Abstract: The concentration of gelatin is an important task in gelatin production. Soft measurement is used in measuring the concentration of gelatin in order to resolve the problem of the off-line sampling and monitoring method with low accuracy, a long time lag, measurement error and so on. Wavelet neural network soft measurement model is used for gelatin concentration and mixed mutation with adaptive genetic algorithm is for parameter optimization. 120 groups of data collected from gelatin production process and dealt with normalization are used for simulation, of which 80 groups as a training and 40 as a predicting. The method can enhance the model’s robustness and improve the convergence rate. The results of simulation indicate that the square error of wavelet neural network model is less than 0.01 after 67 iterations and tends to stable. It is proved that the algorithm is feasible for the concentration of gelatin-line measurement.

Key words: mixed mutation, adaptive, genetic algorithm, gelatin concentration, soft measurement

摘要： 胶液浓度的测量在明胶的生产过程中是一项很重要的任务，为解决明胶浓度胶液手工检测中存在的滞后时间长、测量误差大等实际问题，将软测量技术应用到明胶的浓度测量中。通过对明胶生产工艺的分析，采用小波神经网络建立软测量模型，利用混合变异自适应遗传算法对模型参数进行优化。用现场采集并归一化后的120组数据做仿真实验，其中80组作为训练，40组作为预测。仿真结果表明，该算法迭代67次后误差平方和低于0.01并趋于稳定。证明混合变异自适应遗传算法可增强小波神经网络模型的鲁棒性和提高收敛速度。所以该算法用于明胶浓度的软测量中是可行的。

关键词: 混合变异, 自适应, 遗传算法, 明胶浓度, 软测量