多层前馈网络在商品需求量预测中的应用

利用多层前馈网络进行商品需求量预测。采用一种基于相对敏感性指标的修剪算法来进行网络结构设计，该方法进行迭代回溯选择搜索，删除影响较小的节点，降低了网络的结构复杂度，并可以利用一种保持输入输出总体性能近似不变的算法进行调整剩余权重。结果说明此种方法具有较好的效果，不仅满足了误差目标的要求，而且提高了网络的推广能力，相对敏感性指标还可以反映时间序列数据之间的关系。

Commodity Demand Forecast Using Multilayer Feedforward Network

Multilayer feedforward network is used to do commodity demand forecast, and the network architecture optimized by a network-pruning algorithm based on relative sensitivity index. This technique performs an iterative backward selection search by successively removing nodes with little or no effect on the outputs, reducing the structural complexity of the network. The remaining weights can be adjusted in such a way that the overall input-output behavior learnt by the network is kept approximately unchanged. The results have shown that this technique is better. Not only the error goal is met with, but also the generalization capability is improved. The relative sensitivity index also can reflect the relation in the time series.