Power prediction of waste heat recovery system for a cement plant using back propagation neural network and its thermodynamic modeling

Abstract

This work targets on the prediction of power produced by a waste heat recovery (WHR) system for a cement plant using the back-propagation neural network (BPNN) and comparing its results with the actual cycle. A regression-based predictive model is developed to compare it with an actual WHR cycle that is analyzed thermodynamically. The predictive model is trained using the following parameters as input features: turbine inlet steam, pressure, temperature, and mass flow rate of both stages (i.e., high-pressure (HP) and low-pressure (LP) stages). The mean square error for the validation dataset is found to be 0.283 with 10 neurons in the hidden layer. Moreover, the significance of each parameter on the predicted power investigated shows that the HP stage parameters tend to have more impact on the power generated. The other half of the study comprises energy analysis of the power plant and a detailed parametric analysis to review the impact of each parameter on the power produced and thermal efficiency. The calculated thermal efficiency and generated power of the system are found to be 19.75% and 10.06 MW, respectively. Lastly, from the comparative study of the predictive model and the actual WHR simulated cycle, it is concluded that data science can be used as an alternative to thermodynamic modeling to avoid hefty calculations.