摘要
针对即时(惰性)学习模型频率降低间接导致的精度下降问题,提出一种二阶相似性的即时学习方法。该方法综合顾及到样本集的整体分布特性,在传统一阶相似度准则的基础上建立二阶相似度准则,采用与测试样本具有绝大部分相同近邻的二阶相似样本建立当前时刻的模型;同时将累计相似度因子用于建立局部模型时样本量的确定,并采用相似度阈值的方式判断此刻模型是否需要重新建立。该方法在青霉素发酵过程产物浓度的预测实验中得到了有效的验证。
Aiming at the indirect accuracy reduction caused by the frequency reduction of just-in-time(lazy)learning model,a second-order similarity just-in-time learning method is proposed.This method takes into account the overall distribution characteristics of the sample set,establishes a second-order similarity criterion based on the traditional firstorder similarity criterion,and uses a second-order similarity sample with most of the same neighbors as the test sample to establish the model at the current time.At the same time,the cumulative similarity factor is used to determine the sample size when the local model is established,and the similarity threshold is used to determine whether the model needs to be rebuilt at this time.This method has been effectively validated in the prediction experiment of the product concentration in the fermentation process of penicillin.
作者
祁成
史旭东
熊伟丽
QI Cheng;SHI Xudong;XIONG Weili(School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China;Key Laboratory of Advanced Process Control for Light Industry Jiangnan University,Ministry of Education,Wuxi 214122,China)
出处
《智能系统学报》
CSCD
北大核心
2020年第5期910-918,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61773182)
江苏省自然科学基金项目(BK20170198).
关键词
即时学习
更新频率
二阶相似度
相似度准则
一阶相似度
局部模型
累计相似度因子
相似度阈值
just-in-time learning
update frequency
second-order similarity
similarity criterion
first-order similarity
local model
cumulative similarity factor
similarity threshold
