Deep learning for predicting the residual concentration of sodium hypochlorite in the cooling water OF nuclear power plants.

• Deep neural network was developed to predict the residual concentration of sodium hypochlorite in the cooling water of nuclear power plants. • Method was applied to a real practical problem in Brazilian NPP. • The prediction is based on environmental and process parameters that influence the residual chlorine concentration. • Real samples, collected during months were used to train the predictive model. • The model achieved average mean squared error of 0.002 and correlation coefficient (R2-Score) of 0.934. Sea water is used as cooling water in several operating nuclear power plants, as well as in other industries such as petrochemicals. Biofouling is a common problem in systems that use sea water, causing corrosion in pipes and equipment, blockages, and loss of efficiency in heat exchangers. The use of sodium hypochlorite has proven to be effective in minimizing the damages caused by biofouling, provided that the residual chlorine concentration remains within a specific range, as low dosages will not protect the equipment and high dosages can cause environmental damage. With the aim of reducing maintenance and operation costs of equipment, and particularly ensuring the operability of systems related to the safety of nuclear power plants that use sea water, this study developed deep artificial neural networks (ANN) models capable of predicting the residual chlorine concentration in the main cooling and safety systems of a nuclear power plant. This prediction is based on environmental and process parameters that influence the residual chlorine concentration. Proposed model achieved average mean squared error of 0.002 and correlation coefficient (R2-Score) of 0.934, demonstrating to be a promising tool for predicting residual chlorine concentration. [ABSTRACT FROM AUTHOR]