Additive Model for Monthly Reservoir Inflow Forecast

Abstract

Reservoir inflow forecasting plays an essential role in reservoir operation and management. Considering the characteristics of monthly inflow (trend, seasonality, and randomness throughout the hydrologic year), an additive model is proposed to forecast monthly reservoir inflow. Because different features are represented by different frequency bands of the time series, historical time series of the monthly inflow are decomposed by ensemble empirical mode decomposition into several intrinsic mode functions and a residue. According to frequency signatures analyzed by Fourier spectral representation, all intrinsic mode functions and residue are grouped into three terms: trend term, periodic term, and stochastic term. To accommodate the different characteristics of the three terms, an autoregressive model, a least-squares support vector machine, and an adaptive neuro-fuzzy inference system model are adopted for the three subforecasts, respectively. The additive model is subsequently used to integrate the three subforecasts representing different characteristics to achieve the final forecasting results. The proposed method is applied to the Three Gorges Reservoir in China, using data from January 2000 to December 2012. For comparison, the three terms’ models and two peer models—back-propagation neural network and autoregressive integrated moving average—are adopted for monthly inflow forecasting. Among all six approaches, the present additive model exhibits the best forecasting performance of mean absolute percentage error, 11.36%, normalized root-mean-square error, 0.15, and correlation coefficient 0.97.