With the continuous and rapid developing of our national economy and the improvement of the national trade, the national logistic market is expanding by leaps and bounds. Our national economy is obviously promoted by logistic, which is considered as 'the third source of profit', But nowadays, many people ignore the demand of logistics, the plan they made is obviously not fit for the fact. This had made lots of waste .So it is a foundational premise for the sustained and healthy development of our national logistic demand to establish a scientific and applicable forecasting model for logistic demand and give the accurate estimate. This paper mainly study the problem how to predict the demand of logistics. Firstly, the article simply introduces the development of logistics demand in china and describes its definition. The writer thinks the logistic demand must be emphasized on planning a logistics plan. The research of demand of logistics is a basis of plam design or investment on logistics industry, it provide scientific evidences for them. There is much quantitative research on the logistics demand. The writer thinks the logistics demand has very complicated relations with the economy, trade, consumption and the structure of industry, etc. In the thesis, the author probes into forecasting index selection for logistics demand.The logistics demand in china increasing so rapidly that the traditional methods are hard to predict the demand correctly. How to establish a forecasting model for logistics demand is an important project of logistics study. Researchers, decision makers and employees are all realizing that it is very necessary to forecast logistics demand accurately, but to this day there is not a centralized normal form for which kind of mathematics model can be used.ANN is an artificial nonlinear dynamic system based on the recognition of cerebra neural network theory. ANN is a theoretic cerebra neural network mathematic model and an information processing system based on imitating cerebra neural net structure. Based on previous samples, it can carry on "learn by itself and model identification. And it is proved to be a good instrument for classification and forecasting in practice application just for it's model identification ability. ANN can discriminate the relativity between training samples precisely, so it is better than traditional statistical method on forecasting function. And that, while the training samples is few and there is random error, ANN is much better than ordinary statistical models.ANN forecasting model selection, establishing procedure and achieving method of ANN forecasting model for logistics demand, and structures forecasting theory for logistics demand. In the thesis, based on time sequence statistical data, applying ANN multi-step prediction and rolling prediction, freight quantities.logistics cost in GDP and the total value of logistics, forecasting model is established. The thesis forecasts logistics demand of China by improved three-layer BP network. In order to make it sure that the data is in same quantity rank, the author adopts normalized method to treat the data imported and exported in advance while training BP net. These algorithms can improve network's convergence speed.Another must mention is that ANN is trained through fast BP algorithm with variable learning rate that mixed with momentum factorAt last, The writer points out ANN model is a kind of more effectively forecasting model use in logistic demand, when the designed model is reasonable. |