Research On Soft Sensor Method In Fermentation Process Based On AP Clustering Algorithm

Biological fermentation technology is the foundation of modern biological engineering and related industry. With the increasing demand of the biological fermentation product, status of biological fermentation technology in national economy is more and more important. Therefore,it is an urgent to adopt the advanced control strategy to get the fermentation products of low cost and high quality. However, because of the limitation of the current detection technology, effective real-time detection of some important parameters in the fermentation process is not easy to achieve. It severely limits the application of advanced control in the fermentation production. As the forefront of current detection and control technology, soft sensor technology can establish effective prediction model for serious nonlinear, strong time-varying, and many inherent system. Therefore, this paper analyzes the existing soft sensor modeling method and the characteristics of fermentation process, and studies the related modeling method. The study is based on the affinity propagation clustering algorithm(AP) and least squares support vector machine(LSSVM).This paper firstly studies the existing soft sensor modeling method and biological fermentation process. And then this paper studies the modeling of biological fermentation based on the AP clustering algorithm and LSSVM.According to the characteristics of high nonlinear process and multi period, this paper uses AP clustering algorithm for the stage division of fermentation process. Then the similarity values of AP clustering algorithm were improved to fit the fermentation process’ s data characteristics, and DE-LSSVM method was used to establish prediction model for each stage. In the integration of the global model, this paper analyzes the current fusion method and puts forward a new fusion strategy. The paper adopts the method of constructing local models to achieve fusion model. Simulation results show that the proposed modeling method improves the global model prediction accuracy and generalization performance.In order to solve the off-line model aging problems of multi batch fermentation process, a new online modeling method based on AP clustering algorithm and dynamic time warping(DTW) distance was proposed. This method uses DTW distance as similarity criteria to search the most similar sequences from historical data. And the search sequence from sliding time window. And then the paper uses the local learning theory and online training sample set to establish the online prediction model. In addition, to narrow the search scope, reduce the amount of computation and improve the performance of modeling, this paper uses AP clustering algorithm for condition division of historical data. The result shows that the method improves the soft measurement model’s prediction accuracy and generalization ability and can establish effective online prediction models. This method has certain reference value for solving practical problems in production.