Soft Sensor Modeling And Parameter Optimization For On-Line Estimation Of Gelatin Concentration

Gelatin concentration is one of the most important process parameters in gelatin production process. Gelatin concentration online measurement has the important meaning. But now, gelatin concentration measurement also has focused on manual offline sample. This method has big error, long time and environmental pollution shortcomings. In order to solve this problem, the soft-measurement was introduced to gelatin concentration measurement, and gave gelatin concentration measurement a new way.The research background was QingHai Gelatin Company's gelatin production control system and research object was gelatin concentration, aiming at improving the real-time and robustness of gelatin concentration online measurement. Based on the analysis of the production process, choosing the time and temperature as auxiliary variables, which are very closely related with gelatin concentration and more easily to measure. Then through the analysis of the sample, we have learned that there are complexities, nonlinear and nonstructural characteristic along these samples.According to these characteristics, wavelet neural network was introduced to the soft measurement model because of its non-linear approximation ability. And then improved the algorithm, get the adaptive wavelet neural network. Through the predicted results of them, wavelet neural network has better optimization performance and generalization ability. So the wavelet neural network model was chose as gelatin concentration prediction model.But the structure of wavelet neural network was more complex and it's also had more optimization parameters. In order to fully solve this problem, the genetic algorithm with hidden parallel search performance was used to optimize wavelet neural network model. Later on the adaptive ability issues, got improved adaptive genetic algorithm; on the chromosome structure, got mixed hierarchical genetic algorithm; and on the basis of the two comprehensive performances improvement, got mixed mutation adaptive genetic algorithm. Four optimization algorithms are used to optimize the gelatin concentration wavelet neural network model, running on MATAB environment. Through the predicted results, compared with the traditional genetic algorithm, the adaptive genetic algorithm has better real-time, but the prediction precision was not very well. Compared with the adaptive genetic algorithm, mixed hierarchical genetic algorithm has better real-time and prediction precision. Yet compared with the above three, mixed mutation adaptive genetic algorithm has the best real-time and prediction precision, fully satisfied the gelatin production requirements.At last, this paper discussed the soft measurement model online realized in the concentration of gelatin glue control system through mixing program of VB and MATLAB.