The effects of data preprocessing in the determination of coagulant dosing rate

Abstract

The artificial neural network (ANN) is known as an excellent estimator of nonlinear relationships between accumulated input and output numerical data. Using this nature of the ANN, the optimal coagulant dosing rate can be predicted from the operating data with accuracy and in time. But, the accumulated operating data used in ANN training may have some corrupt and noisy data records. So, to enhance the reliability of the trained ANN, a data preprocessing method is necessary for preparing the train and test data set. In this study, a data preprocessing method was devised, four data sets were prepared using the proposed data preprocessing method and the prediction capabilities of the ANN by each data set were compared in terms of a root-mean-square normalized error (RMSE). The purpose of the data preprocessing method is to remove the outliers of the confidence interval (CI) of the predicted value of the trained ANN from the accumulated operating data set. The data set prepared by data preprocessing shows enhancement of the learning rate and the terminal error. That is, the decrease in the confidence interval of the predicted value leads to an increase in the number of outliers, which results in a rapid learning rate and small terminal error. The ANN trained by the preprocessed data set also improves the prediction capability for the test data set. These results mean that the proper data preprocessing method can facilitate the ANN in formulating the latent structure and in removing real noises and measurement errors within the training data set.