生物发酵过程的温度控制模型研究

Temperature Control Model in Fermentation Process

针对生物发酵过程中温度控制难以建模的问题,基于非线性自回归滑动平均(NARMA)模型,设计了神经网络自回归滑动平均(NN-NARMA)模型.利用径向基神经网络逼近NARMA模型中的映射关系,对神经网络的输出进行了二阶低通滤波,用变异率可调节的遗传算法优化了NARMA模型中的延时参数以及神经网络的输出滤波参数.应用该方法建立了生物发酵过程的温度控制模型,该模型在上温、中温和下温的误差相对于Elman神经网络模型分别减少了38 9%、13 5%和61 3%.该方法具有一定的可操作性,能够较好地解决生物发酵过程中的温度控制建模问题.

To solve the problem of modeling temperature control in the fermentation process, a neural net work nonlinear auto regressive moving average (NN-NARMA) modeling method for nonlinear system is proposed. In the nonlinear mapping relation of NARMA model, which is expressed by radial basis function (RBF) neural network, the output undergoes two-order low-frequency pass filter and the genetic arithmetic that has multi mutation probabilities is employed to optimize the parameters (e.g., delay of NARMA model and coefficients of filter, etc.). A group of contrasting experiments in beer fermentation process are conducted and the results show that the temperature errors from the top, middle and bottom of the fermentation tank by the NN-NARMA model are reduced by 38.9%, 13.5% and 61.3% respectively compared with Elman neural network model. The proposed method is more effective to solve the problem of modeling the temperature control in fermentation process.