Multivariate time series online prediction based on adaptive normalized sparse kernel recursive least squares algorithm

As a kind of kernel methods, kernel recursive least squares has attracted wide attention in the research of time series online prediction. It has low computational complexity and updates in the shape of recursive increment. However, with the increase of data size, computational complexity of calcula...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:2017 Seventh International Conference on Information Science and Technology (ICIST) s. 38 - 44
Hlavní autori: Shuhui Zhang, Min Han, Jun Wang, Dan Wang
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 01.04.2017
Predmet:
adaptive
Coherence
Computational complexity
Decision support systems
Heuristic algorithms
Kernel
kernel recursive least squares
Shape
Time series analysis
time series online prediction
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:As a kind of kernel methods, kernel recursive least squares has attracted wide attention in the research of time series online prediction. It has low computational complexity and updates in the shape of recursive increment. However, with the increase of data size, computational complexity of calculating kernel inverse matrix will raise. In addition, in the process of online prediction, it cannot accommodate dynamic environment commendably. It is difficult to meet the demand of prediction accuracy and efficiency simultaneously. Therefore, this paper presents an improved kernel recursive least squares algorithm for multivariate chaotic time series online prediction. We apply dynamic adjustment and coherence criterions to propose adaptive normalized sparse kernel recursive least squares (ANS-KRLS) method. In our method, the size of kernel matrix can be reduced, so that computational complexity drops. And ANS-KRLS has the ability to operate online and adjust weights adaptively in time-varying environments. The proposed method has been simulated on Lorenz time series and ENSO related indexes time series. Simulation results prove that ANS-KRLS performs well on prediction accuracy and efficiency.
DOI:10.1109/ICIST.2017.7926807