Prediction optimization for system level production control

Patent metadata

Field	Value
Publication number	US-11099529-B2
Application number	US-201916519257-A
Country	US
Kind code	B2
Filing date	Jul 23, 2019
Priority date	Jul 23, 2019
Publication date	Aug 24, 2021
Grant date	Aug 24, 2021

Abstract

A computer-implemented method for controlling a production system includes mapping, by a controller, the production system as a directed acyclic graph. The production system can include multiple plants that are represented as nodes and relations between the plants represented by edges of the directed acyclic graph. The method further includes generating, by the controller, a regression model for each of the plants in the production system. The method further includes predicting, by the controller, an output of each plant based on sensor data associated from each plant. The method further includes adjusting, by the controller, one or more control variables for each plant based on a target output by using machine learning. The method further includes adjusting, by the controller, the one or more control variables for each plant to generate the target output.

First claim

What is claimed is:

1. A computer-implemented method for controlling a production system, the computer-implemented method comprising:
mapping, by a controller, the production system as a directed acyclic graph, the production system comprising a plurality of plants that are represented as nodes and relations between the plants represented by edges of the directed acyclic graph;
generating, by the controller, a regression model for each of the plants in the production system, wherein a first regression model for a first plant of the plurality of plants comprises a gradient-based model and wherein a second regression model for a second plant of the plurality of plants comprises a black-box model;
predicting, by the controller, an output of each plant based on sensor data and the regression model associated from each respective plant;
determining, by the controller, one or more optimized control variables for each plant based on the regression models and a target output by using machine learning, wherein the one or more optimized control variables are selected to maximize the target output, wherein the optimized control variables for the first regression model are determined by decomposing the gradient-based model using a primal dual algorithm, and wherein the optimized control variables for the second regression model are determined using a derivative free optimization and estimating noisy gradients by convolution smoothing;
adjusting, by the controller, one or more set points for each plant based on the optimized control variables to generate the target output, wherein each of the one or more optimized control variables is associated with a set point; and
operating the production system using the adjusted one or more set points.

2. The computer-implemented method of claim 1 , further comprising sending, by the controller, a notification with the optimized one or more control variables.

3. The computer-implemented method of claim 1 , wherein adjusting the one or more set points comprises determining, by the controller, one or more non-linear relations between the plants based on the sensor data.

4. The computer-implemented method of claim 1 , wherein the target output is a multi-period set point comprising a plurality of desired outputs over a predetermined duration.

5. The computer-implemented method of claim 4 , wherein each period in the multi-period set point has respective constraints.

6. A system comprising:
a plurality of plants that form a production system; and
a controller coupled with the plurality of plants for controlling the production system by performing a method comprising:
mapping the production system as a directed acyclic graph, the plurality of plants are represented by nodes and relations between the plants represented by edges of the directed acyclic graph;
generating a regression model for each of the plants in the production system, wherein a first regression model for a first plant of the plurality of plants comprises a gradient-based model and wherein a second regression model for a second plant of the plurality of plants comprises a black-box model;
predicting an output of each plant based on sensor data and the regression model associated from each respective plant;
determining one or more optimized control variables for each plant based on the regression models and a target output by using machine learning, wherein the one or more optimized control variables are selected to maximize the target output, wherein the optimized control variables for the first regression model are determined by decomposing the gradient-based model using a primal dual algorithm, and wherein the optimized control variables for the second regression model are determined using a derivative free optimization and estimating noisy gradients by convolution smoothing;
adjusting one or more set points for each plant based on the optimized control variables to generate the target output, wherein each of the one or more optimized control variables is associated with a set point; and
operating the production system using the adjusted one or more set points.

7. The system of claim 6 , wherein the method further comprises sending, by the controller, a notification with the optimized one or more control variables.

8. The system of claim 6 , wherein adjusting the one or more set points comprises determining, by the controller, one or more non-linear relations between the plants based on the sensor data.

9. The system of claim 6 , wherein the target output is a multi-period set point comprising a plurality of desired outputs over a predetermined duration.

10. The system of claim 9 , wherein each period in the multi-period set point has respective constraints.

11. A computer program product comprising a computer readable storage medium that has computer executable instructions stored thereupon, the computer executable instructions when executed by a processor cause the processor to perform a method comprising:
mapping, by a controller, a production system as a directed acyclic graph, the production system comprising a plurality of plants that are represented as nodes and relations between the plants represented by edges of the directed acyclic graph;
generating, by the controller, a regression model for each of the plants in the production system, wherein a first regression model for a first plant of the plurality of plants comprises a gradient-based model and wherein a second regression model for a second plant of the plurality of plants comprises a black-box model;
predicting, by the controller, an output of each plant based on sensor data and the regression model associated from each respective plant;
determining, by the controller, one or more optimized control variables for each plant based on the regression models and a target output by using machine learning, wherein the one or more optimized control variables are selected to maximize the target output, wherein the optimized control variables for the first regression model are determined by decomposing the gradient-based model using a primal dual algorithm, and wherein the optimized control variables for the second regression model are determined using a derivative free optimization and estimating noisy gradients by convolution smoothing;
adjusting, by the controller, one or more set points for each plant based on the optimized control variables to generate the target output, wherein each of the one or more optimized control variables is associated with a set point; and
operating the production system using the adjusted one or more set points.

12. The computer program product of claim 11 , wherein adjusting the one or more set points comprises determining, by the controller, one or more non-linear relations between the plants based on the sensor data.

13. The computer program product of claim 11 , wherein the target output is a multi-period set point comprising a plurality of desired outputs over a predetermined duration.

14. The computer program product of claim 13 , wherein each period in the multi-period set point has respective constraints.

Assignees

• IBM

Inventors

• Phan Dzung
• Nguyen Lam
• Murali Pavankumar
• Kalagnanam Jayant R

Classifications

• G05B13/0265

  the criterion being a learning criterion · CPC title

• G05B13/041 Primary

  in which a variable is automatically adjusted to optimise the performance · CPC title