Application Of Broad Learning System In Container Throughput Forecasting

In recent years,along with the promotion of the national top-level Cooperation Initiative of "one belt and one road",the marine shipping industry will carry the transportation of our main trade goods in the future,and will play an important role in boosting the social and economic development of our country.However,due to the uncertainties in the growth of cargo volume,the cargo volume of many ports has been oversaturated,and cargo ships have to enter standby status in the near-port waters.Therefore,effective forecasting of port warehousing logistics volume in advance can not only improve the efficiency of port operation and save cost,but also provide decision support for shipping time planning and route optimization.The time series model based on BLS(Broad Learning System)proposed in this paper is applied to the prediction of container throughput in ports.It solves the pain points in the field of engineering application through the hot research in the field of application management operations research.Firstly,the BLS is an efficient incremental machine learning method without deep structure.It is based on the idea of using mapping features as input of RVFLNN(random vector functional link neural network).The feature nodes and enhancement nodes of data complete the model training process in an effective and efficient way.At the same time,the concept of activation function is introduced in BLS algorithm,so that the linear problem can be solved better in the non-linear space.However,since BLS is a typical classification and recognition algorithm,this paper will improve the BLS and build a time series prediction model based on BLS.Then,a prediction model of container cargo throughput based on BLS time series algorithm proposed in this paper will be used to predict container cargo throughput of Shanghai Port.Finally,the prediction results of BLS model are compared with those of linear regression model,support vector regression model,autoregressive integral moving average model and generalized autoregressive conditional heteroscedasticity model.The experimental results show that the method is optimized on the basis of training data and achieves higher prediction accuracy.