基于OPC技术和果蝇优化广义回归神经网络的海洋蛋白酶发酵过程软测量

Soft sensor for marine protease fermentation process based on OPC technology and fruit fly optimization general regression neural network

鉴于海洋生物酶发酵过程中关键生物参数难以实时在线测量的问题,提出了一种基于果蝇优化算法(Fruit fly optimization algorithm,FOA)的广义回归神经网络(Generalized regression neural network,GRNN)与面向过程控制的对象链接与嵌入技术相结合的软测量方法。GRNN的非线性映射能力强、学习速度快,但GRNN的预测性能受平滑因子的影响比较大,因此利用FOA对GRNN的平滑因子进行寻优,以提高模型的泛化能力,采用OPC技术可以实现MATLAB和组态王之间的数据通讯,将预测的关键生物参数值传送给组态王进行实时显示与存储。通过采集海洋蛋白酶发酵过程的实验数据,建立基于FOA优化GRNN的海洋蛋白酶发酵过程关键生物参数(菌体质量浓度、基质质量浓度、酶活)的软测量模型,并与GRNN、BP神经网络、支持向量机(Support vector machine,SVM)进行对比。结果表明,基于FOA优化GRNN的软测量模型对训练样本的拟合能力和对测试样本的预测能力都远远超过GRNN、BP神经网络和SVM,通过OPC技术将MATLAB和组态王进行数据连接,实现了生物参数的实时在线测量,且系统运行的稳定性较好。

Considering the problem that the crucial biological variables in the marine biological enzyme fermentation process are difficult to measure on-line in real time,a soft sensor method based on the fruit fly optimization algorithm(FOA)of generalized regression neural network(GRNN)is proposed which is combined with the OPC technology.GRNN has strong nonlinear mapping ability and fast learning speed,but the prediction performance of GRNN is greatly affected by smoothing factors.Therefore,FOA is used to optimize the smoothing factors of GRNN to improve the generalization ability of the model.The data communication between MATLAB and Kingview can be realized by object linking and embeded for process control(OPC)technology,the predicted key biological parameters are transmitted to Kingview for real-time display and storage.By collecting the experimental data of marine protease fermentation process,a soft sensing model of the key biological parameters(cell concentration,substrate concentration,enzyme activity)in the marine protease fermentation process is established which is based on the optimization of GRNN by FOA,and compared with GRNN,BP neural network and Support Vector Machine(SVM).The results show that the fitting ability to the training samples and the prediction ability to the test samples of the soft sensor model based on the FOA optimized GRNN are far better than GRNN,BP neural network and SVM.Through OPC technology,MATLAB and Kingview are connected to realize the real-time online measurement of biological parameters,and the system has good stability.