Waste Heat Recovery of a Thermal Power Plant

Thermal power plants still provide the bulk of electricity that can be dispatched on the grid in India and the rest of the developing world, but a significant portion of the energy they use is dissipated as low-quality heat into the environment. Much of this rejected energy is lost from the plant with the boiler flue gas at temperatures appropriate to make use of them but not high enough for a conventional steam bottoming cycle. A representative 210 MW subcritical coal fired unit is considered to recover the low-grade heat of the flue gas in a subcritical Organic Rankine Cycle (ORC) along with a machine-learning framework. A first law and second law thermodynamic model was developed in Python, where the flue gas source was initially at a temperature of 160 °C and subsequently cooled to a lower limit set by an acid dew-point. Seven candidate working fluids were considered and R1233zd(E) and R245fa were found to be most promising on both the performance and environmental aspects. A parametric study quantified the effect of evaporator pressure, condensation temperature, superheat, inlet temperature and mass flow of flue-gas on the net power and thermal efficiency. Five regressors have been trained with a dataset of approximately 2050 feasible operating points, with the best accuracy (R² = 0.991, RMSE = 0.049 MW), given by a gradient-boosted-tree (XGBoost) surrogate model, showing close agreement with the underlying thermodynamics in terms of the ranking of features. The surrogate successfully identified an optimum value that recovers ~2.60 MW of net power, which is within 1% of the optimum recovered in the full model, a ~1.2 percentage-point improvement in plant efficiency and an estimated annual abatement of ~18,000 t CO₂ at only over two orders of magnitude less computing cost than direct optimisation.