Prediction for snow melting process of conductive ethylene propylene diene monomer composites based on machine learning approaches

Abstract

Accurate prediction for snow melting process of conductive ethylene propylene diene monomer (EPDM) composites is an important prerequisite for engineering applications. In this paper, the snow melting process of conductive EPDM composites on roads was determined and three machine learning approaches were compared. The snow melting process was tested several times in outdoor environment. Multiple linear regression (MLR), back propagation neural network (BPNN) and support vector regression (SVR) models were analyzed and developed. The three machine learning models were optimized by a cross validation approach. Finally, a suitable model was selected to predict the snow melting process. The results show that conductive EPDM composites can melt snow on roads at low voltages. In addition, the snow melting process is controlled by adjusting the input voltage. The rate of snow melting is affected by snow thickness, input voltage, and ambient temperature. The results of the MLR model are poor with a coefficient of determination (R2) less than 0.85. Moreover, the BPNN model with three neurons has better prediction results. The R2 of the BPNN model ranges from 0.93 to 0.95. The R2 of the SVR model with radial basis kernel function is greater than 0.95. Among the three machine learning models, the SVR model can accurately predict the snow melting process. This study not only provides a new idea for active snow melting but also provides a theoretical basis for the engineering application of conductive EPDM composites.