| Chlortetracycline (CTC) is a broad-spectrum antibiotic of tetracycline, secondarymetabolite of Streptomyces aureofaciens and one of the eight antibiotics for animal feedproducts. CTC is characterized by antibacterial，high feed efficiency，promoting animalgrowth，less residue, lower production costs, mature technology relatively etc., which hasbeen the largest amount of antibacterial agents. But chlortetracycline is a typical secondarymetabolites whose mechanism of microbial synthesis is complex quietly and its mechanismof reaction is very difficult to identify how to be structured. So far scientists in the scopesare still active in exploring the mechanism of the reaction of CTC biochemical process.Therefore, to achieve high yield of CTC, the optimization control of fermentation process isthe best of high efficient method promoting established strain, in addition to well-conductedselect Streptomyces aureofaciens strains and develop a medium more suitable breedingstrains grown. Because of the synthesis result of the genetic character at molecular level,the metabolic regulation at cellular level and the transfer characteristic at reactorengineering level of microbial fermentation process, the fermentation of CTC shows itscharacteristics such as time-varying, nonlinear, large delay and uncertain. The mainscientific research of this dissertation is to study on soft sensor modeling method based onartificial intelligence, information technology and machine learning so as to implement theoptimal control strategy for CTC fermentation process. The specific research work in thisdissertation is summarized as follows:1. For some difficult-to-measured process state variables, a hybrid soft sensormodeling method based on estimation and inference of mathematics is proposed in thedissertation. The method of soft sensor model proposed in this dissertation is to estimate theun-measurable process variables by establishing dynamic and static model, data-drivenmodel or hybrid model based on those measurable physical and chemical (or biological)parameters of the process. By analyzing some basic steps of establishing soft sensor model,such as data choosing, filtering, regularization and model verification methods and so on, amethod proposed in this dissertation is to establish a reliable, robust and generalized softsensor model, which it can be applied to predict the amino nitrogen, biological titer andtotal sugar content in the process of fermentation.2. The three variables of amino nitrogen, biological titer and total sugar content areimportant process parameters but not measured online throughout the CTC fermentation. By integrating local model and global model method and taking advantage of GeneralizedRegression Neural Network (GRNN), Output Recurrent Wavelet Neural Network(ORWNN) and Self-Organizing feature Map&Least Square Support Vector Machine(SOM+LSSVM) mixing Just-In-Time Learning&Recurrent Least Square Support VectorRegression (JITL-RLSSVR) method, three aforementioned variables are constructedrespectively in order to predict changes in state variables and establish the foundation torealize optimizing control in the process of CTC fermentation.3. If the fermentation process is invaded by other harmful bacteria during culturephases, the broth (fermentation liquid) may be contaminated. Once this occurs, the cultureprocess must be halted and the broth must be discharged for preventing contamination ofother fermenting kettles. Obviously, this will cause a huge waste of raw material and upsetthe schedule of plant production. The disastrous of those CTC plants will happen if theother fermenting kettles are invaded by the harmful potent bacteria as a result of the failureof the disinfection work and sterilization to contaminated fermenting kettle non-in time. Inthis dissertation, some important comprehensive feature information of contamination bycombing multiple process information hinting indirect contamination can be acquired usingthe information fusion technology based on Dezert-Smarandeche theory. And simulationexperiments based field data show that the method can predict whether the process of CTCfermentation is contaminated, so if this method is applied into real fermentation production,the safety performance of the culture process will be improved.4. Some abnormal failures of machinery or sensor may affect normal productionduring CTC culture. For three common failures of CTC culture process, i.e. metering tankof sugar failure, pH electrode failure and DO electrode failure, the fault detection anddiagnosis model is proposed in this dissertation based on Just-In-Time Learning+Hierarchical Kernel Partial Least Square (JITL+HKPLS) and Multiway Gaussian MixtureModel+Hierarchical Kernel Partial Least Square (MGMM+HKPLS) hybrid model. Thefield experimental results demonstrate that the model proposed can achieve good faultdetection and diagnosis results.5. CTC fermentation is multiphase fed-batch process, and strains culture process ateach phase requires a different optimal environment for achieving high-quality andhigh-yield metabolites. In order to obtain the goal, optimal control strategy must beimplemented in fermentation process. An automatic phase-division algorithm is studied inthis dissertation, and one phased sugar feed rate regulation method of optimal control strategy based phase-division algorithm is presented. And based on benefit function of CTCproduction, an optimal schedule method for culture process is proposed in order to achievethe optimal CTC production cycle.It is challenge task to achieve optimal control for production process without accurateinternal dynamic and static model because CTC fermentation is a complex biochemicalprocess. By analyzing the characteristics of input and output of CTC fermentation, anoptimal control algorithm for sugar feed is proposed based on constructed soft sensormodel of difficult-to-measured process state variables. On the basis of Labview2010andSQL Server2010software platform, the optimal control strategy system for CTCfermentation is developed. The result of application indicates that this control systemimproves automatic level of production process, and reduces the labor intensity of workersand increases the economic efficiency. |
PDF Full Text Request