A lightweight reconstruction method for missing spatio-temporal data

Abstract

The invention discloses a lightweight missing spatio-temporal data reconstruction method. The overall steps are as follows: 1, performing spatio-temporal data representation: Abstracting the point data and the mesh data of the static reference into a unified space-time state matrix for representation; 2, performing time dimension interpolation: Introducing an average correlation coefficient to automatically select a time window so as to improve the modeling time dependence capability of the SES algorithm; 3, performing spatial dimension interpolation: respectively adopting constant equal distance and a correlation distance based on a Gaussian function to assign a weight to each spatial neighbor so as to improve the capacity of modeling spatial dependence of the IDW algorithm; 4, performingspace-time integration: introducing an extreme learning machine as a learning algorithm of the neural network model, and integrating an estimation result of a space-time dimension to obtain a final predicted value of missing data. According to the method, a plurality of improved lightweight models are integrated, so that the reconstruction precision of mass missing spatio-temporal data is furtherimproved on the premise that the calculation efficiency of the reconstruction algorithm is ensured.

Description

technical field [0001] The invention relates to a data reconstruction method, in particular to a lightweight reconstruction method for missing spatio-temporal data, and belongs to the technical field of spatio-temporal data mining. Background technique [0002] With the continuous popularization and development of sensor networks and mobile positioning technology, the extension of data acquisition and computing units has continued to expand, and earth science has experienced a major revolution from a data-poor field to a data-rich field. These data continue to grow in time and space dimensions, resulting in massive spatio-temporal data. Although the gradual expansion of data scale makes the input information of spatiotemporal data analysis more and more abundant, and the analysis results are correspondingly more accurate, however, the lack of spatiotemporal data is still a common problem faced by the current geospatial big data collection and mining. [0003] There are seve...

AI Technical Summary

Problems solved by technology

However, the traditional neural network learning algorithm (such as BP algorithm) usually only emphasizes its nonlinear fitting ability, and does not consider its slow training speed and easy to fall into local optimal solution...

Images