A Comprehensive Study On Developing Neural Network Models For Predicting The Coagulant Dosage And Treated Water Qualities For A Water Treatment Plant
Determination of the optimum coagulant dosage for water treatment is traditionally carried out using the jar test, which is a time consuming procedure incapable of responding to sudden changes in water qualities. Therefore, data driven modeling techniques such as neural networks are used for develop...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/46832/1/A%20Comprehensive%20Study%20On%20Developing%20Neural%20Network%20Models%20For%20Predicting%20The%20Coagulant%20Dosage%20And%20Treated%20Water%20Qualities%20For%20A%20Water%20Treatment%20Plant.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Determination of the optimum coagulant dosage for water treatment is traditionally carried out using the jar test, which is a time consuming procedure incapable of responding to sudden changes in water qualities. Therefore, data driven modeling techniques such as neural networks are used for developing predictive models for the coagulation process. In this work, three different neural network models, namely, the general regression neural network (GRNN), extreme learning machine single layer feed forward neural network (ELM-SLFN) and the extreme learning machine radial basis function neural network (ELM-RBF) were developed to predict the coagulant dosage, and their performances were compared with the commonly used multilayer perceptron neural network (MLP). It was shown that the ELM and the GRNN models consumed significantly lesser effort and time for training compared to the MLP. The ELM-RBF demonstrated the best tradeoff between prediction accuracy and computational requirement. Therefore, the ELM-RBF was used to develop models for predicting the coagulant dosage, treated water (TW) turbidity and residual aluminum with R values of 0.9752, 0.8239 and 0.9019 respectively. The input parameters required to develop each model was determined using a global exhaustive search algorithm as it was shown that the Pearson correlation coefficient and the principal component analysis were not suitable techniques for selecting input parameters for this study. Thus, inputs used for predicting the coagulant dosage were raw water (RW) turbidity, RW color and alum (t-1). The effectiveness of the coagulant dosage and the TW quality models were improved using an imputation model and a genetic algorithm. The imputation model was developed using K-means clustering with an imputation accuracy similar to a self-organizing map, to cope with failures in hardware sensors causing downtime in fully automated water treatment plants and ensure the continual use of the coagulant dosage model. The imputation model reconstructed missing values of RW turbidity and RW color with R values of 0.9075 and 0.8250 respectively. Subsequently, the reconstructed RW turbidity and RW color were used to predict the coagulant dosage with R values of 0.9742 and 0.9809 respectively, which are highly satisfactory. Meanwhile, the GA improved the R value of the TW turbidity model to 0.8294. The GA significantly improved the ability of the ELM-RBF to identify the required response of TW turbidity to the alum dosage. |
|---|