一种基于BP神经网络算法和低通滤波的水资源评估预测数学模型建模方法

A Method Based on Back Propagation Arithmetic and Low-pass Filter to Build the Evaluation and Prediction Mathematical Model for Water Resources

基于BP神经网络算法建立水资源评估数学模型。首先获得大量相关的评估指标,通过主成分分析法(PCA)剔除不重要指标。对于指标赋权,先用基于蒙特卡罗的层次分析法(AHP-MCA)初步给指标赋权,由于此法具有一定主观性,因此进一步采用BP神经网络算法对所赋权重进行训练调节,过程中需要一个标准比对物,最终得到符合实际情况的权重因子。在建立预测数学模型前,针对可能的突变输入数据,采用低通滤波器将突变高频数据过滤,增加模型适用性,最后通过灰色模型GM(1,1)建立预测模型。

To build an evaluation mathematical model for water resources,original evaluation indicators were selected,unimportant indicators by Principal Component Analysis(PCA)were eliminated and ultimately independent and importantindicators were obtained,which actually carry all information of the original indicators.Then the Analytic Hierarchy Processbased on Monte Carlo Algorithm(AHP-MCA)was used to preliminarily give weights.As this method is subjective,it isnecessary to adjust the weights by Back Propagation Arithmetic(BP).Ultimately,reasonable and real weights were obtained.Before building the prediction model,Low-pass Filter was used to filer abnormal data,which will add adaptabilityof the model.Finally,the prediction model was established by Grey Model GM(1,1) .