Research on Penicillin Concentration Soft Sensing Method based on K-means++ Clustering and Improved SVR Algorithms

For the problem that penicillin concentration cannot be measured by hardware sensors in the production process, a soft sensing method is proposed based on the combination of K-means++ clustering and improved SVR algorithm. Firstly, the production conditions are clustered by using the K-means algorithm to obtain the initial number of clusters. Then, by using K-means++ for improved clustering results, the initial cluster centers are optimized. Secondly, an improved SVR algorithm combining grid search and cross-validation is utilized to construct a predictive model for penicillin concentration. Finally, the data generated by the penicillin fermentation process simulation platform are utilized for simulation experiments. The simulation results are compared among three methods: K-means++ with improved SVR, K-means with improved SVR, and adaptive AP-GPR. The results indicate that the proposed soft sensing has better evaluation indicators and higher prediction accuracy.