Self organizing map and least square support vector machine method for river flow modelling

Abstract

Successful river flow time series forecasting is a primary goal and an essential procedure required in the planning and water resources management. River flow data are important for engineers to design, build and operate various water projects and development. The monthly river flow data taken from Department of Irrigation and Drainage, Malaysia are used in this study. This study aims to develop a suitable model for short-term forecasting of monthly river flow in three catchment areas in Malaysia. The hybrid model based on a combination of two methods of Self Organizing Map (SOM) and Least Square Support Vector Machine (LSSVM) model referred as SOM-LSSVM model is introduced. The hybrid model using the “divide and conquer” approach where SOM algorithm is used to cluster the data into several disjointed clusters. Next, the LSSVM model is used to forecast the river flow for each cluster. This study also provides a method for determining the input structure that will be used by Artificial Neural Network (ANN), LSSVM and hybrid SOM-LSSVM models. There are three techniques used to determine the number of input structures. The first technique is based on the past trend river flow data, the second technique is based on the stepwise regression analysis and the third technique is the best Autoregressive Integrated Moving Average (ARIMA) model. The experiments present a comparison between a hybrid model and a single model of ARIMA, ANN, and LSSVM. The comparison to determine the best of the model is based on three types of statistical measures of Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Correlation Coefficient (r). The results have shown that the hybrid model shows better performance than other models for river flow forecasting. It also indicates that the proposed model can be predicted more accurately and provides a promising alternative technique in river flow forecasting.