Predicting energy consumption in desulfurization wastewater bypass evaporation systems using hybrid artificial neural networks

Abstract

Flue gas evaporation technology for desulfurization wastewater was a mainstream technique to achieve zero liquid discharge (ZLD). However, this technology required extracting a portion of the hot flue gas at the air preheater inlet, which reduced boiler efficiency and increased coal consumption. To accurately predict the energy consumption increase caused by flue gas extraction, this work proposed a hybrid predictive model that combines mechanistic modelling with an artificial neural network (ANN). Using operational data from a 660 MW power plant in Guangdong province as a sample, six parameters were selected as inputs to establish an energy consumption prediction model for the flue gas evaporation technology. The optimal structure of the model is 6 (Input layers) - 9 (Hidden layers) - 1 (Output layer), achieving an R² of 0.99478, with the relative prediction error fluctuating around 1 %, indicating overall good predictive performance. Furthermore, predictions were conducted for four typical operating conditions, with operational costs ranging from 18.21 to 24.5 CNY/m³ . The gas-to-liquid evaporation ratio was identified as a critical parameter affecting energy consumption. The recommended gas-to-liquid ratio range of 10,000–12,000 Nm³ /m³ could ensure complete wastewater evaporation while maintaining relatively low energy consumption. Additionally, this work reviewed two other ZLD demonstration projects, with operational costs for wastewater management around 20 CNY/m³ . The findings of this work supported the low-energy operation of flue gas evaporation technology for wastewater treatment and provided theoretical and technical guidance for ZLD processes.