Fast Multivariate Search for Unequal-Length Fed-Batch Fermentation

A huge amount of high-frequency on-line data are accumulated in the industrial fermentation. These data are used to evaluate the fermentation status through batch similarity search. For the fed-batch fermentation, the fermentation time varies between different batches. Meanwhile, multiple process variables are utilized for the similarity search, such as temperature, pH, aeration rate, etc. Therefore, the multivariate time series search problem should be solved for these unequal-length fermentation batches. In this study, a framework is proposed to reduce the frequency of computationally expensive distance calculations between multivariate time series pairs. The validation result indicates that the proposed method significantly reduces the similarity search time, at the cost of an acceptable loss of accuracy. Therefore, it is feasible for the multivariate search of unequal-length fed-batch fermentation.