Runoff Prediction Based on Hybrid Clustering with WOA Intervals Mapping Model

Abstract

Accurate forecasting of daily runoff plays an important role in water resource management. This paper presents a feed-forward neural network interval-mapping-model-based clustering analysis technique and the and whale optimization algorithm (C-BPELM-WOAM) for the prediction intervals of the daily runoff series. The proposed model is composed of two parts. One part is the daily runoff point prediction. In this part, a combination of the error unequal-weight coefficient with the error back-propagation training and extreme learning machine algorithms was applied to construct a feed-forward neural network (BPELM), which can improve the performance of the prediction model. The second part is a clustering interval-mapping prediction model based on the whale optimization algorithm (WOA). In this part, k-means clustering was used to classify the daily runoff series data into several groups. Then the interval-mapping coefficients corresponding to each group of data were optimized by the WOA so that the prediction interval could be obtained. Finally, the daily runoff data for the Astor River basin were used to verify the efficiency of the C-BPELM-WOAM for daily runoff prediction intervals. The results showed that the C-BPELM-WOAM model obtained higher quality daily runoff prediction intervals.