Ensemble approach for mid-long term runoff forecasting using hybrid algorithms

Abstract

Factor selection and model construction play an important role in mid-long term runoff forecasting. Due to the indeterminacy between the data and mid-long term runoff, identifying key factors for mid-long term runoff forecasting is challenging. Another problem for mid-long term runoff forecasting is the low accuracy, which limits practical application. Aiming to solve these problems, an ensemble approach is proposed in this paper. First, we propose a novel method for constructing a comprehensive runoff index, and apply the partial mutual information approach to calculate the correlation between multiple factors and the comprehensive runoff index. Through this calculation, the key factors for the mid-long term runoff forecasting can be selected. Second, we implement mid-long term forecasting by combining improved particle swarm optimization (IPSO) and extreme learning machine (ELM) algorithms, which can improve the accuracy of runoff forecasting. The novelty of the proposed method lies in combining the construction of comprehensive runoff index, the key factor selection and the forecasting model based on IPSO-ELM for mid-long term runoff. Experimental results demonstrate that the proposed forecasting model significantly outperforms the current state-of-the-art of the extreme learning machine algorithms and other classical data-driven models for runoff forecasting in the Yalong River basin. Moreover, the performance for datasets based on different hydrological impact factors in the conducted experiments proves the robustness of the proposed method.