Fermentation Culture Control Using Linear Models for Stable Output
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for stabilizing amino acid production by fermentation lack real-time control capabilities, as they require pre-set conditions and do not account for varying factors during the culture process, leading to suboptimal production results.
Innovation Solution
A control device and method that utilize a linear model to calculate and adjust culture conditions in real-time based on acquired data, determining the optimal conditions for future production by comparing predicted and target production amounts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-set culture conditions are used based on known information before culture start, then production stability can be maintained, but real-time control capability is lost
Solution Approach 1:
The patent implements a feedback mechanism where culture data is continuously acquired during the fermentation process, fed into a learning model, and used to update predictions of future production amounts. This enables real-time adjustment of culture conditions based on actual process deviations, resolving the contradiction between maintaining stability and achieving real-time control adaptability
Solution Approach 2:
The system performs self-optimization by automatically acquiring culture data, processing it through the learning model, predicting future production amounts, and determining optimal culture conditions without external intervention. This self-service capability enables continuous real-time adaptation while maintaining production stability
2Measurement precision
If black box learning models are used for real-time control, then prediction accuracy improves, but interpretability and understanding of control factors are lost
Solution Approach 1:
The patent transforms the black box learning model into an interpretable system by changing the parameter representation from opaque neural network weights to a linear model with explicit coefficients. These coefficients directly indicate the influence degree of each culture factor on production, maintaining prediction accuracy while enabling clear interpretation of control factors
Data Source
AI summary
A control device performs control of a culture condition in production of an organic compound by a fermentation method. The control device executes processing a plurality of times to acquire culture data, to calculate, using the acquired culture data, a plurality of candidates for a culture condition set in advance, and a linear model set in advance that outputs a production amount of an organic compound at a future time, the production amount at the future time for each of the candidates, to determine an optimum candidate out of the candidates using the calculated production amount at the future time for each of the candidates and a target production amount of the organic compound at the future time set in advance, and to change the culture condition to the determined candidate. The linear model and the target production amount are set for each time.