文献类型：期刊文献

中文题名：基于模糊分析的LSTM交通流量预测

英文题名：Traffic flow prediction of LSTM neural network based on fuzzy analysis

作者：赵刚[1];王梦灵[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2021

卷号：42

期号：4

起止页码：1103

中文期刊名：计算机工程与设计

外文期刊名：Computer Engineering and Design

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：国家自然科学基金项目(61673177);上海市“科技创新计划”人工智能专项基金项目(19DZ1209003);上海市经济和信息化委员会人工智能创新发展专项资金计划基金项目(2019-RGZN-01015)。

语种：中文

中文关键词：交通流预测;LSTM神经网络;历史误差分析;模糊聚类;相似度分析

外文关键词：traffic flow prediction;LSTM neural network;historical error analysis;fuzzy clustering;similarity analysis

摘要：为提高交通流预测模型的准确性及泛化性,提出一种基于模糊分析的LSTM交通流预测方法实现对交通状态的预估分析。对历史数据采用LSTM神经网络进行训练,获取神经网络权值参数,针对交通流时序数据存在周期性,提出基于模糊聚类分析的策略对LSTM模型的历史训练误差进行聚类。根据当前交通流数据与历史数据的相似度预估LSTM预测模型的在线误差。综合LSTM神经网络预测输出以及基于相似度分析的在线误差预测输出预估交通流状态,给出相应的算法步骤。仿真实验验证了提出方法的有效性,其比单一预测预测模型效果更好。
To improve the accuracy and generalization of the traffic flow prediction model,a fuzzy analysis based long-short term memory(LSTM)neural network traffic flow prediction method was proposed to realize the prediction and analysis of traffic status.The historical data were trained through the LSTM neural network to obtain the weight parameters of the neural network.Considering the strong periodicity of the traffic flow time series data,a strategy based on the GK clustering analysis was proposed to cluster the historical training errors of the LSTM model.The online errors of the LSTM prediction model were estimated according to the similarity analysis of the current traffic flow data and the historical data.The prediction results were obtained by combining the prediction output of LSTM neural network with the estimate of online error,and the corresponding algorithm steps were given.The simulation results show that the proposed method is more effective than the single prediction model.