Aviation flow prediction method

Abstract

The invention belongs to the field of aviation flow state prediction, and relates to an aviation flow prediction method based on a space-time diagram convolutional neural network and height hierarchical weighting. The method comprises the following steps: constructing a space-time attention module based on a space attention mechanism and a time attention mechanism, and applying different weights to nodes near waypoints; constructing a space-time convolution module based on the graph convolution in the space dimension and the standard convolution in the time dimension; constructing a space-timegraph convolutional network by utilizing a space-time attention module and a space-time convolution module; learning the constructed space-time diagram convolutional network by using aviation flow historical data to obtain different influence parameters of the three parts, the aviation flow historical data comprising three characteristic values: speed, flow and time occupancy; and inputting the aviation flow historical data of each node of different height layers into the space-time diagram convolutional network according to different flight layers to obtain the aviation flow prediction of each height layer, and then predicting the overall aviation flow through a weighting algorithm.

Description

technical field [0001] The invention belongs to the field of air traffic state prediction, and relates to an air flow prediction method based on a spatio-temporal graph convolutional neural network and highly hierarchical weighting. Background technique [0002] In recent years, with the rapid development of the economy, the air traffic flow has increased accordingly, and the air traffic has become more complex and uncontrollable, followed by the problem of air traffic jams. Therefore, the forecast of air traffic flow has become very necessary. The essence of predicting future traffic with the help of historical traffic flow values ​​is to sort the time series of traffic flow. There are some current research, for example, using multiple linear regression for inference, although the method is simple, but the accuracy is not high; using the least squares support vector machine, using equality constraints instead of inequality constraints, the quadratic programming problem bec...

AI Technical Summary

Problems solved by technology

There are some current research, for example, using multiple linear regression for inference, although the method is simple, but the accuracy is not high; using the least squares support vector machine, using equality constraints instead of inequality constraints, the quadratic programming problem becomes linear Equation solving problem, the algorithm is more advanced and has better generalization ability, but the error in medium and long-term prediction is relatively large

[0003] In the field of civil aviation, aircraft of different types, sizes, and headings have different flight altitudes. Therefore, when using graph convolutional neural networks to calculate route traffic, it is necessary to consider how to weight different flight levels. Research

Images