Deep learning model-based traffic speed prediction method

Abstract

The invention discloses one deep learning model-based traffic speed prediction method, which comprises the following steps of building a GCLSTM model; introducing a GCN model to carry out graph convolution operation on a cell layer state and a hidden layer state separately by taking an Seq2Seq model as a model basis; inputting a traffic speed of a to-be-predicted road section in a previous periodof time into the GCLSTM model; and outputting a predicted traffic speed in the next period of time through calculation. The invention further discloses the other deep learning model-based traffic speed prediction method, which comprises the following steps of building a GLAT model; introducing a temporal attention mechanism to pay attention to a hidden layer vector of an encoder at each moment bytaking the Seq2Seq model as the model basis; inputting the traffic speed of the to-be-predicted road section in the previous period of time into the GLAT model; and outputting the predicted traffic speed in the next period of time through calculation. According to the two traffic speed prediction methods, the traffic speed can be accurately predicted.

Description

technical field [0001] The invention belongs to the field of traffic monitoring, and in particular relates to a traffic speed prediction method based on a deep learning model. Background technique [0002] Real-time and accurate traffic speed prediction is a fundamental yet challenging task in intelligent transportation systems. The traffic network in the real world is composed of a large number of criss-crossing roads, and the traffic of each specific road segment may be affected by its adjacent road segments. The geography of a road affects its traffic conditions. For example, traffic patterns on major and minor roads are different, and traffic jams often occur at intersections. However, traditional traffic prediction models aim to learn the time dependence within each road, which only predicts the future traffic speed of a road based on the previous traffic speed observed from each road. [0003] If directly used as attached figure 1 The Seq2Seq model shown is used to...

AI Technical Summary

Problems solved by technology

Therefore, the influence of spatial dependence between adjacent road segments is ignored, or the information sharing between related roads is ignored.

[0004] In addition, there is a very serious problem in the Seq2Seq model: regardless of the length of the input sequence, the encoder will be encoded into a fixed-length feature vector representation, and the decoder will be limited to this fixed-length vector representation

This problem limits the performance of the model. When the sequence is input, as the sequence continues to grow, the performance of the original time step method is getting worse and worse. This is due to the structure of the original encoder-decoder model design. Defects, especially when the input sequence is relatively long, it is difficult for the model to learn a reasonable vector representation, so the performance of the model will become poor

Images