产品需求预测的演化神经网络算法MLPES

Forecast of Manufacturer's Product Demand Based on MLPES Algorithm

在需求拉动型的供应链中,需求成为供应链的起点和动力源泉。由于制造商在供应链中的特殊地位,制造商成为供应链由需求驱动变为预测驱动的断耦点,以制造商为核心进行准确的需求预测可以在一定程度上减少需求不确定性的影响。本文在多层感知器的框架上,提出了基于演化策略的神经网络预测方法MLPES,改进了在多层感知器中普遍采用的BP算法,并设计了学习算法的流程,通过反复测试确定了模型的参数,最后对预测结果进行了分析。

Product demand is the starting point and motivation in the demand-pulled supply chains. Because of its special position in the supply chain, the manufacturer becomes the decoupling point in the change of supply chain from demand driven to forecast driven. Accurate manufacturer-centric demand forecast can to some extent reduce the affect of the uncertainty of demand. Furthermore, on the basis of the multi layer perceptron model, a neural network forecasting algorithm (Multi Layer Perceptron based on Evolution Strategy, MLPES) combining an evolution strategy is proposed which improves the commonly used ones, back-propagation (BP) algorithms. The process of the learning algorithm and the model parameters obtained by several times of calculating are also presented. Finally, the forecasting result is analyzed and compared with that of a BP algorithm.