基于FLNN的多粘菌素发酵过程建模

为解决由于缺乏传感器使众多状态参数难以在线测量的问题，建立了多粘菌素的发酵过程模型，对许多重要的状态参数进行了预估．通过在FLNN内部增加一个带有局部激活反馈和一个局部输出反馈的自回归滑动平均滤波器使其成为动态的FLNN网络，并把它运用于多粘菌素发酵过程的建模中，结合遗传算法实现对其发酵过程的茵体浓度、总糖浓度和相对效价进行预估，为实际生产和优化控制提供了有利条件．仿真结果表明，基于改进的FLNN建立的多粘菌素发酵过程模型预估效果良好．

Modeling the Mycetozoan Fermentation Based on the Functional-Linked Neural Network

It is essential to establish the model for Mycetozoan fermentation and estimate some important state parameters owing to the shortage of sensors and many parameters not being on-line measured.The functional-linked neural network can approach arbitrary non-linear function by learning examples. The network structure is very simple,and in the training stage its computational cost is smaller,but Mycetozoan fermentation process is dynamic. Therefore, we improved the FLNN structure where an Auto-Regressive Moving Average filter has been placed either before the activation function of the neuron or on the back connection from the output to the neuron's input.Then, the improved network that absorbs genetic algorithm is applied in modeling the process and state estimation such as biomass or substrate or produce concentration .The simulation result proved that the mapping ability of the suggested neural network is stronger and the prediction ability of the corresponding model is better.