基于LSTM-RBF的水路货运量预测

水路运输是交通和货运的重要组成部分，水路货运量的预测对各地经济发展有重要意义。近年来随着经济形势的变化和多式联运的快速发展，水路货运量的数据波动加大，精准预测的难度变得更大。因此提出一种基于长短期记忆网络（Long Short-Term Memory，LSTM）和径向基神经网络（Radial basis function，RBF）的组合预测模型 。在LSTM-RBF预测模型中，第一阶段通过LSTM对各指标进行精准预测，减少指标值误差对目标值预测带来的影响；第二阶段训练RBF神经网络并在未来指标值的基础上对目标值（水路货运量）进行预测。该模型既避免了时间序列预测仅考虑单一影响因素的缺陷，又能够把LSTM的长时记忆优势带入到RBF的回归预测中。实验结果表明，LSTM-RBF预测模型在均方根误差和拟合度方面，均优于其他对比模型，该预测模型对水路货运量的预测有着较高的准确度。

Prediction of waterway freight volume with LSTM-RBF

Water transportation is an important part of transportation and freight, and the forecasting of water freight volume is of great value to the economic development. In recent years, with the change of the economic situation and the rapid development of multimodal transport, the data fluctuation of water transport freight volume has increased, and the difficulty of accurate prediction has become greater. Therefore, a combined forecasting model based on long short-term memory (LSTM) and radial basis function (RBF) was proposed, LSTM is used to accurately forecast each index to reduce the impact of index value error on target value prediction; and then the RBF neural network is trained to forecast the target value (waterway freight volume) which based on the future index value. The LSTM-RBF model not only avoids the defect of single influencing factor in time series forecasting, but also can bring the advantage of LSTM''s long-term memory into RBF''s regression forecasting. The experimental results show that the LSTM-RBF model is superior to the other comparison models in terms of root-mean-square error and fitting degree, and this model has high accuracy in the forecasting of waterway freight volume.