Short-Term Vehicle Traffic Flow Forecasting Grey Model for Intelligent Transportation System Performance

Abstract

Vehicular traffic is continuously increasing around the world and the resulting congestion and pollution is a major concern to transportation specialists and decision makers. As population continues to grow it is a challenge to handle traffic demand, traffic jam, CO2 emission, global warming and economic loss. Road capacity is not adequate. Roads and highways are unlikely to expand due to cost and dwindling land supply. To manage these issues it is critical to integrate intelligent transportation system (ITS) in the transport management systems. Short-term vehicle traffic flow forecasting by ITS is vital in proactively monitoring a vehicle traffic system. Unfortunately, effective traffic flow forecasting is a key problem of ITS. Therefore, performance improvement of the predictive models which can enhance the forecasting ability of ITS is very crucial. Hence the objective of this study was to develop a short-term vehicle traffic flow forecasting grey model (GM) for ITS performance. Hence, in this thesis the precision of the original GM is improved by three proposed methods namely data grouping technique (DGT), relative variable smoothing approach (RVSA) and a three-step approach (TSA). To further improve the GM’s precision these new methods were combined with existing methods such as modification of background value (MBV), modification of initial condition (MIC) and Fourier series error correction approach (FSECA). Consequently, hybrid grey models were established. The accuracy improvement on the conventional grey models were measured by employing measures of model performance, namely root mean square error (RMSE), root mean square percentage error (RMSPE), mean absolute error (MAE) and the mean absolute percentage deviation (MAPD). The evaluation results revealed that the hybrid grey models outperformed the conventional GM in vehicle flow modelling and short-term forecasting. For instance in short-term vehicle traffic flow forecasting the improved models (GGM(1,1) and MBVGGM(1,1)) had good accuracy in the range of 80-90% compared to the corresponding conventional GM(1,1) and MBVGM(1,1) which had reasonable accuracy in the range of 50-80%. On the other hand in validating the DGT in improving the fitting accuracy of the conventional GM(1,3) the accuracy was improved from 60.3270% to 96.9706%. This was great improvement in the conventional GM(1,3)’s fitting accuracy. Further, the results of this research show that the proposed new methods i.e. the DGT, the RVSA and the TSA methods have the potential for improving the prediction accuracy of the conventional GMs. Hence the DGT in hybrid grey models can enhance the short-term forecasting ability of the ITS. A case study based on traffic data collected from Nairobi city, Kenya, was presented and analyzed to show the accuracy improvement in both the univariate (GM(1,1)) and multivariate (GM(1,3)) grey models. For instance from this case study computation of the RMSPE had shown that the fitting accuracy of GM(1,3) was improved from 69.7243% to 99.6281% by the TSA method. Thus an improved multivariate grey model can attain high traffic flow forecasting accuracies compared with an improved univariate grey model. Finally, the performance of the grouping technique based GMs on energy consumption and carbon dioxide emissions, outperformed the conventional GMs. From one of the presented empirical cases the grouping technique based multivariate GGM(1,3) attained an accuracy of 96.9706% against 60.3270% of the conventional GM(1,3). Thus the hybrid grey models developed in this thesis are multidisciplinary. However, in comparison with other state of the art improved GM such as the grey model with cosine term (GM(1,1|cos(ωt))), the performance of the proposed models was below that of the GM(1,1|cos(ωt)). In a recent research GM(1,1|cos(ωt)) had a mean absolute percentage error (MAPE) of 0.1% compared to 0.58% of the original GM(1,1). Therefore, there is need to investigate the performance of the proposed models in this research in comparison with the GM(1,1|cos(ωt)), in the future.