Modelling of Engine Cooling System with a New Modelling Approach Based on Dynamic Neural Network

Thermal management has always played a significant role in reducing emissions and improving the fuel efficiency of the internal combustion engines (ICEs). With a momentous influence on the thermal behavior of the engines, the cooling system has a considerable impact on ICE performance. In this scenario, a method based on artificial neural network (ANN) of the cooling system was proposed in this work. Specific modeling methods were adopted for the various operating conditions and flow circuits of the cooling system. To describe these varied dynamic characteristics, four ANN sub-models were established to simulate the system at different temperature stages. As a closed-loop system, the temperature of the cooling system can be regarded as a result of all the experienced operating points. Therefore, integral parameters describing the trajectory of the system were selected as the input of the ANNs. The training data was segmented into multiple segmentations and parallel training was utilized. With this training method, each segmented data can be regarded as a brand-new learning content since a new trajectory is generated due to the initialization of the segmented data. In this way, more training content was generated and the model was able to represent the system with different initial temperature. Finally, in order to simulate the switching of the cooling and heating processes in the high-temperature stage, a net switching logic considering the engine operation conditions was proposed, based on which the model obtained the ability to predict the supercooling and superheating.