Prediction of port cargo throughput using NeuralProphet-LSTM combination model

Conventional method has not been successful in predicting cargo throughput of ports due to the great fluctuation caused by complex external factors. In order to improve the accuracy of cargo throughput prediction for ports, the combination of NeuralProphet time series model and LSMT neural network is introduced. NeuralProphet model is trained to produce, according to input cargo throughput data, a preliminary prediction with a residual sequence. The LSMT model is used to correct the preliminary prediction and output final prediction. The combination model is verified through process the monthly throughput data from Shanghai Port and Xiamen Port. The advantage of the combination model in handling abnormal fluctuation over other configurations is demonstrated.