Prediction of Suspended Sediment Concentration by ANN with Immune Genetic Algorithm

Abstract

This paper used the immune genetic algorithm (IGA) for optimizing artificial neural network (ANN) and constructed a new model ANN-IGA to predict suspended sediment concentration(SSC) based on the daily SSC and flow discharge data observed at the hydrological station near Puerto Rico, USA. Moreover, the performances of ANN-IGA were compared with those of ANN with particle swarm optimization (ANN-PSO), ANN, generalized regression neural network (GRNN), radial basis neural network (RBNN) and the traditional sediment rating curve (SRC). The root mean square error (RMSE), mean root square error (MRSE), and coefficient of determination (R2) were adopted as indicators for the evaluation of the prediction accuracy of all models. According to the different settings of the input and output variables, the predictions for four different scenarios were carried out. The comparative analysis results show that we can gain the best prediction results when the current day’s flow is designed as the input variable and the current day’s SSC designed as the output variable for the hydrological Quebrada Blanca station. The MRSE value of ANN-IGA, ANN-PSO, ANN, RBNN, GRNN and SRC is respectively 0.7689, 0.8494, 0.9684, 1.0481, 1.0995 and 1.6742. It is obvious that ANN-IGA and ANN-PSO boast superior performances to ANN, GRNN, RBNN, and SRC. Furthermore, ANN-IGA is slightly superior to ANN-PSO, and ANN is slightly superior to GRNN and RBNN.