基于人工神经网络的模型择优预测方法及应用

The forecasting method of selecting-best model based on artificial neural networks and its application

针对组合预测中各模型权重难以合理确定的问题,根据"择优取用"原则将组合预测问题转化为一种模式识别问题,并采用非线性映射能力很强的改进BP人工神经网络方法进行该问题的求解。实例表明,这种择优预测方法不仅有效避免了传统组合预测模型权重的繁琐计算,而且能集各模型所长,概念清晰,计算简便。该法作为变权重组合预测方法的一个特例,在灾害风险预测等中有较高的实用价值。

In order to deal with the problem of how to scientifically determine the weights of combined forecasting models（CFM）, a new method is presented in this paper. According to the principle of the best selection, the combined forecasting model is transformed into a problem of pattern recognition, it can be resolved by the method of improved BP artificial neural network （ANN） which owns the ability of non-linear mapping. The given example shows that the so called selecting-best forecasting model not only successfully avoids the complex progress of computing the weights of combined forecasting models, but also owns the properties of clear concept, easy operation and good characters of the forecasting models. As a special case of variable-weighting CFM, it has some values in application.