基于图拉普拉斯变换和极限学习机的时间序列预测算法

由于时间效率的约束,多元时间序列预测算法往往存在预测准确率不足的问题。对此,提出基于图拉普拉斯变换和极限学习机的时间序列预测算法。基于图拉普拉斯变换对时间序列进行半监督的特征提取,通过散布矩阵将监督特征和无监督特征进行融合。设计在线的极限学习机学习算法,仅需要在线更新网络的输出权重矩阵即可完成神经网络的学习。利用提取的特征在线训练极限学习机,实现对多元时间序列的实时预测。基于多个数据集进行仿真实验,结果表明该算法有效地提高了预测准确率。

TIME SERIES PREDICTION ALGORITHM BASED ON GRAPH LAPLACE TRANSFORM AND EXTREME LEARNING MACHINE

Because of the constraints of time efficiency,classical multivariate time series prediction algorithms suffer from low prediction accuracy.In view of this,I propose a time series prediction algorithm based on the graph Laplace transform and extreme learning machine.It adopted graph Laplace transform to abstract the semi-supervised features of time series,and fused the supervised feature and un-supervised features through scatter matrix.I designed a online extreme learning machine learning method,and it only updated output weight matrix online to finish the neural networks leaning.This algorithm utilized the extracted features to train extreme learning machine,and it realized real time prediction for multivariate time series.The simulation experiments were finished based on several datasets.The results show that the proposed algorithm improves the prediction accuracy of time series effectively.