基于误差修正的菌体浓度软测量

On-line Measurement of Biomass Concentration by Error-corrected Method

针对机理模型中参数易受环境影响,结果常常不准,而数据模型对于复杂系统外推能力差的情况,提出了以机理模型为基础,以数据模型为补充,利用数据模型对机理模型的预测结果进行误差修正的方法。将该方法用于菌体浓度的预测,误差修正模型采用RBF神经网络,包含了影响菌体浓度的主要理化因素:温度、溶解氧和pH,以实际测量值为目标对该网络进行训练。训练好的神经网络用来对机理模型的输出进行修正。试验数据表明该方法能有效提高菌体浓度的预测精度。

A novel method is proposed to estimate the biomass concentration on-line by combining a mechanics model of poor extrapolation with a neural network model of unreliable parameters. An error corrected model is built to correct the output of mechanics model. It is made up of a RBF neural network that consisted of temperature, dissolved oxygen and pH and is trained by the practical values. The laboratory result shows this method can effectively increase the accuracy of forecasting biomass concentration.