Model Order Reduction of Vehicle Air-Conditioning Loop with Neural Networks

To meet the energy demand, passenger’s comfort and sustainability in the air conditioning system of vehicles, the Neural Network plays an intelligent role. In this study, the vehicle air conditioning loop was replaced with the reduced control model using a neural network including Proportional-Integral-Derivative temperature controller. The detailed model of vehicle air conditioning system with Proportional-Integral-Derivative controller and the reduced model with neural network were modelled and simulated using AMESim simulation tool. The five different driving cycles (NEDC, FTP75, US06, EPA-HWFET and SC03) were used to simulate the results of both detailed model and reduced model with neural network. The reduced control model with neural network was trained and validated through more history of data sets. Finally, the output simulation results such as vehicle cabin temperature, relative humidity at evaporator outlet, compressor shaft mechanical power, fidelity, CPU run time and compressor command were compared between detailed model and reduced control model with neural network among each driving cycle. The simulation results showed the greater reduction of CPU run time of a reduced control model with neural network than a detailed model of vehicle air-conditioning system. The same fidelity levels and speed up gains obtained in cabin temperature, relative humidity at evaporator outlet, compressor shaft mechanical power and compressor command. The drastically reduced CPU run time of a reduced model with neural network, results in lower computational complexity and its adaptability at dynamic operating situations make it suitable for real time embedded controlled air-conditioning control in vehicles. The validation of optimized parameters helps the manufacturers significantly break down the number of expensive and time consuming physical sub-models, resulting in adequate cost savings and a more sustainable development process.