East China JiaoTong University
The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)
针对稀土萃取过程进行质量监控时, 存在采集样本重复率高、有效数据少的小样本问题, 提出一种基于混合虚拟样本生成的稀土萃取过程组分含量预测方法. 首先, 以萃取现场的小样本为基础, 采用中点插值法生成虚拟样本输出数据, 再根据随机配置网络（SCN）中隐含层与输出层、输入层与隐含层间的映射关系, 生成虚拟样本输入数据; 鉴于这些虚拟样本仅能在邻近点产生的局限, 采用结合遗传算法（GA）的多分布趋势扩散技术（MD-MTD）生成优化的虚拟样本集进行补充. 依据数据合理性原则, 将虚拟样本与真实小样本进行融合, 建立基于SCN的组分含量预测模型. 通过铈镨/钕萃取现场数据验证和对比实验分析, 结果表明本文方法能有效解决小样本问题, 适用于稀土萃取过程组分含量监控.
Aiming at the small sample problem of high sample collection repetition rate and low effective data during quality monitoring of rare earth extraction process, a method for predicting component content of rare earth extraction process based on mixed virtual sample generation is proposed. First of all, based on the small sample extracted from the field, the output data of virtual sample are generated by midpoint interpolation method. Then, the input data of virtual sample are generated according to the mapping relationship between the hidden layer of the stochastic configuration network (SCN) and the output layer, the input layer and the hidden layer. In view of the limitation that the virtual sample can only be generated at neighboring points, a multi-distribution trend diffusion technology (MD-MTD) combined with genetic algorithm (GA) is used to generate an optimized virtual sample set to supplement. According to the principle of data rationality, the virtual sample and the real small sample are merged, and a component content prediction model based on SCN is established. Through the field data verification and comparative experimental analysis of CePr/Nd extraction, the results show that the method in this paper can effectively solve the problem of small sample and the paper is suitable for component content monitoring in the rare-earth extraction process.