Research On Forecast Methods Preference Of Port Throughput Development

As a key economic indicator of port, the development of throughput is very important for the decision-making of layout, investment, operation management and development strategy for a port.Some existing forecast techniques, including time series methods, causal methods and combined forecasting method, are reviewed in this paper. Some new forecasting models, such as grey method and artificial neural network method with genetic algorithm technique, are introduced in detail. Throughput forecasting is a complex system engineering, which should be analyzed systematically in terms of prescribed procedure. For port group with overlapped hinterland, the Analytia1 Hierarchy Process method (AHP) is used to decide the competitive relation between individual ports.The process of establishing forecasting models and their verification are discussed in the paper. For comparison, several models are selected, including the tri-exponential smoothing model and grey model in time series methods, as well as multivariate linear regression model and artificial neural network in causal methods. A combined forecasting model is introduced to eliminate the random errors in individual forecasting models.Generally, the development of throughput should keep a steady state during short or medium term due to strong inertia existing in the phenomena of society, politics and economy. However, the steady state is impacted quite probably by some unexpected events during a long future time. So, the time series methods are more suitable for the short-term and medium-term forecasting.Principal component analysis (AHP) with correlation analysis is used in causal models to reduce the number of influencing factors, which is useful to improve the forecasting accuracy. In this paper, the tri-exponential smoothing model in time series methods is adapted to obtain the future values of influencing factors instead of qualitative judge. As can be seen from the results, the artificial neural network with genetic algorithm technique is better than the multivariate linear regression model due to its better nonlinear approximation and global search ability. It is more rational to carry out the forecasting work by combining the quantitative analysis and the qualitative judge because the development of throughput is usually impacted by some unexpected events, which could hardly be caught by quantitative method and however maybe is judged easily by the qualitative analysis.The development of throughout of Tianjin port during short-term, medium-term and long-term is forecasted based on the models established in this paper. The results indicate that all of models are reliable in short-term forecasting and the throughput of Tianjin port will develop with annual compound growth rate of about 10%, which is probably reached as long as the macro-environment of Tianjin port keeps stable.