Optimal operation for organic Rankine cycle using a machine learning-based radial turbine with variable nozzles at off-design conditions

To maximize the off-design performance of an organic Rankine cycle using a radial turbine with variable nozzles, this study proposes a new integrated optimization method, including design and off-design stages. The one-dimensional design optimization is conducted to simultaneously determine the optimal design parameters and working fluid in a single step. Then, an off-design model of the variable nozzle turbine (VNT) is developed using three-dimensional computational fluid dynamics and particle swarm optimization-backpropagation neural network method to accurately predict its behavior. In addition, the off-design optimization is performed to obtain the optimal off-design performance and operating parameters under constant pressure and sliding pressure control strategy. The results show that R600 is the optimal working fluid among several candidate fluids. The thermal efficiency and net output power of the designed system are 10.91% and 358 kW, respectively. Under the sliding pressure control strategy, the turbine inlet pressure increases linearly while the nozzle outlet angle decreases with increasing heat source temperature, and the isentropic efficiency of the VNT basically maintained at designed value. The influence of heat source temperature on system off-design performance, turbine inlet pressure, and nozzle outlet angle is much greater than that of heat source flow rate. Compared with the constant pressure control strategy, sliding pressure operation can increase the net power of the system by up to 43.9% and reduce the levelized energy cost by 29.2%.