Plant Data Prediction Using Dynamic k-NN Similarity Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing plant prediction methods, particularly those using the k-NN algorithm, are prone to inaccurate predictions due to the reliance on elaborate models, which can be difficult for experts to construct, leading to poor performance and increased false alarms in plant management.
Innovation Solution
An apparatus and method that enhance prediction accuracy by employing a modified similarity calculation method and k optimization, using a plant modeling unit to generate a k-NN prediction model, a similarity measuring unit to determine similarity between data sets, and a k optimizing unit to set an optimum k value for generating prediction data, thereby improving prediction performance even with inaccurate models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a conventional k-NN algorithm is used for plant prediction, then the system can operate with a simple model structure, but the prediction accuracy deteriorates due to reliance on elaborate models that are difficult to construct
Solution Approach 1:
The patent transforms the prediction approach by changing from model-based parameters to similarity-based parameters. Instead of relying on complex model structures, the system calculates similarity between current plant data and historical data points, using similarity scores as the basis for prediction. This parameter transformation enables accurate predictions without requiring elaborate model construction.
Solution Approach 2:
The patent uses historical data copies (stored plant operation data) to predict current plant state rather than using a theoretical model. By copying and comparing actual historical operation data with current data, the system achieves accurate predictions while maintaining simple system structure, as it relies on empirical data rather than complex modeling.
2Measurement precision
If experts construct elaborate models to improve prediction accuracy, then the prediction performance improves, but the ease of operation deteriorates because experts are unable to always execute modeling
Solution Approach 1:
The patent enables the system to automatically perform prediction without requiring expert intervention for model construction. The system self-services by automatically calculating similarities between current and historical data, determining k values, and generating predictions. This eliminates the need for experts to manually execute modeling tasks while maintaining high prediction accuracy.
Solution Approach 2:
The patent performs preliminary data preparation by storing historical plant operation data in advance. This pre-collected data serves as the basis for future predictions, eliminating the need for experts to perform complex modeling at prediction time. The system simply compares current data with pre-stored historical data to generate accurate predictions.
3Device complexity
If a fixed k value is used in k-NN algorithm, then the device complexity is reduced, but the prediction accuracy deteriorates due to inability to adapt to varying data conditions
Solution Approach 1:
The patent makes the k value dynamic rather than fixed. The system automatically determines the optimal k value based on the similarity calculation results and data distribution. This dynamic adjustment of k allows the algorithm to adapt to varying data conditions, improving prediction accuracy while maintaining relatively simple system structure through automated parameter selection.
Solution Approach 2:
The patent implements feedback by using the similarity calculation results to inform the selection of k value. The system calculates similarities between current data and historical data points, then uses this similarity information to determine the appropriate k value for prediction. This feedback loop enables the system to adapt k to current data conditions, improving accuracy without requiring complex external control.
Data Source
AI summary
A method of predicting plant data in a system generates prediction data based on a plant prediction model and for detecting anomalies of the plant by comparing the prediction data with measurement data. The method can provide precise prediction data in a normal state even though the measurement data contains data in an anomalous state. Anomaly/fault prediction accuracy is enhanced by providing precise prediction data in the normal state. An apparatus using the method includes a plant modeling unit for generating a plant prediction model based on k-nearest neighbors (k-NN) by modeling a plant based on measurement data of multiple tags; and a prediction data generating unit for measuring similarity between the plant prediction model and the measurement data, determining a parameter k value based on the similarity, and generating plant normal state prediction data based on the determined parameter k value and the measured similarity.