Plant Tag Correlation Modeling for Early Abnormality Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing plant abnormality prediction systems face inaccuracies due to improper selection of input and output tags, leading to ineffective modeling and prediction, especially when the relationship between tags is not well understood.
Innovation Solution
A system that preprocesses data, analyzes correlation between tags to determine independent tags, and uses ensemble-based prediction methods to output alarm signals based on differences between measured and predicted data, enabling accurate predictions even with insufficient understanding of the system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If tags are selected based on similar variations over time, then the model construction is simplified, but the prediction accuracy deteriorates due to lack of causal relationship
Solution Approach 1:
The system performs preliminary correlation analysis and clustering operations on tags before model construction. By pre-identifying independent tags through correlation coefficient calculation and clustering, the system prepares high-quality input features in advance, ensuring both ease of model construction and high prediction accuracy without requiring deep system understanding.
Solution Approach 2:
The system introduces correlation analysis and clustering operations as intermediary steps between raw tag selection and model construction. These intermediary processes filter and organize tags based on statistical relationships, transforming arbitrary tag selections into scientifically validated independent tag sets that improve prediction accuracy while maintaining construction simplicity.
2Productivity
If tags without causation are used to create the model, then the model construction is faster, but the abnormality prediction accuracy deteriorates
Solution Approach 1:
The system performs preliminary correlation analysis and clustering operations on tags before model construction. By pre-identifying independent tags through correlation coefficient calculation and clustering, the system prepares high-quality input features in advance, ensuring both ease of model construction and high prediction accuracy without requiring deep system understanding.
Solution Approach 2:
The system replaces the need for expert mechanical understanding of causal relationships with automated statistical methods. By using correlation coefficient calculation and clustering algorithms, the system objectively identifies independent tags without requiring domain expertise, maintaining fast construction speed while improving prediction accuracy through data-driven tag selection.
3Measurement precision
If correlation analysis and clustering operations are performed on all tags, then the prediction accuracy is improved, but the system complexity increases
Solution Approach 1:
The system segments the tag set into independent and dependent tags through correlation analysis and clustering operations. By dividing tags into distinct groups based on their correlation relationships, the system reduces the dimensionality of the problem and identifies a manageable subset of independent tags that can be used for modeling, balancing accuracy improvement with system complexity management.
4Measurement precision
If multiple prediction algorithms are applied to different tag types, then the prediction accuracy is improved, but the operational complexity increases
Solution Approach 1:
The system applies different prediction algorithms to different types of tags based on their specific characteristics. By tailoring the prediction approach to each tag type (independent vs. dependent tags), the system optimizes prediction accuracy for each specific case while maintaining a structured framework that manages operational complexity through clear classification and algorithm assignment.
Data Source
AI summary
A system and method predict whether a plant is abnormal by modeling a relationship equation between tags based on a correlation between the tags, applicable even if modeling is executed without understanding a target to abnormality determination, and implements internal early alarm logic based on a difference between measured data and predicted data over time. The plant abnormality prediction system includes a modeling information output unit including a pre-processing part for pre-processing past data received for a plurality of tags, a correlation analysis part for receiving the pre-processed data for each tag to determine an independent tag among the plurality of tags based on correlation coefficients for any two tags, and a modeling part for generating a relationship equation between the tags by using outputs of the pre-processing part and the correlation analysis part; and a prediction unit for calculating estimated data for the tag based on the relationship equation.