基于RF-GA-BP神经网络的N-乙酰氨基葡萄糖含量预测

为解决微生物发酵法制取氨基葡萄糖(Glucosamine,GlcN)过程中N-乙酰氨基葡萄糖(N-acetylglucosamine,GlcNAc)含量难以在线测量的问题。提出一种由随机森林算法、遗传算法及神经网络算法相结合的改进预测算法。利用随机森林算法中平均不纯度下降的特点,对输入特征进行关联性分析,并通过遗传算法对神经网络初始权值、阈值进行优化。以某氨糖生产企业发酵过程中的数据为基础,建立基于RF-GA-BP算法的预测模型。结果表明：该模型对发酵生产过程中N-乙酰氨基葡萄糖含量具有良好的预测能力,所提出的模型兼顾了高精度与快收敛的需求,测试样本预测平均误差低于7%,优于GA-BP模型与传统BP模型。

Prediction of N-acetylglucosamine Content Based on RF-GA-BP Neural Network

In order to solve the problem that the content of N-acetylglucosamine (GlcNAc) in the process of preparing glucocosamine (GlcN) by microbial fermentation is difficult to measure online, an improved prediction algorithm based on stochastic forest algorithm, genetic algorithm and neural network algorithm is proposed. The algorithm utilizes the feature of decreasing average impurity in random forest algorithm to analyze the relevance of the input characteristics. The initial weights and thresholds of the neural networks are optimized by the genetic algorithm. A prediction model based on the RF-GA-BP algorithm is established based on the data from the fermentation process of an ammonia sugar production enterprise. The results show that the model has a good prediction ability for the content of GlcNAc. The prediction model has a fast convergence rate and a high accuracy. The average error is less than 7%, which is better than the GA-BP model and the traditional BP model.