Utilizing Artificial Neural Network for Boiler Feedwater Quality Parameters Prediction at Perlis Power Plant

This study proposes a model to predict the feedwater quality parameters that can cause corrosion and scaling in the Perlis Power Plant boiler using an artificial neural network (ANN) for four years. The related operating parameters and boiler feedwater chemistry, notably pH, specific conductivity, silica content, ammonia content, and iron content, were input parameters in the ANN model to predict five different boiler feedwater quality parameters. The duration of the data selection was monthly from January 2019 to December 2022. This study utilized the Neural Net Fitting app in MATLAB R2022b to develop the proposed ANN model. After constructing the ANN model, a model performance evaluation was done to determine the effectiveness and accuracy of the trained ANN model. These include the coefficient of determination (R2) and the root mean square error (RMSE). The results showed that the ANN models have good performance, with R from 0.8135 - 0.9991 and R2 from 0.6618 - 0.9982 in training, validation, and testing datasets for the predictions of all five boiler feedwater quality parameters.

I. Introduction

Major issues on the water side of industrial boilers are corrosion and scaling. Therefore, it is critical to maintain the quality of boiler feedwater to avoid corrosion and scaling, which can lead to boiler tube failure. According to a study by [1], the cause of boiler tube failure at Indonesia's Suralaya power plant is the poor quality of boiler feed water which contains high values of sodium, conductivity, and silica content. In power plants, it is essential to maintain the feedwater quality supplied to boilers, as variations in boiler feedwater quality can significantly impact steam production, efficiency, and equipment longevity. This study uses an artificial neural network (ANN) to predict the feedwater quality parameters that can cause corrosion and scaling in the Perlis Power Plant boiler for four years. In the context of boiler feedwater quality prediction, an ANN model is trained using historical data on the feedwater quality. After constructing the ANN model, a model performance evaluation is done to determine the effectiveness and accuracy of the proposed ANN model. These include the coefficient of determination (R2) and the root mean square error (RMSE) [2]. Once the ANN has been trained, it can predict feedwater quality in realtime, which can help power plants optimize their steam production processes, improve efficiency, and protect equipment.