Monthly runoff time series interval prediction based on WOA-VMD-LSTM using non-parametric kernel density estimation

Logical development and effective use of water resources depend heavily on the practicability of runoff forecast. A monthly runoff interval prediction approach based on WOA-VMD-LSTM and non-parametric kernel density estimation is proposed in an effort to address the issue that conventional point prediction method cannot properly convey the uncertainty of prediction findings. Using variational modal decomposition (VMD) optimized by whale optimization algorithm (WOA), monthly runoff series is first divided into a number of relatively stable subsequences. Each subsequence is then predicted using long-term and short-term memory neural network (LSTM), and final point prediction results are obtained by superposition. Non-parametric kernel density estimation is then used to forecast monthly runoff interval and is compared with LSTM, EMD-LSTM, and VMD-LSTM models based on findings of point prediction. Findings indicate that this model's prediction accuracy is noticeably higher than that of other models utilized in this paper. A realistic fluctuation range for runoff interval prediction is also provided by non-parametric kernel density estimation, which can be a useful reference for decision-makers in water resources management.

Automated summary

In this study, a monthly runoff interval prediction method based on WOA-VMD-LSTM and non-parametric kernel density estimation is proposed to address the issue of conveying prediction uncertainty. The approach involves dividing the monthly runoff series into stable subsequences using VMD optimized by WOA, predicting each subsequence using LSTM, and obtaining final point predictions by superposition. Non-parametric kernel density estimation is then used to forecast the runoff interval and is compared with other models. Results show that this model has higher prediction accuracy and provides a useful reference for decision-makers in water resources management.