Prediction Method of Biological Fermentation Data Based on Deep Neural Network

Abstract

This paper proposes a (Takagi-Sugeno-Kang) TSK fuzzy regression model that based on self-supervised learning and deep autoencoder to predict and monitor the real-time concentration of each ingredient in the fermentation process. The entire model consists of the following steps: obtaining and preprocessing sample spectral data to obtain a training set; using the training set to train a self-supervised feature extraction network model to optimize the parameters of the feature extraction network model; training the autoencoder network model to establish a dimensionality reduction model by using the feature-extracted data; performing TSK fuzzy regression on the data selected by the dimensionality reduction model to establish a concentration prediction model; inputting the spectral data of the solution to be tested to predict the concentration of the solution. Combined with the deep autoencoder feature extraction method of self-supervised learning, our model can not only construct a more complex nonlinear map than the traditional principal component analysis (PCA), but also ensure that the extracted features have semantic information that is beneficial to the subsequent regression prediction method. Combined with TSK regression prediction, our model can avoid the problem of excessive spectral data dimension and redundant information, and can give accurate and interpretable results.