Facility Abnormality Prediction Model Using Genetic Algorithm Sensor Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to predict facility abnormalities, especially when data exhibits nonlinear relationships, leading to delayed diagnosis and low reliability due to the limitations of conventional statistical methods.
Innovation Solution
A system and method for establishing a facility abnormality prediction model using a genetic algorithm and support vector machine, which includes a data receiver, abnormality notification time predictor, optimal sensor combination calculator, and facility abnormality prediction model generator to identify and pre-notify potential malfunctions based on sensor data, even in nonlinear relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional statistical methods (multivariate, SPC, PCA) are used to analyze facility data, then the analysis is simple and easy to implement, but the methods fail when data exhibits nonlinear relationships, leading to inability to predict facility abnormality
Solution Approach 1:
The patent transforms the analysis approach by changing from linear statistical parameters to nonlinear machine learning parameters. Support vector machines with kernel functions (e.g., RBF kernel) enable the model to handle nonlinear relationships by mapping data to higher-dimensional feature spaces, while genetic algorithms optimize the model parameters adaptively, resolving the contradiction between handling nonlinear data and maintaining method simplicity.
2Reliability
If non-parametric methodologies are used to predict facility abnormality, then the prediction can be made for nonlinear data, but much time is taken to diagnose the abnormality and high-reliability results are difficult to obtain
Solution Approach 1:
The patent performs preliminary actions by pre-training the support vector machine model with historical facility data and pre-optimizing parameters using genetic algorithms during the offline phase. This preliminary preparation enables the model to quickly and reliably predict facility abnormalities in real-time operation, reducing diagnosis time while maintaining high reliability.
Solution Approach 2:
The patent implements feedback mechanisms where prediction results are continuously monitored and used to refine the model. The genetic algorithm optimizes parameters based on feedback from prediction accuracy, and the system adjusts to new data patterns, improving both reliability and efficiency over time through iterative learning.
3Measurement precision
If all sensor data are used for facility abnormality prediction, then comprehensive information is available, but the complexity of data processing increases and optimal sensor combinations are difficult to identify
Solution Approach 1:
The patent extracts only the most relevant sensor data for abnormality prediction by using genetic algorithms to evaluate and select optimal sensor combinations. The system identifies and extracts key features from the full sensor dataset that have the highest predictive value, removing redundant information and reducing processing complexity while maintaining detection precision.
Data Source
AI summary
A facility abnormality prediction model generation system includes: a data receiver receiving data of sensors of a facility previously obtained during an operation of the facility; an abnormality notification time predictor detecting a malfunction time of a malfunction of the facility based on the data of the sensors and determining an abnormality notification time for pre-notification of the malfunction of the facility based on the detected malfunction time; an optimal sensor combination calculator generating a chromosome based on the data of the sensors and performing a genetic algorithm using the generated chromosome to calculate an optimal sensor combination which is a combination of sensor data related to the determined abnormality notification time; and a facility abnormality prediction model generator generating a facility abnormality prediction model for the pre-notification of the malfunction of the facility, based on the optimal sensor combination.