Abstract:Accurate traffic flow prediction is of great significance to intelligent transportation systems (ITS). The long short-term memory model (LSTM) and gradient boosting regression (GBR) model are widely used in this field and have achieved significant achievements. However, the inputs of these models have a strong influence on the model outputs. Traffic states of roads are correlated spatially, thus the prediction models that only consider features of the objective road segment is sufficient. To improve the prediction accuracy, this paper presents a framework to appropriately refine the short-term traffic prediction results by considering temporal and spatial correlations. In the framework, we use the temporal detrending, top-k correlated and top-k neighboring methods in the spatial dimension, which consider temporal and spatial correlations, to enhance the input features of the prediction models. To verify the prediction accuracy, we use the probe vehicle GPS traces extracted from the on-demand ride service platform of DiDi between August 1, 2016, and September 25, 2016, in Beijing, China, as the dataset in this paper. The first 7-weeks of data are utilized as the training set and the remaining 1-week of data are used as the test set. The results show that the temporal detrending and spatial correlation method can effectively improve the prediction accuracy of both LSTM and GBR. Besides, we compare the top-k correlated method to the top-k neighboring method to show their pros and cons.

Comparative Analysis for Traffic Flow Forecasting Models with Real-Life Data in Beijing

Traffic Flow Prediction Based on Probe Vehicle GPS Traces Considering Temporal and Spatial Correlations

Application Improvement of Nonparametric Regression to Traffic Flow Forecast

Fuzzy Hybrid Framework with Dynamic Weights for Short‐term Traffic Flow Prediction by Mining Spatio‐temporal Correlations

The Combined Short-Term Forecasting Approach to Traffic Flow Based on Non-Parametric Regression

Short-Term Traffic Flow Forecasting by Selecting Appropriate Predictions Based on Pattern Matching

Long-Term Forecasting Oriented to Urban Expressway Traffic Situation

Combined Short-Term Traffic Flow Forecast Model For Beijing Traffic Forecast System

Study of a method for real-time dynamic forecast of traffic volume

Forecasting Transport Mode Sharing in Beijing

Study on effect of flow forecasting error on traffic control performance

Study on Short-term Traffic Flow Combined Forecast Model

Short-term Traffic Flow Forecasting of Urban Network Based on Dynamic STARIMA Model

A Short-term Traffic Flow Forecasting Method Based on Combination of Fuzzy C-mean Clustering and Neural Network

Real-Time Corrected Traffic Correlation Model for Traffic Flow Forecasting

A Comparative Study of Prediction Models on the Duration of Traffic Incidents

Traffic network flow forecasting based on switching model

Short-Term Traffic Forecast System of Beijing

A Novel Networked Traffic Parameter Forecasting Method Based On Markov Chain Model

The Simulation Research of Non-parametric Regression for Short-Term Traffic Flow Forecasting

Big Data-Driven Based Real-Time Traffic Flow State Identification and Prediction