Hybrid Biochemical Fermentation Model for Substrate Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modeling a chemical process, such as bio-fermentation, to predict the production of a chemical product is challenging due to complex interactions within the process, requiring accurate prediction of substrate concentrations and optimal conditions for efficient production.
Innovation Solution
A hybrid model combining a first-principles model and a data-driven model, specifically using a deep neural network, to simulate chemical reactions, estimate substrate concentrations, and set optimal conditions for the chemical process, thereby predicting future production results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a first-principles model is used to simulate chemical reactions, then the model is interpretable and based on physical laws, but the model accuracy is insufficient for complex biochemical processes
Solution Approach 1:
The patent combines a first-principles model with a data-driven deep neural network model to create a hybrid model. The first-principles model provides physical law-based interpretability while the neural network compensates for model inaccuracies by learning from operational data, thereby achieving both reliability and prediction accuracy simultaneously
Solution Approach 2:
The hybrid model functions as a composite modeling approach, integrating two different modeling paradigms (mechanistic and data-driven) into a unified framework that leverages the strengths of both approaches to overcome the limitations of using either model alone
2Measurement precision
If a data-driven model is used to simulate chemical reactions, then the model accuracy improves, but the model becomes a black box without interpretability
Solution Approach 1:
The patent merges a data-driven neural network model with an interpretable first-principles model. The neural network provides accurate predictions while the first-principles model maintains interpretability through its basis in physical laws, creating a hybrid system that achieves both accuracy and transparency
Solution Approach 2:
The first-principles model acts as an intermediary that bridges the gap between the black-box neural network and the physical reality, providing a mechanistic framework that explains the predictions while the neural network refines the accuracy based on empirical data
3Measurement precision
If complex interactions in biochemical fermentation are modeled, then the prediction accuracy improves, but the model complexity increases
Solution Approach 1:
The patent segments the modeling task into two distinct components: a first-principles model that handles the physical law-based aspects and a neural network that handles the complex non-linear interactions. This segmentation allows each component to focus on specific aspects of the system, reducing overall model complexity while maintaining prediction accuracy
Data Source
AI summary
A system for producing a product is disclosed. The system includes a production facility for producing a product using a chemical process involving chemical reactions, and an information processing device comprising a computer processor that simulates, using a hybrid model, the chemical reactions in the chemical process that produces the product to obtain a predicted output, wherein the hybrid model is a combination of a first-principles model and a data-driven model, determines, using an observer model, expected concentrations and levels of all substrates for the simulated process, sets derived optimal conditions for the chemical process based on the estimated concentrations and levels of all substrates, and predicts future production results based on a current status of the production facility.