Research on Soft Measurement Method Combining Grid Search Optimization Clustering and Improved SVR Algorithm

Aiming at the problem that certain parameters in a class of complex industrial engineering cannot be measured in real time by hardware sensors, this paper proposes a multi-model soft measurement method combining grid search optimized clustering and improved SVR algorithm. Firstly, the improved ISODATA algorithm is used to cluster the production process data, and the core idea of its improvement is to initialize the clustering center by using K-Means++ algorithm and apply grid search and cross-validation to optimize the ISODATA clustering parameters. Secondly, according to the clustering results, the improved SVR algorithm is used to construct the soft measurement model after the optimization of sparrow search, respectively. Finally, using the data generated by the penicillin fermentation simulation platform to perform simulation experiments. ISODATA algorithm-improved SVR algorithm compared with soft measurement method of the standard ISODATA algorithm-improved SVR algorithm and the cyclic neural network semi-supervised dynamic soft measurement method of recurrent neural network respectively. The results show that the soft measurement method proposed in this paper has better evaluation indexes and higher prediction accuracy.