Method and system for real time trajectory optimization

Patent metadata

Field	Value
Publication number	US-2023104214-A1
Application number	US-202117760297-A
Country	US
Kind code	A1
Filing date	Apr 9, 2021
Priority date	Apr 9, 2020
Publication date	Apr 6, 2023
Grant date	—

Abstract

Trajectory optimization is process of designing a trajectory of operating variables that optimizes measure of performance while satisfying a set of constraints, when the system moves from one state to another. It is very necessary to achieve optimization in real time. A system and method for real-time trajectory optimization has been provided. The trajectory optimization of a process can be performed in any dynamical automated system. The system is configured to optimize the trajectory in both online and offline mode. In the online mode, the system optimizes the trajectory of the process in real-time. The system has the ability to handle both machine learning and deep learning based time series models along with first principles based models represented by ordinary/partial differential equation or differential algebraic equation based dynamic models of the process to estimate process variables given the disturbance profile and the actuation profile of manipulated variables.

First claim

1 . A processor implemented method for real time trajectory optimization of a process, the method comprising:
receiving, via a data receiver, storage and transmitter unit, a plurality of data from one or more sources as an input data; preprocessing and integrating, via one or more hardware processors, the input data; classifying, via the one or more hardware processors, the preprocessed and integrated input data among one of a set of disturbance variables, a set of process variables, a set of manipulated variables, a set of material properties, and a set of design parameters, wherein the classification is stored in a variable information library; forecasting, via the one or more hardware processors, an expected profile of each of the set of disturbance variables using respective models; fetching, via the one or more hardware processors, most appropriate actuation profiles of the set of manipulated variables corresponding to a trajectory optimization model from a knowledgebase, based on a predefined condition for an applicable set of design parameters, the set of material properties, and expected profiles of the set of disturbance variables; implementing, the via one or more hardware processors, the most appropriate actuation profiles of the set of manipulated variables by informing a set of actuators, wherein the set of actuators are part of the process; predicting, via the one or more hardware processors, the expected profile of each of the set of process variables using their respective models; monitoring, via the one or more hardware processors, the set of disturbance variables, the set of process variables, objectives, and constraints of the process in real time to observe one or more changes; adjusting, via the one or more hardware processors, constituents of a trajectory optimization model in response to the one or more changes observed while monitoring; re-estimating, via the one or more hardware processors, the actuation profiles of manipulated variables using the adjusted constituents; and implementing, via the one or more hardware processors, the re-estimated actuation profiles of manipulated variables by informing the set of actuators.

2 . The method according to claim 1 , wherein the expected profile of the set of disturbance variables is forecasted using one of the current data or by making predictions using models of set of disturbance variables present in a model database.

3 . The method according to claim 1 , wherein the most appropriate actuation profiles of the set of manipulated variables is identified by:
calculating similarity scores between the forecasted expected profiles of the set of disturbance variables and each of the respective profiles already stored in the knowledgebase for a similar set of design parameters and the set of material properties, and selecting the most appropriate actuation profiles for which the calculated similarity score is found to be maximum.

4 . The method according to claim 3 , wherein similarity score can be calculated in terms of distance measure comprising one of Euclidean distance or Mahalanobis distance, or any similar distance measure or a weighted sum of distance measures computed using different distance estimation techniques

5 . The method according to claim 1 , where in monitoring can be performed by
computing the difference between the forecasted expected profiles and the measured profiles of the set of disturbance variables as per the recent data; computing the difference between the expected profiles and the currently measured profiles of the set of process variables with the former estimated by performing model predictions on the recent data; and checking whether all the constraints are satisfied and quantifying the amount by which each constraint is violated, in case if constraints are not satisfied.

6 . The method according to claim 5 , wherein the computed difference can represent an instantaneous difference or an integrated difference.

7 . The method according to claim 1 , wherein adjustment of the relevant constituents of the trajectory optimization model comprises at least one of:
updating the forecast of the set of disturbance variables using the recent data or by updating the models of the set of disturbance variables if the difference between the expected profile and actual profile of disturbance variables is more than a first threshold; updating the models of the set of process variables if the difference between expected profile and actual profile of the set of process variables is more than a second threshold; and relaxing the limits on the constraints if constraints are not satisfied and if there is a scope for relaxing the limits on the constraints;

8 . The method according to claim 7 , wherein the models are updated by one or more of tuning hyper-parameters of machine or deep learning models or by tuning the parameters of the first-principles based models.

9 . The method according to claim 1 , wherein re-estimation of the actuation profiles can be performed by one of the following methods depending on the limitations on the computational time and actuation requirements of relevant sensors:
solving the trajectory optimization model with adjusted constituents for a relatively short horizon; and solving the trajectory optimization model for the remaining horizon with adjusted constituents and the most recent data from sensors.

10 . The method according to claim 1 , wherein one or more sources of data comprises one or more of a distributed control system (DCS), a supervisory control and data acquisition (SCADA) system, a laboratory information management system (LIMS), an enterprise resource planning (ERP) system, a manufacturing execution system (MES), a manufacturing operations management (MOM) system, or a plurality of sensors.

11 . The method according to claim 1 , wherein the step of data preprocessing further comprises:
applying an outlier detection and removal method on the integrated dataset to identify and remove unreliable data and to identify any anomalous behavior of the system; detecting the functioning of each of the plurality of sensors using sensor monitoring method and historical performance data of the sensor; and performing data imputation using a multivariate imputation method for the given set of data to fill in the missing data.

12 . The method according to claim 1 , wherein the models for the set of process variables include one or more of machine learning models, deep learning models, and the first principles based models.

13 . The method according to claim 1 , wherein the models for the set of disturbance variables include one or more auto-regressive machine learning or deep learning models.

14 . A processor implemented method for trajectory optimization of a process in an offline mode, the method comprising:
receiving a plurality of historical data from one or more sources as an input data; preprocessing and integrating, via one or more hardware processors, the input data; classifying, via one or more hardware processors, the preprocessed input data among a set of disturbance variables, a set of process variables, a set of manipulated variables, a set of design parameters, and a set of material properties corresponding to the process, wherein the classification is stored in a variable information library; updating, via one or more hardware processors, existing models or creating new models of the set of process variables and the set of disturbance variables corresponding to pre-defined objectives and constraints, wherein the updated or created models are stored in a model database, wherein each of

Assignees

• Tata Consultancy Services Ltd

Inventors

• Pareek Aditya
• Masampally Vishnu Swaroopji
• Runkana Venkataramana

Classifications

• G05B13/045

  using a perturbation signal · CPC title

• G06F18/27

  Regression, e.g. linear or logistic regression · CPC title

• G06Q10/04 Primary

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G05B13/028 Primary

  using expert systems only · CPC title