Investigation on the real-time control of the optimal discharge pressure in a transcritical CO2 system with data-handling and neural network method

In order to develop an acceptable real-time control approach in terms of accuracy and computation time in industrial and commercial applications, the based Back Propagation Neural Network (BPNN) approach was introduced into the discharge pressure optimization process of the transcritical CO₂ heat pump systems. The relevant characteristic variables concerning to the discharge pressure was minimized by the Group Method of Data Handling (GMDH) method, and the relevance of all the variables with the optimal rejection pressure were investigated one by one. Prediction error of different type neural network were compared with each other. Finally, the performance of neural network based transcritical CO₂ system was compared with that of conventional empirical correlations-based systems in terms of the optimal discharge pressure, which showed that the novel PSO-BP prediction model provides an innovative and appropriate idea for developers and manufacturers.