An Improved BP Neural Network Method For Railway Freight Volume Forecasting

The railway freight volume can reflect the traffic demands of the various national economy departments; it is not only the important basis of the constructions and operations but also the precondition for coordinating development of the internal departments of the railway. Under the condition of market economy, it seems to be particularly urgent and important to predict the changeable railway transportation demand promptly, sensitively, accurately and scientifically. But the railway freight volume is affected by a number of internal and external factors, and the relation to the influencing factors is complex and nonlinear. However the most traditional forecasting methods which based on the time series or the causality, can't reflect the data structure and the complexity of the problem, and also can't take full use of the information. So it's important and necessary to seek a series of scientific forecasting method to forecast the railway freight volume.First, the thesis puts forward the significance of the freight volume forecasting, analyzes the current studies and the problems on freight volume forecasting, and describes the main content of this thesis. Then it introduces the existing methods for forecasting the railway freight volume from qualitative and quantitative view. Empirical analysis is made by the exponential smoothing, linear regression, gray forecast model and the combine-forecast by the use of the real data of railway freight volumes in the last 17 years. From the qualitative and quantitative analysis of the national economy, industrial structure, production of bulk cargo, transportation structure, national policies and other factors, the thesis selects ten factors as the key factors for forecasting the railway freight volume and make the first phase preparations for the neural network algorithm. Based on the theories of the Artificial Neural Network and Genetic Algorithm, the thesis designs an advanced BP neural network——Genetic neural algorithm, and the algorithm is realized by programming in the MATLAB. Finally, the thesis compares the Genetic neural algorithm with the original neural network algorithm and with the other linear forecasting algorithms, and analyzes the advantages and disadvantages of these algorithms.The thesis research shows that:the railway freight volume is closely related to the factors such as GDP, proportion of the second industry, output of coal, output of oil, output of steel, output of cereals, infrastructure investment, throughput goods of port, railway market share, road market share and so on; and the Genetic neural algorithm proposed in this thesis is better than the original neural network algorithm in the convergence rate and the solution quality, and it's feasibility is also approved.