Study on the influence factors of short-time thermal response test based on artificial neural network

摘要

In order to free from the dependence on identification models in the parameter estimation process of soil thermal properties and shorten the time of thermal response tests (TRTs) possibly, this paper proposes to directly establish the mapping relationship between thermal response characteristic parameters and soil thermal properties by artificial neural network (ANN). A large number of virtual TRTs are conducted on the simulation platform verified by testing to obtain data for the ANN. For evaluating the identification accuracy and rapidity under different factors, the prediction results are then compared with the set values. The results show that the less the number of input characteristic parameters, the better the identification accuracy. The smallest error is 4% when the characteristic parameter is the inlet temperature of ground heat exchanger (GHE). In addition, the error is small when using the temperature data of unsteady section for identification, which proves that this method can greatly shorten the test time. Taking the shortest time needed to achieve the same identification accuracy as an indicator, it is found that the optimal test interval is 30 s. Appropriately reducing heating power or increasing fluid velocity can shorten the time.