基于二维矩阵分解的船舶交通流预测
Ship Traffic Flow Prediction with Bidimensional Matrix Mode Decomposition
-
摘要: 为克服船舶交通流的非线性和非平稳性特点造成的预测精度不高的问题,提出了一种融合二维经验模式分解(Bidimensional Empirical Mode Decomposition, BEMD)和时序正则化矩阵分解(Temporal regularized matrix factorization, TRMF)的船舶交通流预测方法。首先,将传统一维船舶交通流时序数据重整为二维交通流量时序矩阵(天×时段),再利用BEMD将二维交通流量数据分解为高频矩阵和低频矩阵,其中高频矩阵体现突变因素对交通流的影响,低频矩阵体现稳定因素对交通流的影响;接着,采用引入正则时序项的TRMF,分别对高频与低频矩阵进行预测,进而融合得到最终的交通流量预测结果;最后,对比分析BEMD-TRMF、GM(1,1)、ARIMA、BPNN、WNN、LSTM和TRMF预测模型,结果表明BEMD-TRMF模型的平均预测误差约为3%,优于对比模型,达到了较好的预测精度。Abstract: Bidimensional empirical mode decomposition and temporal regularized matrix factorization are used for ship traffic flow prediction. Normal traffic flow time series are reorganized as 2 D time matrix（day X Time slot） and the latter is decomposed into high frequency matrix and low frequency matrix. The high frequency matrix reflects the effects of fast changing factors and the low frequency matrix reflects the effects of steady factors. The two matrixes are processed separately with temporal regularized matrix factorization and fused to get the resultant prediction, The resultant prediction is compared with results obtained by using other models, such as BEMD-TRMF, GM（1,1）, ARIMA, BPNN, WNN, LSTM and TRMF, showing the advantage of the BEMD-TRMF model in accuracy（accuracy of 3% is achieved）.
下载: