Travel demand prediction method based on space-time feature extraction

Abstract

The invention discloses a travel demand prediction method based on space-time feature extraction. The method comprises the following steps of collecting non-aggregated travel data and preprocessing the non-aggregated travel data, constructing a travel demand grid chart, constructing a travel demand prediction model based on space-time feature extraction, constructing a time sequence of the local travel demand grid chart, constructing the time sequence of the local travel demand grid chart into a spatial feature training sample set Ds, and constructing a time feature training sample set DT, extracting space-time features, constructing a prediction training sample set DP, and training a prediction module by using the prediction training sample set DP, and splicing the trained space feature extraction module, time feature extraction module and prediction module to obtain an end-to-end prediction model, and taking the end-to-end prediction model as a final travel demand prediction model based on space-time feature extraction. The method improves the prediction precision of travel demands.

Description

technical field [0001] The invention relates to the field of traffic demand forecasting, in particular to a travel demand forecasting method based on spatio-temporal feature extraction. Background technique [0002] In recent years, new forms of travel such as online car-hailing and shared bicycles have gradually become popular, and people's demand for new transportation resources is also increasing day by day. Therefore, it is becoming more and more important to better dispatch traffic resources, reduce enterprise operating costs, and improve the quality of travel services by accurately predicting travel demand. [0003] Among the existing travel demand forecasts, some use traditional linear models and machine learning methods, and some use deep learning methods. The existing deep learning models combine and utilize deep learning structures such as convolutional neural networks, recurrent neural networks and their variant long-short-term memory neural networks, which to a ...

AI Technical Summary

Problems solved by technology

The existing deep learning models combine and utilize deep learning structures such as convolutional neural networks, recurrent neural networks and their variant long-short-term memory neural networks, which to a certain extent play a role in capturing the temporal and spatial correlations in traffic data. However, most of the time-space related attributes have been directly input into the model, without considering the extraction of space-time features.

Images