Sugar + oxygen yeast + CO 2 + alcohol

Size: px
Start display at page:

Download "Sugar + oxygen yeast + CO 2 + alcohol"

Transcription

1 1. Introduction: Yeast Fermentation Sugar + oxygen yeast + CO 2 + alcohol Saccharomyces cerevisiae (aka baker s yeast) - the most studied strain As the benchmark What we want? ---MORE YEAST! Enhance the supply of raw material Sustainable environment and production Future potential 1

2 1. Introduction: Fed batch operation Substrate (Food) AIM: Maximize yeast Minimize alcohol Yeast (biomass) Fermentor Product (after some time) WHY Fed-batch? (1) Dynamic growth behavior (2) Quality (3) Resource saving 2

3 2. Objective To introduce and apply the learning algorithm (Q-Learning) in fermentation system to determine the optimal substrate feeding profile To use the Q-Learning optimized profile as predetermined profile for fermentation system Motivations: Unsupervised learning ability Reduction of human input Applicable to highly dynamic system Multi-objectives algorithm 3

4 3. Methodology Observation and Learning Module QL can be described as the following equation: Q-learning (QL) is an unsupervised learning algorithm Works by learning state-action function that gives expected reward of taking an action in a specific state with a fixed policy. Where s = state s = next state a = action a = next action Q = current Q-valueQ = Expected Q-value in next state t = time α = learning rate R = reward γ = discount factor 4

5 3. Methodology In this research, QL is used to: Observe and learn the effect of input feed flow rate on the various product concentration Maximum reward is chosen for the feed which gives maximum yeast and less alcohol Determine the optimal feed profile by total maximum reward. 5

6 4. Simulation Feed in S KINETIC Material Balances DYNAMIC Product (after some time) Complex process dynamics Substrate and product inhibition 6

7 5. Results and Discussion C on c ent ration of G luc os e (g/l ), E th anol (g/l) yeast glucose ethanol C onc e ntrat ion of Y ea s t (g/ L) C o n c e n t ra tio n o f G lu c o s e (g /L ), E th a n o l (g /L ) yeast glucose ethanol C o n c e n tra t io n o f Y e a s t (g / L ) Nominal exponential feeding (EF) performances: F = 5e.5t Q-learning (QL) optimized feeding performances Comparison of substrate (glucose), undesirable products (ethanol) and biomass (yeast) concentrations for exponential feeding and Q-learning feeding strategy. 7

8 5. Results and Discussion 3 25 Feed flow rate (EF) Feed flow rate (QL) Feed Flow Rate (L/h) QL manage to optimize feed flow rate to the maximum before inhibitions and overflow metabolism which triggers ethanol production happens. 8

9 5. Results and Discussion Concentration of Glucose (g/l), Ethanol (g/l) glucose (EF) glucose (QL) Concentration of Glucose (g/l), Ethanol (g/l) yeast (EF) yeast (QL) Concentration of Glucose (g/l), Ethanol (g/l) ethanol (EF) ethanol (QL) QL-suggested feeding profile is able to achieve multi-objective purposes: reduce substrate usage, maximize yeast production and minimize ethanol. 9

10 6. Conclusion Based on the performances, Q-learning is able to: suggest satisfying optimal profile as predetermined feeding strategy for yeast fermentation based on its objectives; act in dynamic-complex system; suggest an alternative algorithm for biosystem 1