Control objective integration system, control objective integration method and control objective integration program

1. A control objective integration system comprising:

a hardware including a processor;

an expert model unit, implemented at least by the processor, which generates predicted expert control actions based on an expert model which is a machine learning model trained using data collected when an expert operated a plant which is a control target or a plant of the same or similar characteristics;

a transformer, implemented at least by the processor, which constructs metrics or error measures involving the predicted expert control actions from the expert model unit as an objective term; and

a combiner, implemented at least by the processor, which collects different objective terms from the transformer and a learner which outputs machine-learning models as objective terms and computes an optimal set of weights or combinations of the objective terms to construct an aggregated cost function for use in an optimizer.

2. The control objective integration system according to claim 1 ,

wherein based on the current and expected plant situation, the transformer chooses the metric or error measures for measuring the deviation between the expert inputs as predicted by the expert model unit and the optimal inputs desired by a user.

3. The control objective integration system according to claim 1 ,

wherein the combiner collects two different classes of objective terms generated using different principles.

4. The control objective integration system according to claim 3 ,

wherein one class of objective terms represents quantities or measures that are mechanical in nature, the other class of objective terms acts as performance measures related to human skills, sensitivity or preferences.

5. The control objective integration system according to claim 1 ,

wherein the combiner employs Pareto-based multiobjective optimization approaches.

6. A control objective integration method comprising:

generating predicted expert control actions based on an expert model which is a machine learning model trained using data collected when an expert operated a plant which is a control target or a plant of the same or similar characteristics;

constructing metrics or error measures involving the predicted expert control actions based on the expert model as an objective term; and

collecting different objective terms from a learner which outputs machine-learning models as objective terms and computing an optimal set of weights or combinations of the objective terms to construct an aggregated cost function for use in an optimizer.

7. The control objective integration method according to claim 6 ,

based on the current and expected plant situation, choosing the metric or error measures for measuring the deviation between the expert inputs as predicted-by and the optimal inputs desired by a user.

8. A non-transitory computer readable information recording medium storing a control objective integration program, when executed by a processor, which performs a method for:

generating predicted expert control actions based on an expert model which is a machine learning model trained using data collected when an expert operated a plant which is a control target or a plant of the same or similar characteristics;

constructing metrics or error measures involving the predicted expert control actions based on the expert model as an objective term; and

of collecting different objective terms from a learner which outputs machine-learning models as objective terms and computing an optimal set of weights or combinations of the objective terms to construct an aggregated cost function for use in an optimizer.

9. The non-transitory computer-readable recording medium according to claim 8 , based on the current and expected plant situation, choosing the metric or error measures for measuring the deviation between the expert inputs as predicted and the optimal inputs desired by a user.