基于卷积神经网络及长短时记忆网络的短时船舶交通流量预测

    Short-Term Water Traffic Flow Prediction with Convolutional Neural Network and Long Short-Term Memory Network

    • 摘要: 由于船舶交通流具有非线性和复杂性的特征,传统交通流量预测方法的预测结果精度普遍不高,且需大量历史数据进行训练。在考虑船舶交通流数据时间特性的基础上增加了对数据空间特性的考虑,提出一种基于卷积神经网络(Convolutional Neural Networks, CNN)和长短时记忆网络(Long Short-Term Memory, LSTM)的短时船舶交通流量预测模型,并引入动态时间规整(Dynamic Time Warping, DTW)算法构造损失函数,提升模型的预测精度。结果表明:通过与灰色模型(Grey Model, GM)、差分整合移动平均自回归模型(Autoregressive Integrated Moving Average Model, ARIMA)、小波神经网络(Wavelet Neural Network, WNN)、反向传播神经网络(Back Propagation Neural Network, BPNN)和CNN-LSTM等模型对比,所提出的CNN-LSTM-DTW预测模型的预测结果相对误差最小,可信度高,预测精度优于对比模型。

       

      Abstract: The convolutional neural network and long short-term memory network are used together to improve accuracy of short-term water traffic flow prediction and reduce the network training time.Loss function is constructed based on dynamic time warping algorithm.Achieved prediction accuracy of the design is compared to that of other models, including GM(Grey Model), ARIM, WNNA, WNN, BPNN, CNN-LSTM and proved to be superior.

       

    /

    返回文章
    返回