Method and apparatus for constructing spatio-temporal neural network as well as method and apparatus for carrying out prediction by using spatio-temporal neural network

Abstract

The present disclosure relates to methods and devices for building spatio-temporal neural networks and making predictions using them. According to an exemplary embodiment of the present disclosure, the method for constructing a spatio-temporal neural network may include: clustering a spatio-temporal sequence data set to divide the spatio-temporal sequence data set into a plurality of sub-regions in a spatial region; determining the the spatio-temporal correlation of each sub-area in the plurality of sub-areas; and constructing the spatio-temporal neural network based on the spatio-temporal correlation of each sub-area. According to the present disclosure, a heterogeneous spatiotemporal neural network is constructed by considering the heterogeneity of spatiotemporal sequence data. Using this method, the spatiotemporal model can be constructed more reasonably, and can improve the prediction accuracy of spatiotemporal sequence data while improving the fitting performance of training data.

Description

technical field [0001] The present disclosure relates to neural network technology, and more particularly to a method and device for constructing a spatio-temporal neural network and a method and device for predicting using a spatio-temporal neural network. Background technique [0002] Spatiotemporal variables are quantities that change with time or spatial location. There are massive time-space series data accumulated in the fields of environment, meteorology, transportation, economy, and public health. These time-space series usually describe the change of a certain variable over time in different spatial locations. For example, 35 air monitoring stations set up in a city The PM2.5 concentration value of the site within 24 hours of a certain day, the daily average temperature of each city in a certain province in October 2014, the total annual GDP of each province from 2000 to 2013, etc. Time-space sequence modeling can describe the correlation between space-time variabl...

AI Technical Summary

Problems solved by technology

[0007] However, spatio-temporal data are not always homogeneous, but often heterogeneous, that is, temporal and / or spatial correlations of different categories of spatio-temporal data are different

However, the above spatio-temporal neural network construction method uses the same processing method for all data, so the spatio-temporal neural network is not suitable for heterogeneous spatio-temporal data.

Images