Stochastic modeling of curling-induced loads on demountable ice rinks using ARIMA, ANN, and LSTM

• Field loading tests were conducted on a prefabricated ice rink to measure curling-induced loads employing wireless instrumented insoles. • A machine learning-based modeling framework was developed to simulate ice sports-induced loads during curling. • The ARIMA, ANN, and LSTM models were validated against experimental measurements, and compared with existing load models. Demountable ice rinks which can accommodate various activities through structural transformations, are becoming increasingly popular. Compared to traditional solid-foundation ice rinks, these rinks exhibit lower structural stiffness, making human-induced loads from ice sports critical for vibration serviceability design. While human-induced load models in civil buildings are well-established, research addressing ice-specific scenarios remains insufficient. Therefore, this study presents a machine learning-based modeling framework for ice sports-induced loads during curling activities on ice rinks. Plantar pressure data across multiple scenarios were captured via wireless instrumented insoles. Load characteristics were subsequently extracted through time-history and energy distribution analyses. Moreover, the measured load database containing five typical ice sports movements (walking, backward walking, jumping, ice-sweeping, and gliding) was established. Finally, ARIMA, ANN, and LSTM machine learning models were implemented, validated against experimental measurements, and compared with available load models. The study could provide reliable input loads for accurate dynamic analysis of demountable ice rinks.