Process Value Prediction Model With Virtual Error Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing data-driven simulators in plant process control face challenges in preventing the propagation of prediction errors, especially in multi-input multi-output non-linear systems, leading to reduced prediction accuracy and generalization performance.

Innovation Solution

A data-driven simulator is generated by adding virtual error data to input samples and training a machine learning model to predict process values, using time-series data and input-output data sets to mitigate the impact of previous errors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If a dynamic prediction model is trained using machine learning with operation data, then the simulator can be easily generated without domain knowledge and adapts to linear and nonlinear processes, but prediction errors propagate successively over time reducing accuracy

Engineering Contradiction:
Improveease of model generationVSAvoidprediction accuracy
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent applies preliminary action by adding virtual error data to the training dataset before training the dynamic prediction model. This pre-processing step prepares the model to handle errors by exposing it to error conditions during training, thereby preventing error propagation during actual simulation operations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements beforehand cushioning by incorporating virtual error data into the training process. This creates a buffer or cushion against future errors by teaching the model to tolerate and correct errors, thereby cushioning the impact of prediction errors that would otherwise propagate through the simulation.

Inventive Principle:
Principle #11Beforehand cushioning (Prior cushioning)

2Reliability

If ensemble prediction with multiple models is used to reduce inaccurate prediction results, then prediction robustness improves, but the complexity of the simulator increases

Engineering Contradiction:
Improveprediction robustnessVSAvoidsimulator complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by modifying the training dataset to include virtual error data rather than changing the model architecture or ensembling multiple models. This approach improves prediction robustness by changing the training parameters (data characteristics) rather than changing the system structure, thereby avoiding increased complexity.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If more training data is used to improve prediction accuracy, then the model performs better in accurate domains, but the amount of data required increases time and resource consumption

Engineering Contradiction:
Improveprediction accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent introduces virtual error data as an intermediary element in the training process. This intermediary component allows the model to learn error correction without requiring additional real-world data, thereby improving prediction accuracy without proportionally increasing training time and resource consumption.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4600766A1Information processing apparatus, generation method, and generation program
Publication Date: 2025.08.13 YOKOGAWA ELECTRIC CORP
  • EP4600766A1 patent drawingFigure 1
  • EP4600766A1 patent drawingFigure 2
  • EP4600766A1 patent drawingFigure 3

AI summary

An information processing apparatus 10 includes an acquisition unit configured to acquire time-series data related to a process value that represents an operation amount with respect to a process and a state of the process, a generation unit configured to generate an input-output data set in which an operation amount and a process value at a first time are adopted as an input sample and a process value at a second time subsequent to the first time is adopted as an output sample, an adding unit configured to add virtual error data to each process value of the input sample, and a training unit configured to train a machine learning model that outputs a process value at the second time while adopting the operation amount and the process value at the first time as input, by using the input-output data set to which the virtual error data is added.