Container Throughput Prediction Paradigm With Major Events

Port container throughput is a barometer of the world’s economic development and is the basic premise for government authorities to formulate macroeconomic and industrial development policies.In recent years,the frequency of major events such as natural disasters,economic environment,geopolitics and public health(epidemic)has an obvious increasing trend,mainly manifested as a huge impact on the whole social transportation operation,whose impact scale and duration will cause fluctuating and oscillating changes in port container throughput;however,the container throughput observation data contains a variety of complex superimposed components,and the traditional forecasting technology However,the container throughput observation data contains a variety of complex superimposed components,and the traditional forecasting techniques cannot effectively capture the impact of major events on the forecast results,so it is difficult to obtain the forecast results that can be used to guide practice.In order to remedy the above deficiencies,it is necessary to innovate the method for container throughput forecasting by taking the quantitative analysis of the impact of major events as the core,i.e.,constructing a forecasting model based on the analysis of major events,and then proposing a highly scalable port container forecasting method that is applicable to the universal scenario of major events.To achieve the above objectives,this paper proposes a decomposition integration forecasting method(EMD-Event Analysis-ArimaSVR,EEAS)based on the impact analysis of major events.EMD)technique to decompose the original observation data into a finite number of Intrinsic Mode Function(IMF),and the impact signal of significant events can be extracted using IMF;(2)then,the data characteristics of different IMFs are analyzed by means of smoothness test,modal fluctuation scale analysis,structural breakpoint test,etc.for event impact analysis judgment The impact of significant events on the variation of port throughput series and define its scale degree according to statistical methods;(3)then,according to the data characteristics and the results of significant event determination,the IMFs are put into the applicable forecasting model groups(ARIMA forecasting group and optimized SVR forecasting group)and combined into new finite IMF modes by modal reconstruction,and subsequently,these new modes are forecasted individually;(4)finally,the prediction results of each new IMF are integrated by simple addition(ADD)for aggregation to obtain the final prediction results.In order to verify the forecasting performance of the new method,this paper compares and analyzes the forecasting performance and accuracy of the new method(EEAS)and EMD-SVR,SVR,and ARIMA by using the container throughput forecasting of the world’s largest container throughput transportation port(Shanghai Port)as an example.The results show that:(1)the new model can effectively capture the impact of two important events,namely,the new crown epidemic and the "big congestion" in Shanghai port,which makes the prediction accuracy significantly higher than that of the comparative prediction model;(2)the results of the cross-validation prove that the event impact hybrid analysis model can define the prediction lag d of the SVR model,which makes the prediction performance of the SVR model’s prediction performance is effectively improved.Through the above study,it is worth pointing out that this method has high scalability,which is mainly reflected in two aspects: first,the method can automatically extract major event impact components from the initial observation data in a data-driven manner,which is theoretically capable of analyzing container volume forecasting under any kind of major events that have significant impact on the forecast object,such as new crown epidemic,earthquake,economic policies,etc.,and has very high Second,the method allows the selection of suitable specific forecasting models according to the data characteristics,and has a very high scalability of specific forecasting models.