1. Title, Journal and Authors

Title: Artificial Neural Network ensemble modeling with conjunctive data clustering for water quality prediction in rivers

Journal: ScienceDirect

Authors: Sung Eun Kim, Il Won Seo

 

2. Summary

In this study, an ensemble modeling technique and clustering methods are employed to reduce the errors of Artificial Neural Network (ANN) model in one-day ahead water quality prediction. The errors are caused by the initial weight parameter problems and the imbalanced training data set. The optimal initial weight parameters were significantly influenced by the training algorithms and data structures. However, the commonly used method which is random initialization of the weight parameters sometimes does not reflect reality properly. 

Thus, conjunctive clustering methods with CDbw are used to improve model accuracy through input vectors adjustment in this study. Moreover, the multilayer feedforward neural network with one hidden layer is applied as network architecture of ANN ensemble modeling.

 

The ANN models for the training target data set show low interval RMSE in pH, DO, TP, and TN. But the Turb did not produce good results due to ill-training of the network caused by imbalanced training data set. The three-clustered ANN model is employed to solve this problem, the interval RMSE of Turb is also greatly reduced by comparison with the model without clustering.

 

Consequentially, ANN ensemble model and clustering methods useful for water quality prediction with imbalanced data. Particularly, clustering methods can improve the precision of initialization of the weight parameters.

 

3. Originality and Creativity

Ensemble ANN model, an ensemble of individual ANN models developed under each specific condition, shows good results in environmental prediction.

4. Application to research

 

 

5. Contact

Suki Park / Intern student

Environmental Systems Engineering Lab.

School of Earth Sciences and Environmental Engineering Gwangju Institute of Science and Technology

1 Oryong-dong Buk-gu Gwangju, 500-712, Korea

Phone : +82-10-9082-5579

E-mail : flskdpffps@naver.com