Blockwise Recursive PLS for Delayed Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing modeling techniques for real-world chemical processes, such as those in crude oil refineries, struggle with nonlinear changes and noisy inputs, leading to computational deficiencies and challenges in determining when a new model update is necessary.
Innovation Solution
A blockwise recursive partial least squares (PLS) technique is employed, using a condition number and forgetting factor to detect changes in model outputs, assess new models with prediction metrics, and replace the existing model in real-time if performance improves.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If recursive least squares (RLS) and ordinary least squares (OLS) are used for modeling, then modeling capability is provided, but computational deficiencies occur with ill-conditioned matrices, inferior quality data, and highly correlated inputs
Solution Approach 1:
The patent transforms the modeling approach by changing from traditional RLS/OLS to Partial Least Squares (PLS) regression, which fundamentally alters how the model handles correlated and noisy data. PLS changes the parameter estimation method to be more robust against computational deficiencies while maintaining modeling capability.
Solution Approach 2:
The patent employs a virtual model that replicates the complex chemical process behavior without directly modeling the physical complexity. This virtual copy captures the essential relationships between inputs and outputs, avoiding the need to computationally handle the full complexity of the actual process.
2Reliability
If real-world chemical processes are modeled, then process understanding is achieved, but nonlinear changes and noisy inputs make the processes challenging to model
Solution Approach 1:
The patent applies PLS regression which changes the approach from direct physical modeling to statistical relationship modeling. This parameter change allows the model to handle nonlinearities and noise by focusing on the correlation structure between inputs and outputs rather than attempting to model the physical mechanisms directly.
Solution Approach 2:
The patent creates a virtual representation that copies the input-output relationships of the chemical process without replicating the physical nonlinearities and noise. This virtual model captures the essential behavior patterns while filtering out the challenging aspects of real-world process complexity.
3Measurement precision
If continuous model updating is performed, then model accuracy is maintained, but computational resources are consumed and model stability is compromised
Solution Approach 1:
The patent implements periodic model updating rather than continuous updating. The model is updated at specific intervals or when certain conditions are met, which maintains accuracy while significantly reducing computational resource consumption compared to continuous updates.
Solution Approach 2:
The patent performs preliminary model validation and selection before implementing updates. By pre-assessing whether an update is necessary and preparing multiple candidate models in advance, the system avoids unnecessary computational resources while maintaining accuracy.
4Adaptability or versatility
If model updates are triggered frequently, then adaptability to changing processes is improved, but system stability and operational continuity are compromised
Solution Approach 1:
The patent implements a feedback mechanism that continuously monitors process performance and model accuracy. This feedback allows the system to adapt to changing processes by triggering updates only when actual performance degradation is detected, thereby maintaining adaptability while preserving system stability during normal operation.
Solution Approach 2:
The patent makes the model updating process dynamic rather than static. The system can switch between multiple models based on real-time performance conditions, allowing it to adapt to changing processes while maintaining stability by selecting the most appropriate model for current operating conditions.
Data Source
AI summary
Most chemical processes are nonlinear in nature and are time-varying in nature. Some processes are fast changing and some are slow changing in nature. As the operating conditions change, it is important that the model and its parameters be updated to realize the benefits of model-based control. Systems and methods are provided for updating a model, the method including detecting a change in an output of the model responsive to providing a set of inputs to the model, in response to detecting the change in the output, triggering a new model detection flag using a blockwise recursive partial least squares (RPLS) algorithm. The model is further strengthened by using a delayed just-in-time update.