Artificial neural network discrete-time biomass controller for a continuous stirred tank reactor

• Published in: Journal of computational science Vol. 91; p. 102688
• Main Authors: Hapoglu, Hale, Balas, Egemen Ander, Altuntaş, Semin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.10.2025
• Subjects: Analytical hierarchy process; Artificial neural network controller; Discrete-time biomass controllers; Generalized Predictive Control; Weighted aggregate sum product assessment; Analytical hierarchy process; Generalized Predictive Control; Artificial neural network controller; Discrete-time biomass controllers; Weighted aggregate sum product assessment
• ISSN: 1877-7503
• DOI: 10.1016/j.jocs.2025.102688

The employment of stirred tank reactors in the field of treatment technology is well-established. In this regard, a bioreactor model is commonly utilized for conducting simulations, identifying parameters, and developing control applications. Control of biomass concentration is independent of scale through manipulation of the dilution rate. To enable discrete-time control, an equivalent model incorporating a zero-order hold element and a 0.1-h sampling time has been formulated for controlling biomass concentration. In this study, the various well-known controllers performed effectively to track set points. Further, to mitigate the effects of load disturbances, the generalized predictive controller, the proportional integral derivative controller, and the controllers designed based on pole placement have been employed to obtain process control responses. The performance of these controllers has been evaluated through a weighted aggregate sum product assessment technique that employs an analytical hierarchy process. Due to the significant nonlinearity present in the closed loop bioprocess with substrate inhibition, the feedforward artificial neural network controller is trained using a closed-loop dataset, and its performances are compared with the conventional controllers. The controller has demonstrated its ability to manage realistic feed fluctuations without risking upset to the culture. The biomass concentration showed only minor deviations, settling swiftly back to the desired value by smoothly adjusting the dilution rate. This controller with tansig and purelin functions overcomes nonlinearities and time delays better than conventional controllers. The results suggest that the artificial neural network controller offers the desired simplicity and effectiveness for industrial applications.