Novel Lactobacillus Fermentation Prediction Using Deep Learning

Abstract

In recent years, due to the vigorous development of artificial intelligence in various fields, many various applications have appeared. However, due to the biological uncertainty, only a few research apply artificial intelligence to manage the biological production process. The fermentation process of lactic acid bacteria has biologically uncertain, and the parameters in the fermentation process are difficult to set with fixed values to be automatically executed. Therefore, the current fermentation process is carried out manually. Due to the uncertainty in the production process, once human error occurs, it often causes hundreds of thousands or even millions dollars of losses. Therefore, if the fermentation effect can be improved, the subsequent production efficiency can be directly improved. In order to automate the fermentation process, in this project, we hope that by combining artificial intelligence (AI) with the background of lactic acid bacteria cultivation, the current complicated manual fermentation process can be transformed into automation as the goal of Industry 4.0. Based on the logs of the experiments of Lactobacillus fermentation, we use Long Shorten-Memory (LSTM) to predict the output amount of fermentation results. In the experimental results, we collects 9 trials of experimental results (4 case for over 3*109, 5 cases for approaching 3*109 and 7 cases for 0 output). And then, all the results are randomly separated into training and testing datasets for 20 different runs. The training dataset average accuracy of 20 runs is 100%. And the testing dataset average accuracy of 20 runs is 95%. Hence, according to the experimental results, we can know the proposed methods really can predicted the amount of the fermentation products.