Regression Model Time-Lag Control for Process Variable Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for constructing regression models in plant management struggle to control the number of time lags associated with analysis target data, making it difficult to accurately predict process variables.
Innovation Solution
An information processing apparatus with a setting unit, selection unit, and determination unit is used to set and control the number of time lags in a regression model by selecting appropriate time-lag candidates for explanatory variables and determining regularization parameters based on regularization paths, ensuring the number of time lags is limited, thus constructing a regression model that predicts process variables effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automatic extraction of a large amount of features is performed with a penalized regression model to construct a regression model, then the model can predict process variables, but the number of time lags associated with each piece of analysis target data cannot be controlled
Solution Approach 1:
The patent applies dynamics by making the time lag selection flexible and adjustable rather than fixed. The system dynamically determines the number of time lags based on the regularization parameter λ, allowing the model to adaptively select the appropriate number of time lags for each analysis target data. This resolves the contradiction by enabling both accurate prediction (through optimal time lag selection) and operational control (through adjustable time lag numbers) simultaneously.
Solution Approach 2:
The patent changes the parameter λ (regularization parameter) to control the number of time lags in the regression model. By adjusting this parameter, the system can control which time-lagged features are selected and how many time lags are associated with each analysis target data. This parameter change approach enables precise control over time lag numbers while maintaining prediction accuracy, directly resolving the technical contradiction.
2Measurement precision
If the number of time lags is increased to improve prediction accuracy, then the model performance improves, but the model complexity increases
Solution Approach 1:
The patent applies partial action by selecting only the necessary number of time lags for each analysis target data rather than using all available time lags. Through the penalized regression model with parameter λ, the system performs partial selection of time-lagged features, including only those that contribute meaningfully to prediction accuracy. This reduces model complexity while maintaining prediction performance by avoiding unnecessary time lag components.
Solution Approach 2:
By changing the regularization parameter λ, the system controls the degree of model complexity. When λ is adjusted, it automatically determines the optimal number of time lags to include, balancing prediction accuracy against model complexity. This parameter-driven approach resolves the contradiction by enabling the model to adapt its complexity level based on the specific requirements of each prediction task.
Data Source
AI summary
According to an embodiment, an information processing apparatus is configured to set a candidate for a time lag until analysis target data including at least one of a measurement item and a setting item for use in control of a process controller affects an objective variable, and a time-lag number allowed in a regression model; select, as a candidate for an explanatory variable, at least one of the measurement item measured at a time corresponding to the candidate for the time lag and the setting item set at the time; and determine a regularization parameter of the regression model such that a number of the time lag is equal to or less than the time-lag number, based on a regularization path indicating transition of a regression coefficient for the candidate for the explanatory variable, the regression coefficient varying in accordance with a value of the regularization parameter.