arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2102.05314cs.LGmath.STstat.MLstat.TH
收藏

基于非负矩阵分解的部分观测时间序列预测

Time series forecasting from partial observations via Non-negative Matrix Factorization

Institut Camille Jordan, Ecole Centrale Lyon(让-卡米尔·约当研究所,中央理工大学);Institut Universitaire de France(法国大学研究院)

Yohann de Castro, Luca Mencarelli

AI总结:

提出滑动掩码方法(SMM)结合非负矩阵补全进行多非负时间序列预测,通过掩码原型矩阵分解(mAMF)和掩码归一化非负矩阵分解(mNMF)实现,理论证明恢复误差与噪声成比例,实验优于Transformer、LSTM等方法。

摘要:

在现代时间序列问题中,我们旨在预测可能包含缺失值和噪声的多重时间序列。本文引入滑动掩码方法(SMM),通过非负矩阵补全来预测多个非负时间序列:将观测到的噪声值和预测/缺失值收集成矩阵形式,并通过将其行表示为少量非负向量(称为原型)的凸组合来实现学习。我们提出了两种估计方法,掩码原型矩阵分解(mAMF)和掩码归一化非负矩阵分解(mNMF),它们可以与SMM方法结合。我们证明这些估计能以与噪声成比例的误差恢复真实原型。我们使用近端交替线性化方法(PALM)来计算原型和凸组合权重。我们在真实数据上将我们的估计器与最先进的方法(Transformer、LSTM、SARIMAX...)进行了多时间序列预测比较,结果表明我们的方法在大多数实验中优于它们。

英文摘要:

In modern time series problems, one aims at forecasting multiple time series with possible missing and noisy values. In this paper, we introduce the Sliding Mask Method (SMM) for forecasting multiple nonnegative time series by means of nonnegative matrix completion: observed noisy values and forecast/missing values are collected into matrix form, and learning is achieved by representing its rows as a convex combination of a small number of nonnegative vectors, referred to as the archetypes. We introduce two estimates, the mask Archetypal Matrix factorization (mAMF) and the mask normalized Nonnegative Matrix Factorization (mNMF) which can be combined with the SMM method. We prove that these estimates recover the true archetypes with an error proportional to the noise. We use a proximal alternating linearized method (PALM) to compute the archetypes and the convex combination weights. We compared our estimators with state-of-the-art methods (Transformers, LSTM, SARIMAX...) in multiple time series forecasting on real data and obtain that our method outperforms them in most of the experiments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑