Abstract: Nonlinear time series are often superimposed and coupled with multiple forms of changes such as trend changes, seasonal changes, cyclic fluctuations and random changes, which are manifested as cyclic changes of variable length and amplitude. For the time series with nonlinearity and instability, a dynamic prediction model combining empirical modal decomposition (EMD) and directed visible graph (DVG) network is proposed. First, the original time series is decomposed into multiple intrinsic mode functions (IMFs) by empirical modal decomposition, and then the decomposed high-frequency and low-frequency IMFs are used to obtain their respective periods by fast Fourier transform; based on each period, subsequences of varying lengths are intercepted from the tail of the original time series, and then the directed visible graph algorithm is used to convert the original time series into multiple directed networks, using random wandering in each directed network to find A node similar to the last node of the time series; finally, the next value of the time series is predicted based on the parallel line method. The time series of crude oil price is a typical non-linear and unstable series, and this model is used to empirically analyze the daily price of WTI crude oil time series. The research results show that this model can not only effectively predict the time series trend, but also has a high prediction accuracy. The prediction method proposed in this paper is based on the network method rather than the classical statistical method, which makes full use of the characteristics of the nodes themselves in the network and provides a new way of thinking for nonlinear time series prediction.

Key words: time series prediction; empirical modal decomposition; directed visibility graph; ink prediction