基于改进即时学习算法的镨/钕元素组分含量预测

just-in-time learning (JITL); extraction process; component content; prediction; similarity criterion; updating strategy of local models

中文关键词  即时学习  萃取过程  组分含量  预测  相似度准则  局部模型更新策略

英文关键词  Just-in-time learning (JITL)  extraction process  component content  prediction  similarity criterion  updating strategy of local models

中文摘要

      针对镨/钕(Pr/Nd)萃取过程元素组分含量难以在线实时检测的现状, 引入加权相似度准则和局部模型更新 策略, 提出一种基于改进即时学习算法的稀土元素组分含量快速估计方法. 首先, 为了保证即时学习算法学习集选 取的合理性, 充分考虑输入输出变量之间的相关程度, 采用互信息加权的相似度准则选择建模邻域, 以最小二乘支 持向量机(LSSVM)作为即时学习算法的局部模型; 其次, 依据由相似度阈值更新和数据库更新组成的模型更新策 略校正LSSVM局部模型, 改善组分含量预测模型的精度和实时性; 最后, 基于镨/钕萃取现场数据进行仿真对比试 验, 结果表明所建模型具有精度高、实时性好等优点, 适用于稀土萃取生产现场元素组分含量的快速预估.

英文摘要

      Considering the problem that element component content in the praseodymium/neodymium (Pr/Nd) extraction process is difficult to detect accurately online, an improved just-in-time learning algorithm is used to build prediction model of the element component content. In the meantime, the weighted similarity criterion and updating strategy of local models are introduced into this algorithm. Firstly, In view of the fact that the correlation between input and output variables of the database are different in the learning set of the just-in-time learning algorithm, the similarity criterion of mutual information weight is lead up to select the modeling neighborhood, which ensures the rationality of algorithm’s learning set. Then, least squares support vector machine (LSSVM) is adopted as the local model of JITL algorithm. Secondly, the model update strategy consisted of similarity threshold updates and database updates is used to correct LSSVM local model, so as to enhance the performance of real-time and self-adaption for the component content model. Finally, The simulation results through Pr/Nd extraction field data show that the model has high precision and performance of real-time is excellent. This method is suitable for rapid estimation of element content in rare earth extraction production sites.