method, the method for configuring the control device comprising:
a) a temporal sequence of training data for the machine learning method is captured, wherein the training data comprise state data specifying states of the technical system and control action data specifying control actions of the technical system,
b) a change in the control action data over time is specifically captured,
c) the change in the control action data over time is correlated with changes in the state data over time within different time windows, wherein a time-window-specific correlation value is ascertained in each case,
d) a resulting time window is ascertained on the basis of the ascertained correlation values,
e) the training data situated within the resulting time window are extracted on a time-window-specific basis, and
f) the control device is trained on the basis of the extracted training data using the machine learning method and is therefore configured to optimize control of the technical system.
2. The method as claimed in claim 1 , wherein the technical system comprises a gas turbine, a wind turbine, a solar power plant, a power station, an internal combustion engine, a robot, a production plant, a motor vehicle, a traffic control system, an energy control unit for a building, a power transmission system, a 3D printer, or another machine.
3. The method as claimed in claim 1 ,
wherein respective correlation values for different time windows are compared, and
wherein a time window which is optimized with respect to the respective correlation values is ascertained as the resulting time window.
4. The method as claimed in claim 1 ,
wherein a multiplicity of different time windows are generated, wherein the generation is controlled by means of an optimization method in such a manner that a respective correlation value of the generated time windows is optimized.
5. The method as claimed in claim 4 ,
wherein a gradient descent method, particle swarm optimization and/or a genetic optimization method is used as the optimization method.
6. The method as claimed in claim 1 ,
wherein the control action data comprise action components specifying different control actions,
wherein the change in the control action data over time is specifically captured for a respective action component,
wherein the correlation values are specifically ascertained for a respective action component, and
wherein the resulting time window is ascertained by combining action-component-specific time windows.
7. The method as claimed claim 1 ,
wherein the state data comprise state components specifying different state parameters of the technical system,
wherein the correlation values are specifically ascertained for a respective state component, and
wherein the resulting time window is ascertained by combining state-component-specific time windows.
8. The method as claimed in claim 1 ,
wherein a multiplicity of changes in the control action data over time are captured and are each correlated with a multiplicity of changes in the state data over time using different time windows defined relative to the respective change in the control action data over time, wherein a time-window-specific correlation value is ascertained in each case.
9. The method as claimed in claim 1 ,
wherein the resulting time window is ascertained by means of a combination, an average and/or another combination of different time windows.
10. The method as claimed in claim 1 ,
wherein the state data and/or the control action data are aggregated using an aggregation time window, and
wherein the resulting time window is ascertained on the basis of the aggregated state data and/or the aggregated control action data.
11. The method as claimed in claim 10 ,
wherein different aggregation time windows are used,
wherein the correlation values are specifically ascertained for a respective aggregation time window, and
wherein an aggregation time window which is optimized with respect to the correlation values is ascertained and used.
12. The method as claimed in claim 1 ,
wherein training data situated outside the resulting time window are suppressed or rejected when training the control device.
13. The control device for controlling the technical system, designed to carry out the method as claimed in claim 1 .
14. A computer program product, comprising a non-transitory computer readable hardware storage device having computer readable program code stored therein, said program code executable by a processor of a computer system to implement the method as claimed in claim 1 .
15. A non-transitory computer-readable storage medium having the computer program product as claimed in claim 14 .