Application of a Hybrid Model Based on Secondary Decomposition and ELM Neural Network in Water Level Prediction

Abstract

Accurate water level forecasting is essential for agricultural water resources management, hydropower generation, flood control, drought relief, and watershed planning. A combined model (ICEEMDAN-VMD-WOA-ELM) is proposed based on improved adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN), variational mode decomposition (VMD), extreme learning machine (ELM) and whale optimization algorithm (WOA). First, the historical data with high similarity to the forecast date are extracted using a hierarchical clustering method, and the extracted data are analyzed by box plot. Then the analyzed river level data sets from different Heihe River stations are preprocessed with ICEEMDAN, and the obtained high-frequency subsequence intrinsic mode functions 1 (IMF1) is decomposed secondarily by VMD. Afterwards, the ELM parameters are optimized during the prediction of each subsequence using the global optimization capability of WOA, and the predicted values are accumulated to generate the ultimate water level prediction results. Experiments using five data sets and seven comparative models show that (1) the prediction accuracy of the single model is much lower than that of the combined models; (2) the secondary decomposition of ICEEMDAN-VMD has smaller prediction errors than the single decomposition of ICEEMDAN; and (3) the RMSE values of the ICEEMDAN-VMD-WOA-ELM model in the five data sets are 0.04365, 0.19644, 0.16856, 0.06412, and 0.37150, respectively, with much higher prediction accuracy than the ELM single model, which verifies the important research value of the proposed model in water level prediction.