Closed-Loop Process Controller Identification Using Noise Model Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Standard closed-loop identification techniques often result in biased or inaccurate model parameter estimates for industrial process controllers due to insufficient knowledge of true process and noise model structures, particularly when using direct identification methods without external excitation.
Innovation Solution
The proposed solution involves generating noise models using high-order ARX identification and filtering closed-loop data with an inverse noise model, followed by output-error (OE) model identification to estimate process model parameters, thereby reducing bias and enabling accurate model parameter estimation even without external excitation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If direct identification methods are used without external excitation, then the industrial process controller can operate continuously without offline experiments, but the model parameter estimates become biased or inaccurate
Solution Approach 1:
The patent introduces an auxiliary noise model as an intermediary component that mediates between the process model and the measured data. This noise model captures the statistical characteristics of measurement errors and process disturbances, allowing the identification algorithm to distinguish between true process dynamics and noise effects, thereby enabling accurate parameter estimation from closed-loop data without external excitation
Solution Approach 2:
The patent transforms the identification problem by changing the parameterization approach - instead of directly identifying process parameters from raw closed-loop data, the method first identifies noise model parameters and then uses these to correct the data before final parameter estimation. This two-stage parameter identification approach resolves the bias issue while maintaining continuous operation
2Measurement precision
If noise models are generated and data filtering is applied, then unbiased model parameter estimates are achieved, but the complexity of the identification process increases
Solution Approach 1:
The patent segments the identification process into distinct stages: first identifying the noise model structure and parameters, then using this information to filter and correct the closed-loop data, and finally performing the process model parameter estimation. This segmentation makes the complex overall task more manageable and computationally efficient
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A method includes obtaining (402) closed-loop data associated with operation of an industrial process controller (106, 204), where the industrial process controller is configured to control at least part of an industrial process using at least one model (144, 230). The method also includes generating (404) at least one noise model associated with the industrial process controller using at least some of the closed-loop data. The method further includes filtering (406) the closed-loop data based on the at least one noise model. In addition, the method includes generating (408) one or more model parameters for the industrial process controller using the filtered closed-loop data.