基于矩阵分解的船舶交通流预测方法研究  

Research on Ship Traffic Flow Prediction Method Based on Matrix Factorization

作　　者：高广旭 刘敬贤[1,2,3] 刘奕 李宗志[4] GAO Guangxu;LIU Jingxian;LIU Yi;LI Zongzhi(School of Navigation, Wuhan University of Technology, Wuhan 430063, China;Hubei Key Laboratory of Inland Shipping Technology, Wuhan University of Technology, Wuhan 430063, China;National Engineering Research Center for Water Transport Safety, Wuhan 430063, China;Illinois Institute of Technology, Chicago IL 60616, USA)

机构地区：[1]武汉理工大学航运学院,武汉430063 [2]武汉理工大学内河航运技术湖北省重点实验室,武汉430063 [3]国家水运安全工程技术研究中心,武汉430063 [4]伊利诺伊理工大学,芝加哥IL60616

出　　处：《武汉理工大学学报（交通科学与工程版）》2022年第1期171-176,共6页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：国家自然科学基金(51709219);湖北省科技创新专项(2019AHB053)。

摘　　要：针对船舶交通流具有非线性及复杂性等特点,同时在传统交通流预测方法中对于交通流时空特征考虑不足,而造成预测结果精度不高.提出一种基于低秩稀疏矩阵分解(low rank and sparse decomposition,LRSD)与贝叶斯概率矩阵分解(Bayesian probabilistic matrix factorization,BPMF)结合的船舶交通流预测方法,将一维船舶交通流时序数列转换为二维船舶交通流矩阵,通过LRSD对二维船舶交通流矩阵中平稳性数据和波动性数据分解为低秩矩阵和稀疏矩阵.对于分解后的低秩稀疏矩阵运用BPMF模型进行预测,将两个预测后的矩阵进行恢复,得到最终的预测结果.同时与GM,ARIMA,WNN,BPNN,LSTM,BPMF模型的预测结果进行对比.实验结果表明:所提出的基于LRSD与BPMF组合的预测结果均方根误差平均值为2.62,标准差平均值为0.034,预测精度及结果可信度高于各对比预测模型.In view of the nonlinear and complex characteristics of ship traffic flow,the traditional traffic flow prediction method has not considered the spatio-temporal characteristics of traffic flow sufficiently,which leads to low prediction accuracy.A ship traffic flow prediction method based on low rank sparse matrix decomposition(LRSD)and Bayesian probabilistic matrix factorization(BPMF)was proposed,which converts one-dimensional time series of ship traffic flow into two-dimensional ship traffic flow matrix.The stationarity data and volatility data in two-dimensional ship traffic flow matrix were decomposed into low-rank matrix and sparse matrix by LRSD.BPMF model was used to predict the decomposed low rank sparse matrix.The final prediction result was obtained by restoring the two predicted matrices.Meanwhile,the prediction results of GM,ARIMA,WNN,BPNN,LSTM and BPMF models were compared.The experimental results show that the average root mean square error and standard deviation of the proposed prediction results based on the combination of LRSD and BPMF are 2.62 and 0.034,respectively,and the prediction accuracy and result reliability are higher than those of other comparative prediction models.

关 键 词：交通安全 船舶交通流 预测 矩阵分解 组合模型 

分 类 号：U692[交通运输工程—港口、海岸及近海工程]