A learning data generation device for a plant controller, a learning data generation method for a plant controller, and a plant controller.

Abstract

To provide an improved learning data generation device for plant controllers, a learning data generation method for plant controllers, and a plant controller. [Solution] This disclosure relates to a plant controller learning data generation device 1 for generating learning data for a plant controller that controls a plant, wherein the plant controller controls a controlled object using machine learning, and comprises a learning data generation unit 10 that generates predetermined learning data DB1 including a controlled variable, an manipulated variable and boundary conditions based on predetermined calculation conditions, and outputs a label indicating the calculation conditions on which the predetermined learning data was generated, associating it with the predetermined learning data, and an analysis unit 20 that analyzes the learning data generated by the learning data generation unit based on the label.

Claims

1. A plant controller learning data generation device that generates learning data for a plant controller that controls a plant, The aforementioned plant controller uses machine learning to control the target object. A learning data generation unit generates predetermined learning data including a controlled variable, an manipulated variable, and boundary conditions based on predetermined calculation conditions, and outputs a label indicating the calculation conditions on which the predetermined learning data was generated, associated with the predetermined learning data. An analysis unit that analyzes the training data generated by the training data generation unit based on the labels. A learning data generation device for plant controllers, equipped with the following features.

2. The labels include a measurement label indicating that the training data was generated based on actual measurement data relating to the controlled object, and a simulator label indicating that the training data was generated by a simulation relating to the controlled object. A learning data generation device for a plant controller according to claim 1.

3. The simulator label includes a scenario label indicating that the training data was generated based on a predetermined scenario, and a model label indicating that the training data was generated by a predetermined model. The learning data generation device for a plant controller according to claim 2.

4. The model label includes at least one of the following cases: when the predetermined model is a PID control model, when the predetermined model is a machine learning model, or when the predetermined model is a plant model. The learning data generation device for a plant controller according to claim 2.

5. Based on the analysis results output by the analysis unit, the predetermined calculation conditions are modified. The learning data generation device for a plant controller according to claim 3.

6. The plant simulator is updated based on actual measurement data. The learning data generation device for a plant controller according to claim 5.

7. Among the training data having the simulator label, training data that falls within a predetermined range from the training data having the measured label is replaced with the measured data. A learning data generation device for a plant controller according to claim 6.

8. The analysis results output by the analysis unit include evaluation values ​​obtained by evaluating the training data generated by the training data generation unit using predetermined indicators. The learning data generation device for a plant controller according to claim 7.

9. The predetermined indicators include a confidence level that indicates the reliability of the training data. The learning data generation device for a plant controller according to claim 8.

10. The predetermined indicator includes recency, which indicates the elapsed time since the training data was generated. The learning data generation device for a plant controller according to claim 8.

11. The predetermined indicator includes a coverage degree that indicates the range covered by the training data. The learning data generation device for a plant controller according to claim 8.

12. The predetermined indicators include at least one of the following: confidence level, indicating the reliability of the training data; recency, indicating the elapsed time since the training data was generated; or coverage, indicating the range covered by the training data; or a combination of any two of the following: confidence level, recency, and coverage. The learning data generation device for a plant controller according to claim 8.

13. The analysis results include the ratio of training data with the measured labels to training data with the simulator labels in the training data generated by the training data generation unit. The learning data generation device for a plant controller according to claim 8.

14. A method for generating training data for a plant controller that controls a plant, The aforementioned plant controller uses machine learning to control the target object. Set the predetermined calculation conditions, Based on the predetermined calculation conditions, predetermined training data including a controlled variable, a manipulated variable, and boundary conditions is generated, and a label indicating the calculation conditions on which the predetermined training data was generated is output in association with the predetermined training data. The generated training data is analyzed based on the labels. Method for generating training data for plant controllers.

15. A plant controller equipped with a machine learning model, The machine learning model is trained using training data generated by the training data generation device for plant controllers described in any one of claims 1 to 14. A plant controller equipped with a machine learning model.

Images