Building trust for traffic flow forecasting components in intelligent transportation systems via interpretable ensemble learning

Traffic flow forecasting constitutes a crucial component of intelligent transportation systems(ITSs).Numerous studies have been conducted for traffic flow forecasting during the past decades.However,most existing studies have concentrated on developing advanced algorithms or models to attain state-of-the-art forecasting accuracy.For real-world ITS applications,the interpretability of the developed models is extremely important but has largely been ignored.This study presents an interpretable traffic flow forecasting framework based on popular tree-ensemble algorithms.The framework comprises multiple key components integrated into a highly flexible and customizable multi-stage pipeline,enabling the seamless incorporation of various algorithms and tools.To evaluate the effectiveness of the framework,the developed tree-ensemble models and another three typical categories of baseline models,including statistical time series,shallow learning,and deep learning,were compared on three datasets collected from different types of roads(i.e.,arterial,expressway,and freeway).Further,the study delves into an in-depth interpretability analysis of the most competitive tree-ensemble models using six categories of interpretable machine learning methods.Experimental results highlight the potential of the proposed framework.The tree-ensemble models developed within this framework achieve competitive accuracy while maintaining high inference efficiency similar to statistical time series and shallow learning models.Meanwhile,these tree-ensemble models offer interpretability from multiple perspectives via interpretable machine-learning techniques.The proposed framework is anticipated to provide reliable and trustworthy decision support across various ITS applications.

Hourly traffic flow forecasting using a new hybrid modelling method

Short-term traffic flow forecasting is a significant part of intelligent transportation system.In some traffic control scenarios,obtaining future traffic flow in advance is conducive to highway management department to have sufficient time to formulate corresponding traffic flow control measures.In hence,it is meaningful to establish an accurate short-term traffic flow method and provide reference for peak traffic flow warning.This paper proposed a new hybrid model for traffic flow forecasting,which is composed of the variational mode decomposition(VMD)method,the group method of data handling(GMDH)neural network,bi-directional long and short term memory(BILSTM)network and ELMAN network,and is optimized by the imperialist competitive algorithm(ICA)method.To illustrate the performance of the proposed model,there are several comparative experiments between the proposed model and other models.The experiment results show that 1)BILSTM network,GMDH network and ELMAN network have better predictive performance than other single models;2)VMD can significantly improve the predictive performance of the ICA-GMDH-BILSTM-ELMAN model.The effect of VMD method is better than that of EEMD method and FEEMD method.To conclude,the proposed model which is made up of the VMD method,the ICA method,the BILSTM network,the GMDH network and the ELMAN network has excellent predictive ability for traffic flow series.

Research on traffic flow forecasting model based on cusp catastrophe theory

This paper intends to describe the relationship between traffic parameters by using cusp catastrophe theory and to deduce highway capacity and corresponding speed forecasting value through suitable transformation of catastrophe model. The five properties of a catastrophe system are outlined briefly, and then the data collected on freeways of Zhujiang River Delta, Guangdong province, China are examined to ascertain whether they exhibit qualitative properties and attributes of the catastrophe model. The forecasting value of speed and capacity for freeway segments are given based on the catastrophe model. Furthermore, speed-flow curve on freeway is drawn by plotting out congested and uncongested traffic flow and the capacity value for the same freeway segment is also obtained from speed-flow curve to test the feasibility of the application of cusp catastrophe theory in traffic flow analysis. The calculating results of catastrophe model coincide with those of traditional traffic flow models regressed from field observed data, which indicates that the deficiency of traditional analysis of relationship between speed, flow and occupancy in two-dimension can be compensated by analysis of the relationship among speed, flow and occupancy based on catastrophe model in three-dimension. Finally, the prospects and problems of its application in traffic flow research in China are discussed.

A Short-Term Traffic Flow Forecasting Method Based on a Three-Layer K-Nearest Neighbor Non-Parametric Regression Algorithm

Short-term traffic flow is one of the core technologies to realize traffic flow guidance. In this article, in view of the characteristics that the traffic flow changes repeatedly, a short-term traffic flow forecasting method based on a three-layer K-nearest neighbor non-parametric regression algorithm is proposed. Specifically, two screening layers based on shape similarity were introduced in K-nearest neighbor non-parametric regression method, and the forecasting results were output using the weighted averaging on the reciprocal values of the shape similarity distances and the most-similar-point distance adjustment method. According to the experimental results, the proposed algorithm has improved the predictive ability of the traditional K-nearest neighbor non-parametric regression method, and greatly enhanced the accuracy and real-time performance of short-term traffic flow forecasting.

Improved Social Emotion Optimization Algorithm for Short-Term Traffic Flow Forecasting Based on Back-Propagation Neural Network

The back-propagation neural network(BPNN) is a well-known multi-layer feed-forward neural network which is trained by the error reverse propagation algorithm. It is very suitable for the complex of short-term traffic flow forecasting; however, BPNN is easy to fall into local optimum and slow convergence. In order to overcome these deficiencies, a new approach called social emotion optimization algorithm(SEOA) is proposed in this paper to optimize the linked weights and thresholds of BPNN. Each individual in SEOA represents a BPNN. The availability of the proposed forecasting models is proved with the actual traffic flow data of the 2 nd Ring Road of Beijing. Experiment of results show that the forecasting accuracy of SEOA is improved obviously as compared with the accuracy of particle swarm optimization back-propagation(PSOBP) and simulated annealing particle swarm optimization back-propagation(SAPSOBP) models. Furthermore, since SEOA does not respond to the negative feedback information, Metropolis rule is proposed to give consideration to both positive and negative feedback information and diversify the adjustment methods. The modified BPNN model, in comparison with social emotion optimization back-propagation(SEOBP) model, is more advantageous to search the global optimal solution. The accuracy of Metropolis rule social emotion optimization back-propagation(MRSEOBP) model is improved about 19.54% as compared with that of SEOBP model in predicting the dramatically changing data.

Flow Direction Level Traffic Flow Prediction Based on a GCN-LSTM Combined Model

Traffic flow prediction plays an important role in intelligent transportation systems and is of great significance in the applications of traffic control and urban planning.Due to the complexity of road traffic flow data,traffic flow prediction has been one of the challenging tasks to fully exploit the spatiotemporal characteristics of roads to improve prediction accuracy.In this study,a combined flow direction level traffic flow prediction graph convolutional network(GCN)and long short-term memory(LSTM)model based on spatiotemporal characteristics is proposed.First,a GCN model is employed to capture the topological structure of the data graph and extract the spatial features of road networks.Additionally,due to the capability to handle long-term dependencies,the longterm memory is used to predict the time series of traffic flow and extract the time features.The proposed model is evaluated using real-world data,which are obtained from the intersection of Liuquan Road and Zhongrun Avenue in the Zibo High-Tech Zone of China.The results show that the developed combined GCNLSTM flow direction level traffic flow prediction model can perform better than the single models of the LSTM model and GCN model,and the combined ARIMA-LSTM model in traffic flow has a strong spatiotemporal correlation.

Short-term traffic flow online forecasting based on kernel adaptive filter

Considering that the prediction accuracy of the traditional traffic flow forecasting model is low,based on kernel adaptive filter(KAF)algorithm,kernel least mean square(KLMS)algorithm and fixed-budget kernel recursive least-square(FB-KRLS)algorithm are presented for online adaptive prediction.The computational complexity of the KLMS algorithm is low and does not require additional solution paradigm constraints,but its regularization process can solve the problem of regularization performance degradation in high-dimensional data processing.To reduce the computational complexity,the sparse criterion is introduced into the KLMS algorithm.To further improve forecasting accuracy,FB-KRLS algorithm is proposed.It is an online learning method with fixed memory budget,and it is capable of recursively learning a nonlinear mapping and changing over time.In contrast to a previous approximate linear dependence(ALD)based technique,the purpose of the presented algorithm is not to prune the oldest data point in every time instant but it aims to prune the least significant data point,thus suppressing the growth of kernel matrix.In order to verify the validity of the proposed methods,they are applied to one-step and multi-step predictions of traffic flow in Beijing.Under the same conditions,they are compared with online adaptive ALD-KRLS method and other kernel learning methods.Experimental results show that the proposed KAF algorithms can improve the prediction accuracy,and its online learning ability meets the actual requirements of traffic flow and contributes to real-time online forecasting of traffic flow.

Expressway traffic flow prediction using chaos cloud particle swarm algorithm and PPPR model

Aiming at the real-time fluctuation and nonlinear characteristics of the expressway short-term traffic flow forecasting the parameter projection pursuit regression PPPR model is applied to forecast the expressway traffic flow where the orthogonal Hermite polynomial is used to fit the ridge functions and the least square method is employed to determine the polynomial weight coefficient c.In order to efficiently optimize the projection direction a and the number M of ridge functions of the PPPR model the chaos cloud particle swarm optimization CCPSO algorithm is applied to optimize the parameters. The CCPSO-PPPR hybrid optimization model for expressway short-term traffic flow forecasting is established in which the CCPSO algorithm is used to optimize the optimal projection direction a in the inner layer while the number M of ridge functions is optimized in the outer layer.Traffic volume weather factors and travel date of the previous several time intervals of the road section are taken as the input influencing factors. Example forecasting and model comparison results indicate that the proposed model can obtain a better forecasting effect and its absolute error is controlled within [-6,6] which can meet the application requirements of expressway traffic flow forecasting.

Traffic simulation and forecasting system in Beijing

Transport system is a time-varying, huge and complex system. In order to have the traffic management department make pre-appropriate traffic management measures to adjust the traffic management control program, and release travel information to travelers, to provide optimal path options to ensure that the transport system operates efficiently and safely, we have to monitor the changing of the state of road traffic and to accurately evaluate the state of the traffic, then to predict the future state of traffic. This paper represents the construction of the road traffic flow simulation including the logical structure and the physical structure, and introduces the system functions of forecasting system in Beijing.

时空图神经网络在交通流预测研究中的构建与应用综述

交通流量预测是城市交通管理和规划中的关键问题,而传统预测方法在面对数据稀疏性、非线性关系和复杂动态性等挑战时表现不佳。图神经网络是一种基于非欧结构数据的深度学习方法,近年来在各种复杂网络建模和预测任务中得到广泛应用。为了应用于交通流量预测领域,提出了时空图神经网络,其能够捕捉空间和时间相关性,相较之前的预测模型有显著进步。对近年来使用时空图神经网络进行交通流量预测的模型进行分析,概述和比较了多种邻接阵的构造方式,然后从空间相关性和时间相关性的角度列举了构建交通流预测模型的常用组件,并对不同的时空融合方式进行了分类和对比;在应用方面,根据时间尺度的不同将时空图神经网络模型分为长期预测、短期预测与兼顾长短期的预测三类,分析了各自的目标与要求,并列举比较了近年来较为突出的新模型。最后,讨论了现有研究的局限性,对相关模型的未来研究做出展望。

基于流计算和大数据平台的实时交通流预测

目前交通流预测实时性差,很难满足在线分析和预测任务的需求,基于此提出一种Flink流计算框架和大数据平台结合的实时交通流预测方法。基于流计算框架实时捕捉和预处理数据,包括采用Flink的transform算子对数据进行校验和处理,将处理后的数据sink到大数据的HDFS文件系统,交由下一步的大数据并行框架进行分析建模与训练,实现基于流计算和大数据平台的实时交通流预测。实验结果表明,Flink能够实时捕捉和预处理交通流数据,把数据准时无误送入分布式文件系统中,在此基础上借助大数据框架下的并行分析和建模优势,在实时性数据分析与预测方面取得了较好的效果。

面向动态交通流多步预测的时空图模型

为更好表征交通路网中节点之间的动态隐式关系,提出一种基于时空数据嵌入的动态图卷积交通流预测模型。基于路网中节点之间的共现关系,利用深度游走算法将时空数据映射到嵌入空间中学习节点的向量表示;引入时隙嵌入特征与二维空间嵌入特征共同构建三维嵌入邻接张量,用于捕获时空依赖关系;在图卷积网络中添加自适应更新机制,利用循环组件演化图卷积网络的参数,以捕获图序列的动态性。将所提模型应用于基于真实交通数据集的交通流预测,结果验证了其有效性和提取路网隐式关系的准确性。

基于图自编码器和GRU网络的分层交通流预测模型

准确的交通流预测信息不仅可以为交通管理人员提供交通决策的坚实基础,还可以减少交通拥堵情况。在交通流预测任务中,获得有效的交通流的时空特性是保证预测效果的前提。现有的方法大多是用未来时刻的数据进行监督学习,提取的特征具有局限性。针对现有预测模型无法充分挖掘交通流的时空特性的问题,提出了基于改进的图自编码器和门控循环单元的分层交通预测模型。首先使用图注意力自编码器以无监督的方式深度挖掘交通流的空间特性,然后使用门控循环单元进行时间特征提取。分层结构采用分开训练的方式进行时空依赖关系的学习,旨在获取路网天然存在的空间拓扑特征,使其可以兼容不同时间步下的交通流预测任务。大量实验证明,所提出的GAE-GRU模型在不同数据集下的交通预测任务中取得了优异的表现,MAE,RMSE和MAPE指标均优于基线模型。

高速公路施工控制区动态交通流预测的LSTM-BiGRU-Attention模型

为提前准确预知高速公路施工控制区交通流变化趋势,解决交通流时间序列中的长期依赖问题,文中建立了高速公路施工控制区动态交通流预测的LSTM-BiGRU-Attention模型。首先,将预处理后的动态交通流数据集按时间步长顺序输入到LSTM网络,对交通流信息建模和学习。然后,引入BiGRU和Attention机制以更好地捕捉上下文信息和提供更具针对性的权重分配。最后,将构建的LSTM-BiGRU-Attention模型与其他模型进行交通流预测对比,评估模型性能。实验以G35济广高速公路某施工控制区交通运行情况为案例进行研究,结果显示该模型的平均绝对误差MAE为1.91,均方根误差RMSE为2.83,决定系数R^(2)为0.79,平均绝对百分数误差MAPE为3.23。对比其他模型,LSTM-BiGRU-Attention模型的4个评估指标均有所下降,说明该模型可为高速公路施工控制区提供更加精准的预测。

基于自适应动态图卷积循环网络的交通流预测

针对实际交通状况中节点之间存在的动态变化关系,提出一种自适应动态图时空预测模型TAGGRU,基于编码器-解码器网络结构对交通数据动态时空特征融合建模.将节点嵌入与时间编码结合为时空编码,并以此构建动态邻接图,用以表示节点关系的时间演化.将交通流数据与动态邻接矩阵共同输入编码器,通过自适应门控循环单元进行特征提取.编码器和解码器之间添加交互注意力模块,将历史特征进行转换,以生成未来特征表示,通过特征维度变换得到最终输出.结果表明:该模型有较优的预测性能.

基于VMD-ISSA-LSTM的短时交通流预测研究

针对城市短时交通流随机波动性强、可靠性低、预测精度差等问题,将变分模态分解(VariationalMode Decomposition,VMD)和改进麻雀搜索算法(ImproveSparrowSearchAlgorithm,ISSA)与长短期记忆(LongShort-Term Memory, LSTM)神经网络相结合,建立一种短时交通流预测模型(VMD-ISSA-LSTM)。首先利用VMD对历史原始交通流数据进行分解;然后采用佳点集、正弦函数扰动和Tent混沌映射等策略对标准的SSA算法加以改进,增强ISSA算法的寻优能力;最后,将每个分量送入ISSA-LSTM中进行预测,同时将预测结果线性叠加,得到交通流量预测值。以上海市中山北路-曹杨路口2018年11月1日—30日的历史交通数据对模型进行验证。结果表明,与LSTM、VMD-LSTM、VMD-SSA-LSTM等传统预测模型相比,VMD-ISSA-LSTM模型的预测结果的平均绝对百分比误差为1.278 4%,能够更好地应用于短时交通流预测中。

基于完全自适应噪声集合经验模态分解的短时交通流组合预测

为了提高短时交通流预测的准确性,鉴于短时交通流非平稳、难预测的特征,提出了基于完全自适应噪声集合经验模态分解(CEEMDAN)短时交通流组合预测方法。利用CEEMDAN将原始短时交通流信号进行分解得多个复杂度、频率不同的时间序列分量,利用排列熵算法(PE算法)计算各分量的复杂度;然后根据复杂度和随机性的不同分为高频和低频,分别使用ATT-TCN-BIGRU模型和ARIMA模型对高频分量和低频分量进行预测,最后叠加高频和低频的每个分量预测结果作为最终短时交通流预测值。仿真分析结果表明:与ARIMA模型、TCN模型、BIGRU模型、ATT-TCN-BIGRU模型相比,此模型的平均绝对误差及平均绝对百分比误差为最小,预测精度更高。

基于时空位置关注图神经网络的交通流预测方法

针对现有交通流预测方法存在难以构建空间和时间依赖关系的问题,提出了新的利用时空位置注意力的图神经网络(ST-PAGNN)方法。首先,该图神经网络中包含有位置关注机制,由此能够更好地对城市道路网络中交通节点的空间依赖关系进行有效捕捉;然后,利用带有trend adaptive Transformer(Trendformer)的门控递归神经网络来捕捉交通流序列在时间维度上的局部和全局信息;最后,利用改进的网格搜索优化方法对模型的引入参数进行优化,并以较高的时间效率获得全局最优解。实验结果表明,在数据集PEMS-BAY中,预测步长分别为15 min,30 min,60 min时,ST-PAGNN的评价指标RMSE、MAE和MAPE分别为1.37,2.57,2.67%,1.55,3.64,3.37%,1.97,4.37,4.43%;在数据集METR-LA中,预测步长分别为15 min,30 min,60 min时,ST-PAGNN的评价指标RMSE、MAE和MAPE分别为2.73,5.16,7.13%,2.99,5.97,7.86%,3.53,7.16,9.96%。结论表明,ST-PAGNN在不同粒度下的评价指标中均高于现有模型,从而说明了ST-PAGNN在解决交通预测问题方面的有效性和优越性。

基于时空关系的高速公路交通流量预测

高速公路交通流量预测对于交通拥堵预警、分流诱导和智慧高速公路建设具有重要意义.交通流具有复杂的时空依赖性,各个交通节点之间的空间关系随时间动态变化,时空关系的融合也缺乏高效的手段,因此对交通流量进行准确的预测具有挑战性.对此,提出一种基于动态图卷积网络与时空特征提取模块的高速公路交通流量预测方法.首先,通过动态图调节模块,提取交通流量序列的空间关系,根据提取到的空间特征,计算不同路网节点的道路相似性,并调整交通路网图结构;其次,通过时空特征提取模块,利用更新后的空间结构,结合时序处理方法,对交通流量数据的时空依赖关系进行建模.为检验模型效果,在美国加州高速公路性能测量系统(Performance measurement system,PeMS)所制作的数据集PeMS03、PeMS04、PeMS08和福州京台线高速公路数据集中进行实验对比,平均绝对误差分别为15.6、19.7、16.8和5.21,结果表明,本文提出的方法在高速公路交通流量预测中具有较好的表现.

基于边缘计算的交叉口信号自动控制

交叉口信号控制需及时响应交通流量的变化,特别是在高峰时段,需要快速、准确地进行调整,以确保交通顺畅.为此,提出基于边缘计算的交叉口信号自动控制方法.该方法利用边缘计算预测交通交叉路口的车流量.并以最小化车辆延误时间及强化通行能力为目标,引入二流理论描述研究区域路网的服务水平和拥堵状态,建立交通路段的流量-密度模型,进而完成信号灯自动控制.实验结果表明:提出的方法能够准确预测交叉路口的车流量,并有效缩短车辆的排队长度,能够对交叉路口的信号灯实现有效控制.