Time sequence prediction method based on vector autoregression and convolution

The invention relates to the technical field of sequential sequence prediction, in particular to a sequential sequence prediction method based on vector autoregression and convolution, and the method specifically comprises the following steps: 1, carrying out the data preprocessing of the original data of a sequential sequence; step 2, mapping the time sequence data into linear sequence data and nonlinear sequence data by using vector autoregression mapping; 3, coding and normalizing the linear sequence data and the nonlinear sequence data respectively; 4, the data enter a multi-scale feature extraction neural network model, and feature extraction of different scales is carried out on linear sequence data and nonlinear sequence data by utilizing cavity causal convolution; and 5, carrying out dynamic weight addition on the final scale characteristics of the linear sequence data and the nonlinear sequence data, and outputting a final prediction label after passing through a linear layer. The technical problems