A novel IBAS-ELM model for prediction of water levels in front of pumping stations

Abstract

It is necessary but difficult to accurately predict the water levels in front of pumping stations of an open-channel water transfer project because of the complex interactions among hydraulic structures. In this study, an improved beetle antennae search - extreme learning machine (IBAS-ELM) water level prediction model is proposed by combining the improved beetle antennae search (IBAS) algorithm with extreme learning machine (ELM). First, the water level data is divided into regulated and unregulated periods according to whether the pump unit is regulated or not. Second, the model inputs in the regulated period are improved by taking into account the regulation time of pumping units and the flow before and after regulation. Then, the beetle antennae search (BAS) algorithm is improved by increasing the search direction and introducing the optimal individual and levy flight. Finally, the IBAS-ELM model is established to predict water levels in front of Huibu and Dongsong pumping stations of the Jiaodong Water Transfer Project, China. The results show that: (1) the values of mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) of all models under group prediction are 15.60–24.65%, 10.18–23.86%, and 14.81–26.96% lower than those under direct prediction, respectively, and the values of Nash efficiency coefficient (NSE) are 0.33–2.14% higher; (2) the MAE, RMSE and MAPE values after improvement of model inputs are 27.03–38.85%, 24.64–43.12% and 28.44–40.58% lower than those without improvement of model inputs, respectively, and the NSE values are 4.20–8.84% higher; (3) the MAE, RMSE and MAPE values of the IBAS-ELM model are 0.66–28.50%, 1.07–30.18% and 0–27.10% lower than those of other models, respectively, and the NSE value is 0.09–3.05% higher. Therefore, the proposed model has higher prediction accuracy and can provide support for the operation of water transfer projects.