生物发酵过程效益函数的在线预报

提出了一种基于神经网络的效益函数预报方法，为了提高预报精度，引入滚动学习预报技术处理过程的时变性，该预报技术中，输入输出数据对通过移动窗口获得。在每一次采样时间后，由输入输出数据对构成的训练数据库将被更新，然后程序重复进行下一轮的预报。为检验所提出方法的有效性，以头孢菌素C生产为例，对效益函数进行了超前24h预报，结果显示该方法的预报精度高于已有的方法。

The On-line Prediction of Profit Function for Fed-Batch Bioprocesses

A novel approach for the on-line prediction of the profit function was proposed based on neural networks. In order to improve the prediction accuracy, a rolling learning-prediction procedure was introduced to deal with the time variant property of the bioprocess. In this procedure, the input-output data pair is obtained using a moving data windows technique. The training database, which is composed of the set of input-output data pairs, is updated after each sampling interval, and the learning-prediction is repeated thereafter. To test the validity of the proposed approach, the profit function in Cephalosporin C fedbatch cultivation is predicted 24 hours ahead as an example. The result indicates the proposed approach outperforms the previous method in prediction accuracy.