基于TEL-BiLSTM-PSO-CKDE模型的短时交通量多步预测研究

Research on Shot-Term Traffic Flow Step-Prediction on TEL-BiLSTM-PSO-CKDE Model

摘要: 在智能交通系统(Intelligent Transportation System, ITS)快速发展的背景下，短时交通量预测技术作为交通调控与出行管理的重要支撑，其研究价值日益突出。当前研究大多集中于单步预测，忽视了复杂交通环境下对多步预测和不确定性建模的双重需求。为提升短时交通流多步预测的准确性与鲁棒性，文章提出一种基于数据二次分解和不确定性建模的混合模型——TEL-BiLSTM-PSO-CKDE。该模型融合时变滤波经验模态分解(TVF-EMD)与局部均值分解(LMD)形成数据二次分解架构(简称TEL)，用于处理原始交通数据的非线性与非平稳性；同时引入双向长短时记忆神经网络(BiLSTM)进行子序列预测，并结合粒子群优化(PSO)条件核密度估计(CKDE)对预测残差的概率密度函数进行建模，最终生成多步预测区间(Prediction Interval, PI)。实验结果表明：文章提出的混合模型在PICP、PINAW、CWC等指标上优于ARIMA、BP神经网络等传统方法，具备良好的预测稳定性与泛化能力，可为ITS系统提供有效支持。尽管模型结构较为复杂，未来研究可考虑合并部分分解模块或简化组合方式，以提升整体的可解释性与部署效率。

Abstract: In the context of the rapid development of Intelligent Transportation System (ITS), short-term traffic volume prediction technology is an important support for traffic regulation and travel management, and its research value is becoming increasingly prominent. Most of the current research focuses on single-step prediction, ignoring the dual needs of multi-step prediction and uncertainty modeling in complex traffic environments. In order to improve the accuracy and robustness of multi-step prediction of short-term traffic flow, this paper proposes a hybrid model based on data quadratic decomposition and uncertainty modeling, TEL-BiLSTM-PSO-CKDE. The model fuses time-varying filtering empirical mode decomposition (TVF-EMD) and local mean decomposition (LMD) to form a data quadratic decomposition architecture (TEL), which is used to deal with the nonlinearity and nonstationarity of the original traffic data. At the same time, the bidirectional long short-term memory neural network (BiLSTM) was introduced for subsequence prediction, and the probability density function of the prediction residuals was modeled by combining particle swarm optimization (PSO) conditional kernel density estimation (CKDE), and finally the multi-step prediction interval (PI) was generated. Experimental results show that the hybrid model proposed in this paper is superior to traditional methods such as ARIMA and BP neural network in terms of PICP, PINAW, CWC and other indicators, and has good prediction stability and generalization ability, which can provide effective support for ITS system.