Patent · US Active

Method for constructing prediction model of auto trips quantity and prediction method and system

US11995982B2 · kind B2 · utility

0Cited by
3References
5Claims
0Family size

Assignee

Inventors

Key dates

Filing dateDec 10, 2020
Grant dateMay 28, 2024
Priority date
Expiry dateJan 10, 2042

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG08G1/0104
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

A method for constructing a prediction model of an auto trips quantity and a prediction method and system are disclosed. The prediction model construction method designs a deep neural network Multitask GCN-LSTM based on GCN and LSTM for predicting the auto trips quantity. The deep neural network comprises three modules, wherein the three modules are respectively used for extracting a spatial correlation, a temporal correlation and a feature fusion. The prediction method and system predict the auto trips quantity based on a model constructed. By considering a road segment local relationship and a road segment global relationship and taking an auto arrival quantity as a related task in constructing the model, the prediction model construction method uses a multi-task learning method to avoid overfitting of the deep neural network and reduce a prediction error of the auto trips quantity effectively.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.