MPC Model Adaptation Using Automated Data Screening and Repair

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Multivariable Predictive Control (MPC) performance degrades over time due to process changes, requiring frequent model updates and data screening, which is inefficient and labor-intensive, especially in identifying and excluding unsuitable data for model quality estimation and identification.

Innovation Solution

An automated four-tier data screening and selection system that detects and excludes unsuitable data segments, repairs bad data samples using internal MISO models, and maximizes data usage by interpolating missing values, thereby minimizing data loss and improving model quality estimation and adaptation in MPC applications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual data screening and selection is performed by control engineers, then data quality for model identification can be ensured, but the process becomes labor-intensive and time-consuming

Engineering Contradiction:
Improvedata qualityVSAvoidscreening time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs automated data screening and selection using algorithms that independently evaluate data quality metrics, identify bad data segments, and select suitable data for model identification without requiring manual engineer intervention. The automated system serves itself by implementing the entire data screening workflow through computational methods.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The manual mechanical process of engineers visually inspecting time series plots and marking bad data is replaced with automated computational algorithms that programmatically evaluate data quality, detect anomalies, and select appropriate data segments based on predefined criteria and metrics.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual data screening is performed to exclude bad data segments, then model identification reliability is improved, but productivity decreases due to hours to days of engineering work

Engineering Contradiction:
Improvemodel identification reliabilityVSAvoiddata screening efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system autonomously performs data screening, evaluation, and selection tasks that were previously requiring engineer expertise and manual effort. The automated algorithms independently identify bad data segments and select suitable data for model identification, eliminating the need for manual intervention while maintaining reliability.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system changes the approach from manual visual inspection to automated computational evaluation by introducing new parameters such as data quality metrics, anomaly detection thresholds, and automated selection criteria that enable rapid and reliable data screening without manual effort.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If bad data segments are excluded from model identification, then data quality is improved, but data loss increases reducing available data for modeling

Engineering Contradiction:
Improvedata qualityVSAvoiddata loss
Core Design Contradiction:
Measurement precisionVSLoss of substance

Solution Approach 1:

The system extracts and excludes only the specific bad data segments that meet predefined quality criteria and anomaly detection thresholds, rather than excluding large portions of data. This targeted extraction approach removes harmful data while preserving as much useful data as possible for model identification.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system applies different quality assessments to different segments of the data, identifying and excluding only the local bad segments while preserving the quality of the surrounding good data. This localized approach ensures that data exclusion is precise and minimizes overall data loss.

Inventive Principle:
Principle #3Local quality

4Ease of manufacture

If conventional data screening approaches are used, then implementation is simple, but they are not suitable for frequent online model adaptation runs

Engineering Contradiction:
Improveimplementation simplicityVSAvoidonline adaptation suitability
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The system transitions from static manual screening processes to dynamic automated screening that can be executed frequently and rapidly for online model adaptation. The automated algorithms enable the system to adapt to changing process conditions in real-time by performing repeated data screening and model identification runs as needed.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system introduces new parameters and computational methods that enable frequent online execution, including automated data quality metrics, rapid anomaly detection algorithms, and efficient data selection criteria that can be evaluated quickly for each model adaptation run.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2825920B1Apparatus and method for automated data selection in model identification and adaptation in multivariable process control
Publication Date: 2021.01.20 ASPEN TECHNOLOGY
  • EP2825920B1 patent drawingFigure 1
  • EP2825920B1 patent drawingFigure 2
  • EP2825920B1 patent drawingFigure 3

AI summary

A computer-based apparatus and method for automated data screening and selection in model identification and model adaptation in multivariable process control is disclosed. Data sample status information, PID control loop associations and internally built MISO (Multi-input, Single-output) predictive models are employed to automatically screen individual time-series of data, and based on various criteria bad data is automatically identified and marked for removal. The resulting plant step test/operational data is also repaired by interpolated replacement values substituted for certain removed bad data that satisfy some conditions. Computer implemented data point interconnection and adjustment techniques are provided to guarantee smooth/continuous replacement values.