Soft sensor of generalized dynamic fuzzy neural network for marine protease fermentation process based on dynamic data exchange

Abstract: The crucial biological variables (such as biomass concentration, substrate concentration, and product concentration, and so on) of the microbial fermentation process are difficult to measure online, which has a great influence on the quality of fermentation production. In this paper, a soft sensing method based on a generalized dynamic fuzzy neural network (GD-FNN) was proposed. The configuration software windows control center (WinCC) possesses the advantages of powerful practicality and flexible configuration. A complex interactive graphical interface can be generated by WinCC, but its ability to perform data processing is weak. So it is unable to achieve soft sensor modeling of biological parameters and estimate the value of biological parameters by WinCC. MATLAB is professional software for mathematical analysis and engineering operations. It has the characteristic of powerful data processing capabilities and an open application programming interface, but direct data communication can not be realized between MATLAB and the industrial control equipment. In order to solve this problem, combining MATLAB with WinCC to achieve respective advantage, taking Excel as the middle bridge, the real-time communication between MATLAB and WinCC was established by dynamic data exchange (DDE,DDE is the message mechanism based on Windows, two Windows applications carry on DDE Conversation through mutual transfer DDE message, and thus complete the data request, response, and transmission).Finally the real-time display and monitoring of crucial biological variables was realized. In this paper, the typical microbial fermentation process (the marine protease fermentation process) was taken as an example. First, in MATLAB, a soft sensor model based on GD-FNN(The algorithm of GD-FNN was based on an elliptical basis function. In the algorithm, fuzzy ε-completeness was used as the distribution mechanism of on-line parameters, the importance of fuzzy rules and input variables were evaluated, and this algorithm which has salient advantages in the aspect of learning efficiency and performance was established by using the training sample set for the fermentation process. The established model was verified by the test sample set. Second, the real-time collection data was transferred from configuration software WinCC to Excel by DDE technology. The data of Excel was called by MATLAB programming, and crucial biological parameters were predicted by the established model and the value transferred back to Excel. Finally, the real-time display and monitoring of the biological parameters were realized by DDE settings and the friendly human-machine interface of WinCC, and the intelligent monitoring system of a marine protease fermentation process based on WinCC was established. The application results showed that the prediction accuracy of soft sensor modeling based on GD-FNN is higher，and connecting MATLAB and WinCC by DDE technology has the characteristic of efficient programming, convenient use, and good general performance. The real-time monitoring was processed by WinCC for the marine protease fermentation process, which met the requirements of optimal control of the marine protease fermentation process and enhanced the automation level of the fermentation process and improved the product yield and economic benefit. These lay the foundation for the industrial production of a marine protease fermentation process.