A Traffic Data Imputation Method Based on a Context-Enhanced Graph Network

The present invention discloses a traffic data imputation method based on a context-enhanced graph network, relating to the technical field of electrical data processing. The method comprises the following steps: organizing multi-source traffic data into a sample-variable data matrix and constructing a bipartite graph; establishing relation-enhanced edges between sample nodes according to traffic-domain relationships, such that a graph neural network generates baseline imputation values; extracting observed variables within the same sample as contextual information and aggregating them into a context vector, based on which a context adapter generates residual correction terms and gating coefficients; selectively adding the residual correction terms to the baseline imputation values through an adaptive fusion coefficient jointly controlled by observation coverage and the global missing rate, wherein the method falls back to the baseline values when the coverage is below a predefined threshold; and, after traversing all missing entries, outputting a complete data matrix. By integrating inter-sample traffic-domain relationships with intra-sample observed context and employing an adaptive gated residual mechanism for selective correction, the present invention improves the accuracy and stability of missing-data imputation for multi-source traffic data.