Method for dynamically predicting product concentration in fermentation process based on incremental learning

• Main Authors: WANG XU, YANG LEYAN, JI HAIPENG, LYU HUA, DONG YONGFENG, LIU JING
• Format: Patent
• Language: Chinese; English
• Published: 30.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; PHYSICS

The invention discloses a fermentation process product concentration dynamic prediction method based on incremental learning. The method comprises the following steps: S1, constructing a method framework comprising a similarity calculation module based on feature dimension reduction and a self-adaptive updating module based on incremental learning; s2, training a historical model by adopting a fermentation historical sample; s3, performing feature dimension reduction on newly added samples and historical samples in the fermentation process, and querying samples similar to the newly added samples from the historical samples according to a similarity calculation algorithm to form a similar sample set; s4, respectively calculating a loss gradient of a random training batch of the similar sample set and the historical model training sample, judging a gradient vector included angle, and calculating to obtain a new gradient; and S5, updating the historical model by using the new gradient, wherein the updated model