The invention discloses a novel method for carrying out real-time fault monitoring on a
penicillin fermentation process. The method comprises two stages of off-line modeling and on-line monitoring. The off-line modeling comprises the following steps: firstly,
processing three-dimensional data of a
fermentation process; then respectively calculating a
Laplacian matrix of the data of each
fermentation batch to represent
local structure information of the data in each batch; and finally, carrying out modeling by using a
noise reduction Laplace automatic
encoder (DLAE) to construct monitoring statistics, and determining a control limit by using a
kernel density estimation method. The on-line monitoring comprises the following steps:
processing newly collected data according to a model, calculating the statistical magnitude of the data, comparing the statistical magnitude with a control limit, and judging whether the fermentation process runs normally or not. According to the method, the
local structure of data in batches can be effectively utilized, and meanwhile, the training cost and the hardware requirement of the Laplace automatic
encoder are reduced. Meanwhile, the robustness of the model is enhanced by adopting a
noise reduction training mode, and the fault monitoring accuracy is relatively high.