基于数据预处理灰色神经网络组合和集成预测

当研究的系统扰动因素过大或系统行为在某个时川点发生突变，出现严重扰动系统的异常数据时，提出不应直接按原始数据建模预测，而应根椐实际情况适当地对数据预处理．提出了基于数据修正的改进型灰色神经网络组合和集成预测，并根据南昌火车站旅客发送量时间序列建立了多个模型，从模型预测效果对比中说明数据修正、改进型灰色模型和改进型灰色神经网络、灰色神经网络组合和集成确实能提高预测精度．另外，修正数据要把握一个度，不能修正全部数据，只能修正较异常的数据，要在数据的趋势性和预测的灵敏性间取得平衡。

Grey neural network and integrated forecasting based on preprocessed data

When a system disturbance is too great or a sudden change occurs, the resulting abnormal data can severely disturb the forecasting system. In this situation,runninga forecasting model before abnormal ities in the original data are identified produces misleading results. In this paper, an improved grey neural network forecasting model and integrated forecasting method are proposed on the basis of data modifica-tion. Several forecasting models were tested based on time sequences of passenger volume in Nanchang Railway Station. After comparing model predictions with real data, it became clear that prediction accuracy is considerably improved with revised data, or an improved grey model, or a combined grey neural network. But the data modification must be done properly. Not all data should be modified, it is only necessary to modify abnormal data in order to maintain balance between the data tendency and forecasting sensitivity.