Hybrid Plant Models for Chemical Process Control With Limited Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chemical plants face challenges in accurately determining physical-chemical values, especially with limited historic data, and often lack well-fitting physical models, making it difficult to monitor and control processes efficiently, particularly for new plants or those producing specialty products.
Innovation Solution
A computer-implemented method using a hybrid plant model that combines a mechanistic model with a data-driven model, where the data-driven model is trained with historical data to determine physical-chemical parameters, allowing for accurate monitoring and control with minimal resources and quick adjustments to maintain high product yield and minimize side products and emissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a data-driven model is used to determine physical-chemical parameters, then measurement precision is improved, but the model requires sufficient historic data for training which is often unavailable
Solution Approach 1:
The patent combines a data-driven model (neural network) with a mechanistic model (physical-chemical equations) into a hybrid model. The neural network predicts key parameters (temperature, pressure, concentrations) which are then used as inputs to the mechanistic model to calculate physical-chemical values. This merging allows the system to achieve accurate predictions with limited historic data by leveraging both data-driven patterns and physical laws.
2Measurement precision
If a hybrid model combining neural network and physical model is used, then determination accuracy is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the essential physical-chemical equations needed for calculating specific parameters of interest (e.g., conversion rates, selectivity) rather than implementing a complete detailed mechanistic model. The neural network handles the complex pattern recognition, while the extracted physical equations provide the necessary theoretical framework. This extraction reduces overall model complexity while maintaining accuracy.
3Measurement precision
If more historic data is collected for training, then model accuracy is improved, but loss of time occurs during data collection and model training
Solution Approach 1:
The patent performs preliminary action by collecting and utilizing available historic data from process control systems before the model is deployed. The neural network is trained offline using this pre-collected data, and once trained, the model can operate in real-time without requiring additional data collection. This preliminary training phase separates the time-consuming data preparation from the real-time operation, minimizing ongoing time loss.
Data Source
AI summary
The present invention relates to a computer-implemented method for monitoring and/or controlling a chemical plant. Specifically, the present invention relates to a computer-implemented method for monitoring and/or controlling a physical-chemical process in a chemical plant comprising:(a) receiving sensor data related to the physical-chemical process,(b) determining at least one physical-chemical parameter by providing the sensor data to a plant model, wherein the plant model comprisesa mechanistic model containing at least two equations each representing a part of the physical-chemical process anda data-driven model associated to the mechanistic model, wherein the total number of scalars as output parameters from the data-driven model is lower than the number of equations of the mechanistic model, and(c) outputting the at least one physical-chemical parameter determined by the plant model.