Traffic flow model training method based on attention mechanism

Abstract

The invention discloses a traffic flow model training method based on an attention mechanism, being characterized in that a pre-trained and fused model is applied as a building block of a deep architecture model to measure traffic data for prediction, and a pre-trained station model having a plurality of fused layer architectures is proposed, and the model takes into account traffic network structure and traffic status of each station to predict traffic speed over a network range. The traffic flow model training method captures spatial features and temporal dependencies from historical data using multiple fusion layers, and the proposed model can process missing values in input data by using a masking mechanism. Experiments are established in a real data set, and compared with other classic and most advanced models, results show that the model of the invention is superior to other models in terms of accuracy and robustness.

Description

technical field [0001] The invention belongs to the technical field of data analysis, and in particular relates to a traffic flow model training method based on an attention mechanism. Background technique [0002] In the past ten years, the economy has developed rapidly, and the number of motor vehicles has also increased rapidly. By the end of 2017, the number of motor vehicles in the country has exceeded 200 million. It is estimated that by 2020, the number of motor vehicles will reach 300 million. A large number of motor vehicles And traffic demand has brought a series of problems such as traffic congestion and parking difficulties. At present, the road and traffic management departments mainly adopt the following methods to solve the traffic congestion problem: (1) strengthen the construction of road infrastructure, such as widening roads, building new roads, etc.; (2) developing intelligent transportation systems to implement intelligent management. [0003] The intel...

AI Technical Summary

Problems solved by technology

[0006] Although there are various deep learning solutions available in traffic forecasting problems, existing methods still suffer from some shortcomings of previous methods, which only build models for time series data without considering these spatial relationships; on the other hand, It is difficult to determine the selection boundary of traffic flow characteristics in previous prediction methods, resulting in insufficient or misleading information, which significantly restricts the accuracy of traffic flow prediction

Some methods predict speed based on historical data, but historical data only focus on the segment itself, these methods predict traffic speed based on time series values, ignoring the characteristics of the segment and the influence of nearby segments on it

Images