ARIMA Based Forecasting of stream flows of Three Georges Dam for efficient Water Resource Planning and Management

Abstract

Stream-flow forecasting is one of the major aspects in improving the efficiency of water resource planning and management for the water reservoirs in various geographical regions. Different forecasting techniques have been implemented for forecasting the inflow rate in the past with Support Vector Machine (SVM) being very popular and accurate among them. Similarly, the outflow forecasting is essential to estimate the usage and also encompasses different models for forecasting. Muskingum model is found to be prevalent in forecasting outflow but it has the limitation of inefficiency in taking care nonlinearity. In this paper, both inflow and outflow rate prediction is done by Auto-Regressive Integral Moving Average (ARIMA) model as prediction of stream flow was never modelled as ARIMA in the literature earlier. The performance evaluation parameter considered is the root mean square error (RMSE) for comparison with the existing SVM models. It is found that in case of inflow rate, RMSE obtained by ARIMA model shows a decrease of 36% in the best case and increase of 4.3% in the worst case when compared with the SVM models. Likewise, the outflow rate RMSE when compared with Muskingum model gives better results taking non-linearity into consideration. The results of this study will help in not only planning efficient water resource management but also prediction of flood frequency in future.