Time series prediction based on LSTM neural network for top tension response of umbilical cables

The umbilical cable is an essential component of offshore oil and gas extraction systems. The severe marine environment poses a high challenge to the safety of the umbilical cable structure during operation. The analysis of an umbilical cable requires complex and resource-demanding finite element time-domain simulations to obtain their nonlinear dynamic response. Therefore, in order to solve the problem of structural safety monitoring and real-time assessment of remaining life of umbilical cables under extreme sea states, there is a great need to predict the dynamic response of umbilical cables quickly and accurately during operation, for ease of making fast decisions for system operation and maintenance before the arrival of extreme sea states. Given the strong nonlinear function-approximation ability of the neural network, this study proposes an efficient method for the prediction of the time series of umbilical cable top tension response based on LSTM (long short-term memory) neural network. We use LSTM neural network and ARIMA (autoregressive integrated moving average) model in a real engineering case for time series prediction of the top tension response of the umbilical cable, and the results of the two models are analyzed and compared, and the efficiency and accuracy of the LSTM neural network model are verified. Furthermore, the hyperparameter, dataset and generalization ability of LSTM model are discussed. The results indicate that feasibility of the tension response prediction of umbilical cables under dynamic load in complex marine environments. •The application of neural network to dynamic response analysis of umbilical cable is studied.•A method for constructing a prediction model of the umbilical cable tension response based on LSTM neural network .•The effects of the network structure on the quality of the model were analyzed.•The applicability of the model at different positions of the umbilical cable and under different sea states is verified.•The effect of the data sampling frequency on the simulation and actual prediction is discussed.