Due to the high operation and maintenance cost of urban road monitoring equipment, the number of road installation equipment is limited, so it is necessary to use observable data and its potential spatiotemporal dependence to estimate the data of no sensor location to solve the problem of data sparsity, so as to realize the macro management of the road network. The main challenge of this problem is how to effectively model and exploit the spatiotemporal dependencies in the data. The existing models mainly use GNNs to model the spatiotemporal dependence, but this only considers the local spatiotemporal relationship of each node. Therefore, this paper proposes an AIGNN model, which aggregates the information of local neighbor nodes through graph neural network, and then captures the global spatiotemporal dependencies of each node through the attention mechanism, and uses the induction strategy to train the model to solve the spatiotemporal Kriging task. Experiments are carried out on three real-world spatiotemporal datasets, and the results show that AIGNN has performance advantages over the traditional linear model and the GNN-based Kriging model.