基于混合神经网络-上下限估计法的河流流量区间预测

Interval Prediction of River Flow Based on Hybrid Neural Network and LUBE Method

为对混合模糊识别和连续性方程的神经网络模型(HNN)进行不确定性分析,采用上下限估计法(LUBE)直接得到神经网络输出层的两个节点值作为区间上下限,将综合评价指标CCWC作为目标函数,寻找最优的预测区间(PI),并以美国乔治亚州Yellow River上测站的日流量和小时流量为例,预测下游河道的区间流量。结果表明,HNN模型在90%、95%置信水平区间具有有效性,其可得到包含更多观测值且宽度更小的预测区间。同时,进一步验证了LUBE在区间预测中应用的合理性。

In order to address uncertainty analysis on a hybrid neural network(HNN)with fuzzy recognition and continuity equation,the lower upper bound estimation(LUBE)method was used to determine two-nodes outputs of neural network as the intervals of the lower and upper bounds.Coverage width-based criterion(CCWC)was taken as the objective function for searching optimal prediction interval(PI).The daily and hourly flows of Yellow River in Georgia of USA were studied as a case to predict the interval flows at the downstream stations.The results demonstrate the suitability of HNN-LUBE model in producing PI in both 90%and 95%confidence levels.It is capable of generating narrower intervals which also contain more observations.Furthermore,the efficiency of LUBE is validated in interval prediction.