Optimisation and control of fed-batch yeast production using q-learning

In this work, the optimal production of yeast with minimal production of ethanol in fed-batch yeast fermentation is investigated. Q-learning (QL) is a heuristic approach suggested for the process dynamic handling to achieve the multiobjective optimisation. The QL agent interacts with the fermentatio...

全面介紹

Saved in:
書目詳細資料
主要作者: Helen, Chuo Sin Ee
格式: Thesis
語言:English
English
出版: 2013
主題:
在線閱讀:https://eprints.ums.edu.my/id/eprint/41805/1/24%20PAGES.pdf
https://eprints.ums.edu.my/id/eprint/41805/2/FULLTEXT.pdf
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
實物特徵
總結:In this work, the optimal production of yeast with minimal production of ethanol in fed-batch yeast fermentation is investigated. Q-learning (QL) is a heuristic approach suggested for the process dynamic handling to achieve the multiobjective optimisation. The QL agent interacts with the fermentation environment will gain experience on the state transitions, which are represented by the change of substrate, yeast, oxygen and ethanol concentration and the system volume. In the present study, multistep action (MSA) has been implemented in consideration of the inborn process delay for the substrate feeding to take effect on the yeast growth. Parameter deviated model has been implemented in the QL to test the robustness of the algorithm besides to identify the process disturbance. From the result, QL was able to perform multiobjective decision making for the optimal substrate feeding profile. The final yeast production using QL-optimised feeding profile is 20.86% higher compare to the nominal exponential feeding (EF), and 19.59% higher compare to EF with process disturbance. To cater for the process disturbance, Q-learning with exploration (QLE) has been included in this work for online optimisation. QLE signifies the importance of exploration from time to time based on the developed “past experience” in Q-table to optimise the process. The performance of QLE in both nominal and disturbance cases yielded 51.00% and 46.87% higher yeast production than EF respectively, while maintaining low ethanol production. In a nutshell, QL is an alternative that can be considered to perform multiobjective optimisation in a frequently changing bioenvironment and suggest a substrate feeding profile that satisfied the process goal. The QLE can cope better with the process disturbance.