Dynamic Modelling Of Automotive Air-conditioning Systems With A Data-driven Method

This paper discusses a new method for transient modeling of automotive air conditioning systems,that is,a black box modeling method with big data as support.Compared with the traditional white box model,which relies on the experience of the engineer,all the results of this method have data endorsement,so the reliability of the model is improved and the modeling cost is greatly reduced.In this paper,a set of data acquisition,cleanup,analysis and processing procedures are designed and prepared for the summer road test data of the automotive air conditioning system.The knowledge discovery and sensitivity analysis module based on artificial intelligence is integrated in the process.A time series clustering analysis module,for the model selection,establishment and optimization,as well as data filtering and slicing,is also developed.Sensitivity analysis module points out the dependence of variables in road test data,and the sensitivity of each state variable to external variables.These results will help people gain a deeper understanding of the various aspects of the system.The clustering analysis provides the basis for the selection of time slices.The most prominent temporal features are extracted from 1030 time slices for transient modeling,which reduces the computational complexity of model calibration by 85.8%.These results also provide an improved direction for the subsequent road test.Based on the result of the data analysis process and the analysis of the crew compartment model and HVAC system model,four black-box models were implemented and tested,and their modeling accuracy was analyzed with road test data.The four models are the linear state space model,the linear ARMAX model,the nonlinear NARX wavelet network and the nonlinear NARX neural network.The modeling results can meet the requirements of the design and optimization of the controller.