Research For Data-driven Methods Of Short-term Traffic Prediction On Urban Arterial Road Intersections

With the rapid development of communication,control,and network technology,Intelligent Transportation Systems(ITS)have gradually become a primary means of alleviating traffic congestion and other related problems.Short-term traffic volume prediction is one of the critical technologies of ITS.How to use intelligent prediction models to make reliable predictions of traffic conditions and improve traffic operation efficiency has become a focus of many scholars.However,short-term traffic volume prediction models still need to be optimized and improved,and their prediction accuracy and reliability have not yet met practical requirements.This article is devoted to improving the accuracy and reliability of short-term traffic volume prediction at urban arterial intersections,systematically studying short-term traffic volume prediction from four aspects: single-step and multi-step,determinism and uncertainty,and constructing corresponding prediction models.Taking the short-term traffic volume data of three entrance roads(entrances A,B,and C)at an urban arterial intersection in Chongqing as an example,the effectiveness of the methods proposed in this article is analyzed in detail.The results show that the proposed model has high prediction accuracy and reliability and can provide technical support for future traffic status judgment.The specific research contents are as follows:1.Preprocessing and statistical analysis of measured data.Data loss,repetition,and errors are inevitable in collecting and storing short-term traffic volume data.Directly using incorrect data for analysis will inevitably lead to severe distortion of the prediction model results.At the same time,the prediction model’s performance is closely related to the data characteristics.Studies have shown that due to the influence of complex traffic operation environments,short-term traffic volume data has complex features and often exhibits significant nonlinearity and non-stationarity.In this article,the short-term traffic volume data is first preprocessed by removing duplicate data and repairing incorrect and missing data.Secondly,based on statistical analysis methods,the statistical characteristics of different types of traffic volume data are identified.Then,several commonly used data decomposition methods are introduced,and their advantages and disadvantages are clarified.Finally,according to the statistical data characteristics and the performance characteristics of the prediction model,the applicability of different characteristic data to the prediction model is explored,providing theoretical support for model selection and construction.2.Construction of a deterministic prediction model.The prediction accuracy is a direct manifestation of the prediction model’s performance and a measure of the certainty of the future evolution of traffic volume.To further improve the accuracy of short-term traffic volume prediction,this paper proposes a new hybrid prediction model based on data decomposition techniques with high nonlinearity and non-stationary interpretability.Firstly,the time-varying filtering empirical mode decomposition(TVF-EMD)and local mean decomposition(LMD)are fused to establish a new method of data secondary decomposition(TVF-EMD-LMD)to analyze the nonlinear and non-stationary components of short-term traffic volume data in detail.Secondly,an extreme learning machine(ELM)model with solid learning ability is established for each decomposed subsequence for training and prediction.Finally,the predicted results of all sub-sequences are superimposed to obtain the final prediction result,and the model’s prediction performance is evaluated systematically.Case studies show that this hybrid model has high accuracy in short-term prediction.3.Construction of an uncertainty prediction model.To further meet the needs of traffic risk management,the prediction model also needs to provide highly reliable uncertainty prediction results by quantifying the impact of uncertain factors on the prediction results.Based on the aforementioned deterministic prediction model,this paper combines the CKDE model to research uncertainty prediction.Firstly,the relevant theories of uncertainty prediction methods are expounded,and an adequate measure of uncertainty factors’ influence,i.e.,interval prediction,is determined.Secondly,the advantages of the target CKDE model over other prediction models are clarified,and a data secondary decomposition-based uncertainty prediction model(i.e.,TVF-EMDLMD-ELM-CKDE)is established.Finally,the reliability of the uncertainty prediction model is evaluated using indicators such as the prediction interval coverage rate and the mean coverage error.Case studies show this model can provide highly reliable singlestep uncertainty prediction results.4.Construction of multi-step prediction model.Single-step prediction can only provide forecast results one step ahead(with less available data).However,multi-step prediction can provide the evolving pattern of traffic volume in the future for a more extended period,which can give more data support for traffic management and control.Based on secondary data decomposition,this paper combines Bi-LSTM and CKDE models to construct a short-term traffic volume multi-step prediction model,which simultaneously achieves both deterministic and uncertain predictions.Specifically,first,the Bi-LSTM model and relevant theories are explained to illustrate this model’s superiority in the multi-step deterministic forecast field.Secondly,based on the secondary decomposition,the Bi-LSTM multi-step deterministic prediction model is established,and the detailed solution process is given.Then,the Bi-LSTM model is combined with CKDE to construct the multi-step uncertain prediction model.Finally,eight evaluation indicators are selected to evaluate the accuracy and reliability of the model.Case studies show that the prediction model constructed in this paper has high accuracy and reliability in multi-step prediction.