摘要
基于结构化类比思想,提出针对时间序列的预测算法SAP-TS。通过类比建立条件概率分布,解决了以往概率模型在此时遭遇的精度问题、空间问题或缺值问题,使用综合置信指标在预测的同时评估预测准确性。在预测PTA共沸精馏塔塔顶醋酸含量的工程应用中,SAP-TS的预测精度高于目前实际使用的广义回归神经网络算法。误差分析表明其综合置信指标对预测准确性的评估是有效的。
Based on the thoughts of structured analogy forecasting, a novel algorithm is proposed to solve the numeric time series probability prediction problem named Structured Analogy Prediction for Time Series（SAP-TS）. SAP-TS constructs the conditional probability distribution through analogies, which avoids the obstacles encountered by classical probability methods, either weak predictability or intractable extremely large contingency tables, which also incurs lack of data problem. Furthermore, SAP-TS offers integrated confidence index to evaluate the prediction accuracy instantaneously. When applying SAP-TS to predict the acetic acid amount of Purified Terephthalic Acid（PTA） solvent system, the prediction results are more precise than the results of Generalized Regression Neural Network（GRNN）. The previous best method, and the integrated confidence index also effectively evaluate the prediction accuracy,
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第1期211-214,共4页
Computer Engineering
关键词
时间序列预测
结构化
类比
贝叶斯网络
数据挖掘
time series prediction
structured
analogy
Bayesian networks
data mining
