Spatio-temporal Autoregressive Neural Network Interpolation Method

Spatio-temporal interpolation is an extension of spatial interpolation and an important basis for analyzing the characteristics of the spatio-temporal distribution of data.Traditional spatio-temporal interpolation method based on mathematical model is limited by the model structure,the parameter selection is easily affected by subjective factors,and it is difficult to fit the complex nonlinear relationship between spatiotemporal distance and spatio-temporal weights.For this reason,this paper constructs a spatio-temporal autoregressive neural network interpolation method,which uses neural network to fit spatio-temporal weights to improve the accuracy of spatio-temporal interpolation.The main research contents of this paper are as follows:(1)In view of the subjective influence of model selection and parameter estimation in traditional spatio-temporal interpolation methods,it is difficult to fit the complex nonlinear relationship between spatio-temporal distance and spatio-temporal weights,a spatio-temporal autoregressive neural network interpolation method is designed,using the self-learning and abstract ability of the neural network for nonlinear relationships to realize the mapping from time distance and space distance to spatio-temporal weights,avoiding the influence of complicated parameter estimation process and subjective factors.(2)Combining classic neural network training and optimization strategies,and according to the structural characteristics of the model network,a training framework of the spatio-temporal autoregressive neural network is designed,including the overall training process,optimization strategies,activation function selection and parameter initialization settings to improve the model’s training efficiency and interpolation accuracy.(3)Taking the inorganic nitrogen and temperature in the water quality monitoring data of Zhejiang coastal waters as the research object,the comparison experiments between the spatio-temporal autoregressive neural network interpolation method and the existing interpolation methods represented by spatial kriging and spatio-temporal inverse distance weighting is designed to verifiy the effectiveness and applicability of this method.