Traffic flow predicting method based on Conv1D-LSTM neural network structure

Abstract

The invention discloses a traffic flow predicting method based on a Conv1D-LSTM neural network structure. The method includes the following steps of firstly, preprocessing road traffic flow data and establishing a road traffic flow state matrix data set; secondly, obtaining traffic flow states of different road sections at the same moment and extracting the space features of the traffic flow data;thirdly, extracting data time features on the basis of the traffic flow data including the space features, wherein the traffic flow space features output by a one-dimensional convolution network serves as the input of an LSTM neutral network, and the time features in the road traffic flow data are further extracted; fourthly, using road tariff flow time and space features as the input of a regression predicting layer, calculating a predicting result corresponding to the current input, defining a model loss function, continuously optimizing model parameters through a counterpropagation algorithm according to the value of the loss function, and obtaining real-time traffic flow data to serve as model input for predicting the real-time road traffic flow. By means of the method, the traffic flow short-period predicting accuracy is improved.

Description

technical field [0001] The invention relates to a traffic flow prediction method based on a Conv1D-LSTM neural network structure, and belongs to the field of traffic prediction. Background technique [0002] With the continuous improvement of the socio-economic level and the continuous acceleration of the pace of life, people's demand for vehicles is also increasing, followed by serious traffic jams. How to effectively alleviate traffic congestion and allocate traffic resources more efficiently has become a top priority. The emergence of intelligent transportation system has effectively solved these problems to a certain extent, and road traffic flow prediction as a part of intelligent transportation system plays an irreplaceable role in this process. [0003] The existing road traffic flow prediction methods mainly include: time series method, Markov prediction, Kalman filter method, support vector machine, BP neural network, etc. Some of these methods are based on probabi...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the shortcomings of the low accuracy of existing traffic flow forecasting, the present invention provides a traffic flow forecasting method based on the Conv1D-LSTM (one-dimensional convolution and long short-term memory) neural network structure, which uses one-dimensional convolution and The long-short-term memory neural network obtains the spatial and temporal information in the road traffic flow data, fully excavates the spatio-temporal characteristics of the road traffic flow data, overcomes the shortcomings of the existing methods for feature extraction, and thus improves the short-term prediction of traffic flow. accuracy

Images