Performance prediction of a cryogenic organic Rankine cycle based on back propagation neural network optimized by genetic algorithm

In this paper, a performance prediction model of the cryogenic ORC was presented based on the back propagation neural network optimized by the genetic algorithm (BPNN-GA). Firstly, an experimental setup was established to obtain the database for BPNN-GA model training and test. Then, the expander output power, working fluid mass flow rate, and the cold energy efficiency were selected as the BPNN-GA model outputs and the model structure was determined as 9-12-3. Finally, the accuracy of the BPNN-GA model was verified, and the parametric study was further conducted. The mean absolute relative errors (MARE) are 1.1876%, 0.9037%, and 2.6464%, the root mean square errors (RMSE) are 5.3789 W, 1.0260 kgh −1 , and 0.3151%, and the correlation coefficients (R) are 0.9974, 0.9977, and 0.9665 for the expansion work, the working fluid mass flow rate, and the cold energy efficiency, respectively. The BPNN-GA is proved as a promising methodology, which could provide direct guidance for the determination of relevant parameters in experimental analysis and control strategy optimization. • The performances of a recuperative cryogenic ORC were predicted by BPNN-GA. • 10270 sets of data for BPNN training and test were obtained with experiments. • The effect of hidden neuron number on the BPNN-GA model was discussed. • BPNN was determined as 9-12-3 with the correlation coefficients in 0.9665–0.9977. • Parametric analyses of the recuperative ORC were carried out.