Prediction of the Profit Function for Industrial 2‐Keto‐L‐Gulonic Acid Cultivation

Abstract

The profit function is the generic criterion to describe the cost effect of a batch process. To focus on the prediction of the profit function for 2-keto-L-gulonic acid (2-KGA) cultivation, which is potentially applicable for process monitoring and optimal scheduling, rolling learning-prediction (RLP) based on a support vector machine (SVM) is applied. The RLP implies that the SVM training database is rolling updated as the batch of current interest proceeds, and the SVM learning is then repeated for the prediction. The database is further updated after termination of a batch. The updating procedures are investigated in detail. Pseudo-online prediction is carried out using the data from industrial-scale 2-KGA cultivation under actual and hypothetical inoculation sequences. The results indicate that the average relative prediction error is less than 5 % in the later phase of fermentation in all inoculation sequences. Introduction2-Keto-L-gulonic acid (2-KGA) is the precursor for the synthesis of L-ascorbic acid. It is typically produced by a mixed culture of Bacillus megaterium (B. megaterium) and Gluconobacter oxydans (G. oxydans) with L-sorbose as substrate [1][2][3]. In 2-KGA fermentation workshops, batch-to-batch variation in process performance is often observed, even though the same culture conditions are applied. This may result from many factors such as fluctuations in the seed quality and raw materials composition. For control engineers, batch-dependent variation between batches implies the potential to improve the process performance by dynamic optimization or optimal scheduling [4][5][6][7][8]. The profit function, which is defined as the gross profit of a batch over the production time, reflects the cost effect of the batch process and can be used as the index to evaluate the batch performance. Predicting the profit function will benefit process monitoring and optimal scheduling. Here, the focus is on the data-driven prediction of the profit function.In recent years, computational intelligence techniques such as artificial neural networks (ANNs) [9][10][11][12][13][14] and support vector machine (SVM) [14][15][16][17][18][19] received more and more attention because of their ability to model nonlinear systems. SVM, developed by Vapnik, is based on the statistical theory [20,21]. It implements the structural risk minimization (SRM) principle which is an approach to minimize the upper bound risk related to the generalization performance. The solid theoretical basis makes it advantageous in generalization and convergence [19][20][21]. In this work, rolling learning-prediction (RLP) based on an SVM is practiced to predict the profit function of industrial-scale 2-KGA cultivation. The RLP mechanism is adopted to deal with the time-variant property of fermentation, and it makes full use of the available data of the predicted batch by adding these real-time data into the training database of SVM to start a new round of SVM learning. When the prediction approach is online operated, new batches will arrive in ...

Online Application‐Oriented Optimal Scheduling for 2‐keto‐l‐gulonic Acid Production

“…The profit function of industrial 2‐KGA cultivations, J ( t ), is defined as the average profit of a batch over its cultivation time: where t is the production time, P ( t ) the total 2‐KGA formation of the batch assuming recovery to be done at t , U sale the unit price of 2‐KGA, and C cost ( t ) is the total cost of the batch up to time t , which includes the cost of energy, raw materials, labors, capital, and so on. T gap is the production time interval of two batches successively carried out with the same reactor.…”

Simulation support system and modelbase technology in cold rolling process control system

“…Based on the features of the system, the design idea of building the simulation support system based on field modelbase is presented [3,4]. As the example, simulation system architecture, the extract and design of reusable framework, and the building of domain modelbase in cold continuous rolling process control are discussed.…”