论文部分内容阅读
针对一类因检测困难而导致检测数据稀少的连续工业过程,提出了基于离散Walsh变换的过程神经网络建模方法。在对稀疏样本数据进行预处理的基础上,采用递推式非邻均值生成法对样本数据进行扩充,以此建立可产生任意密集预测数据的过程神经网络模型,并采用在线滚动学习的方法进一步提高所建立的预测模型的精度。以味精发酵过程菌体浓度预测为例,验证了所建立的过程神经元网络预测模型可以得到非常高的预测精度。
Aiming at a series of continuous industrial processes with few detection data due to the detection difficulty, a process neural network modeling method based on discrete Walsh transform is proposed. Based on the preprocessing of the sparse sample data, the recursive non-adjacent mean value generation method is used to extend the sample data to establish a process neural network model that can generate arbitrary and intensive prediction data, and the method of online rolling learning is further used Improve the accuracy of the established prediction model. Taking the microbial biomass concentration in the process of MSG as an example, it is verified that the established neural network prediction model can obtain a very high prediction accuracy.