Abstract Crude oil, as an important energy source, plays a pivotal role in global economic development. However, due to the dynamic nature and characteristics of crude oil data, which are affected by the uncertainty of market volatility and complex price change mechanisms, the existing crude oil price forecasting methods are ineffective. Therefore, we propose a novel crude oil price combinatorial link prediction method based on the MF-VMD-CDVGNets-SLRNN model, which converts crude oil price data into a periodic directed visible graph network. Compared with traditional forecasting methods, it can comprehensively analyze multi-period historical data and current data, derive the intrinsic characteristics of the data, and solve the problem of information loss in historical data. First, the original data are denoised and then decomposed and reconstructed into a sequence of components with different frequencies. The periodicity of each component is determined by applying the Fast Fourier Transform (FFT). Next, a Cyclic Directed Visibility Graph Network (CDVGNet) is employed to convert these components into a network that reflects the periodic structure of the time series. A random walk algorithm is then applied to measure the similarity between nodes across different temporal periods. In addition to extracting multi-period historical and current data features as prediction inputs, the method optimizes the convergence of the algorithm by controlling the network size. Finally, the combined prediction utilizes multiple artificial intelligence models, taking into account the nonlinear relationship between nodes. To validate the effectiveness of the proposed method, we conducted a West Texas Intermediate (WTI) crude oil price prediction experiment. The results show that the proposed method has higher prediction accuracy compared to other baseline methods.
Zhao et al. (Fri,) studied this question.