# US 2024/0004356 A1 · System and method for building and deploying a sustainable dynamic reduced-order model (sdrom) for an industrial process

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2024/0004356 A1
- Document ID: US20240004356A1
- Status: Published application
- Kind: A1
- Application number: 17/810,169
- Filed: 2022-06-30
- Published: 2024-01-04
- Earliest claimed: 2022-06-30
- Assignees: AspenTech Corporation
- Applicants: AspenTech Corporation
- Inventors: Hong Zhao; Christopher Quan; Shu Wang; Melisa Morales Cárdenas
- Web page: https://patentgrep.com/patent/US20240004356A1

## CPC

- G05B 13/048: {using a predictor}
- G05B 19/41885: {characterised by modeling, simulation of the manufacturing system}
- G05B 11/32: with inputs from more than one sensing element; with outputs to more than one correcting element
- G05B 13/0265: {the criterion being a learning criterion}
- G05B 13/042: {in which a parameter or coefficient is automatically adjusted to optimise the performance}
- G05B 17/02: electric
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)

## Abstract

Embodiments of the present disclosure provide functionality for creating a sustainable dynamic reduced-order model (SDROM) for operating a real-world industrial process. The model is based upon a reduced order model (ROM) trained using data obtained from simulations performed using a first-principles model (FPM) of the real-world industrial process. The trained ROM is divided into multiple-input, single-output (MISO) sub-models, which are partitioned into component terms for incorporation of respective gain factors. The SDROM is deployed online to operate the real-world industrial process with one or more optimization objectives and the SDROM is periodically calibrated and validated using historical operation data.

## Claims

### Claim 1

A computer-implemented method of creating a model for operating a real-world industrial process, the method comprising:
training a reduced order model (ROM) of a subject industrial process using simulation data generated from one or more simulations of the subject industrial process performed using a first-principles model (FPM) of the subject industrial process;
constructing a sustainable dynamic reduced-order model (SDROM) of the subject industrial process by:
(i) dividing the trained ROM into a plurality of multiple-input single-output (MISO) sub-models;
(ii) partitioning each MISO sub-model into multiple component terms; and
(iii) inserting a gain factor into each partitioned component term to form a gain vector for each MISO sub-model; and
deploying the SDROM online to operate the subject industrial process.

### Claim 2 (depends on 1)

The method of claim 1 wherein deploying the SDROM online includes:
obtaining operation data of the subject industrial process; and
periodically calibrating and validating the deployed SDROM using the obtained operation data.

### Claim 3 (depends on 2)

The method of claim 2 wherein periodically calibrating and validating the deployed SDROM includes:
monitoring a given MISO sub-model of the plurality of MISO sub-models by:
(i) retrieving historical operation data from a plant history database;
(ii) filtering elements of the retrieved historical operation data with a dynamic filter attached to a given input of the given MISO sub-model;
(iii) wherein the gain vector formed for the given MISO sub-model is a first gain vector, estimating gain factors of a second gain vector of the given MISO sub-model based on the filtered elements of the retrieved historical operation data;
(iv) comparing the estimated gain factors of the second gain vector with inserted gain factors of the first gain vector to determine model performance metrics; and
(v) displaying, to users, a visual representation of the determined model performance metrics.

### Claim 4 (depends on 2)

The method of claim 2 wherein periodically calibrating and validating the constructed SDROM includes:
adapting a given MISO sub-model of the plurality of MISO sub-models of the constructed SDROM with historical operation data of the subject industrial process by:
receiving allowance to perform adaptations to at least a part of the given MISO sub-model of the constructed SDROM from a user or an auto-system; and
updating the given MISO sub-model by calibrating gain factors of the given MISO sub-model.

### Claim 5 (depends on 4)

The method of claim 4 wherein:
updating the given MISO sub-model by calibrating gain factors of the given MISO sub-model includes:
retrieving the historical operation data, as a time-series, from a plant history database, for a window of time represented within the plant history database;
filtering elements of the retrieved historical operation data with a dynamic filter attached to a given input of the given MISO sub-model;
feeding the filtered elements of the retrieved historical operation data as inputs into the given MISO sub-model and predicting an output time-series corresponding to the filtered elements;
comparing the predicted output with the retrieved historical operation data of the subject industrial process output in R2 statistics;
adjusting, according to the comparing, and subject to one or more suppression constraints, values of the gain factors for the given input of the given MISO sub-model, thereby calibrating the gain factors of the given MISO sub-model;
updating the SDROM to include the given MISO sub-model with the calibrated gain factors; and
validating the updated SDROM with a testing dataset.

### Claim 6 (depends on 1)

The method of claim 1 wherein deploying the SDROM online includes:
receiving a stream of real-time data of the subject industrial process;
filtering elements of the received stream of real-time data with a dynamic filter attached to a given input of a given MISO sub-model;
predicting an output of the given MISO sub-model and estimating local gains of the given MISO sub-model around an updated operating point determined from the filtered elements of the received stream of real-time data; and
determining optimal operation settings of the subject industrial process based on the predicted output and estimated local gains.

### Claim 7 (depends on 1)

The method of claim 1 wherein deploying the SDROM online includes:
determining optimal operation settings of the subject industrial process using the SDROM and an optimization objective function, the optimal operation settings including at least one of a temperature setpoint, a pressure setpoint, a flow rate setpoint, an operating throughput, and a transition path from a current operating state to the determined optimal operation settings; and
operating the subject industrial process in accordance with the determined optimal operation settings.

### Claim 8 (depends on 1)

The method of claim 1 wherein constructing the SDROM further includes:
attaching respective dynamic filters to inputs of each MISO sub-model to align, with respect to time, input data.

### Claim 9 (depends on 8)

The method of claim 8 wherein attaching respective dynamic filters to inputs of each MISO sub-model includes at least one of:
obtaining at least one of the respective dynamic filters from a dynamic finite impulse response (FIR) model by use of a linear model order reduction technique; and
obtaining at least one of the respective dynamic filters from input-output time-series data of the subject industrial process by use of a subspace identification technique.

### Claim 10 (depends on 1)

The method of claim 1 wherein training the ROM of the subject industrial process includes:
dividing the generated simulation data into a training dataset and a testing dataset;
establishing the ROM by fitting the training dataset to a polynomial-based implementation of the FPM with polynomials of a 2nd, or higher, order; and
validating the established ROM with the testing dataset.

### Claim 11 (depends on 1)

The method of claim 1 wherein training the ROM further includes at least one of:
configuring the FPM of the subject industrial process based on at least one of: (i) a flowsheet of the subject industrial process, (ii) a design of the subject industrial process, and (iii) operational parameters of the subject industrial process; and
generating the generated simulation data by performing the one or more simulations with variable values encompassing input variations and historical operation of the subject industrial process.

### Claim 12 (depends on 11)

The method of claim 11 wherein configuring the FPM of the subject industrial process includes at least one of:
building one or more simulation blocks and connections, according to the flowsheet of the subject industrial process; and
configuring one or more designs of the subject industrial process, or operational parameters of the subject industrial process, in a given simulation of the one or more simulations.

### Claim 13 (depends on 11)

The method of claim 11 wherein generating the generated simulation data by performing the one or more simulations includes:
running multiple simulations of the subject industrial process based on a plurality of combinations of input values and operating conditions; and
recording simulation output data from the multiple simulations as the generated simulation data.

### Claim 14

A computer-based system for creating a model for operating a real-world industrial process, the system comprising:
a processor; and
a memory with computer code instructions stored thereon, the processor and the memory, with the computer code instructions, being configured to cause the system to:
train a reduced order model (ROM) of a subject industrial process using simulation data generated from one or more simulations of the subject industrial process performed using a first-principles model (FPM) of the subject industrial process;
construct a sustainable dynamic reduced-order model (SDROM) of the subject industrial process by:
(i) dividing the trained ROM into a plurality of multiple-input single-output (MISO) sub-models;
(ii) partitioning each MISO sub-model into multiple component terms; and
(iii) inserting a gain factor into each partitioned component term to form a gain vector for each MISO sub-model; and
deploy the SDROM online to operate the subject industrial process.

### Claim 15 (depends on 14)

The system of claim 14 wherein, in deploying the SDROM online, the processor and the memory, with the computer code instructions, are further configured to cause the system to:
obtain operation data of the subject industrial process; and
monitor a given MISO sub-model of the plurality of MISO sub-models by:
(i) retrieving historical operation data from a plant history database;
(ii) filtering elements of the retrieved historical operation data with a dynamic filter attached to a given input of the given MISO sub-model;
(iii) wherein the gain vector formed for the given MISO sub-model is a first gain vector, estimating gain factors of a second gain vector of the given MISO sub-model based on the filtered elements of the retrieved historical operation data;
(iv) comparing the estimated gain factors of the second gain vector with inserted gain factors of the first gain vector to determine model performance metrics; and
(v) displaying, to users, a visual representation of the determined model performance metrics.

### Claim 16 (depends on 14)

The system of claim 14 wherein, in deploying the SDROM online, the processor and the memory, with the computer code instructions, are further configured to cause the system to:
adapt a given MISO sub-model of the plurality of MISO sub-models of the constructed SDROM with historical operation data of the subject industrial process by:
receiving allowance to perform adaptations to at least a part of the given MISO sub-model of the constructed SDROM from a user or an auto-system; and
updating the given MISO sub-model by calibrating gain factors of the given MISO sub-model.

### Claim 17 (depends on 16)

The system of claim 16 wherein, in updating the given MISO sub-model by calibrating gain factors of the given MISO sub-model, the processor and the memory, with the computer code instructions, are further configured to cause the system to:
retrieve the historical operation data, as a time-series, from a plant history database, for a window of time represented within the plant history database;
filter elements of the retrieved historical operation data with a dynamic filter attached to a given input of the given MISO sub-model;
feed the filtered elements of the retrieved historical operation data as inputs into the given MISO sub-model and predict an output time-series corresponding to the filtered elements;
compare the predicted output with the retrieved historical operation data of the subject industrial process in R 2 statistics;
adjust, according to the comparing, and subject to one or more suppression constraints, values of the gain factors for the given input of the given MISO sub-model, thereby calibrating the gain factors of the given MISO sub-model;
update the SDROM to include the given MISO sub-model with the calibrated gain factors; and
validate the updated SDROM with the testing dataset.

### Claim 18 (depends on 14)

The system of claim 14 wherein, in deploying the SDROM online, the processor and the memory, with the computer code instructions, are further configured to cause the system to:
receive a stream of real-time data of the subject industrial process;
filter elements of the received stream of real-time data with a dynamic filter attached to a given input of a given MISO sub-model;
predict an output of the given MISO sub-model and estimate local gains of the given MISO sub-model around an updated operating point determined from the filtered elements of the received stream of real-time data; and
determine optimal operation settings of the subject industrial process based on the predicted output and estimated local gains.

### Claim 19 (depends on 14)

The system of claim 14 wherein, in deploying the SDROM online, the processor and the memory, with the computer code instructions, are further configured to cause the system to:
determine optimal operation settings of the subject industrial process using the SDROM and an optimization objective function, the optimal operation settings including at least one of a temperature setpoint, a pressure setpoint, a flow rate setpoint, an operating throughput, and a transition path from a current operating state to the determined optimal operation settings; and
operate the subject industrial process in accordance with the determined optimal operation settings.

### Claim 20

A computer program product for creating a model for operating a real-world industrial process, the computer program product comprising:
a non-transitory computer-readable medium having computer-readable program instructions stored thereon, the instructions, when executed by a processor, causing the processor to:
train a reduced order model (ROM) of a subject industrial process using simulation data generated from one or more simulations of the subject industrial process performed using a first-principles model (FPM) of the subject industrial process;
construct a sustainable dynamic reduced-order model (SDROM) of the subject industrial process by:
(i) dividing the trained ROM into a plurality of multiple-input single-output (MISO) sub-models;
(ii) partitioning each MISO sub-model into multiple component terms; and
(iii) inserting a gain factor into each partitioned component term to form a gain vector for each MISO sub-model; and
deploy the SDROM online to operate the subject industrial process.
