58 / 2021-10-14 21:20:01

LSTM neural network models for the analysis of flexible cables and risers

Flexible risers; LSTM neural network; Nonlinear response prediction; Top tension

Flexible risers and cables are important components of offshore oil and gas production platforms. Usually complex Finite-Element time-domain simulation tools are used to obtain its serious nonlinear dynamic response. Such a method usually takes more time to calculate. It is considered that the neural network can accurately predict and quickly respond to the nonlinear behavior. In this paper, according to the Long Short-Term Memory (LSTM) neural network model, the past values of floating body displacement and expected response is taken as input information. A LSTM neural network model for predicting the response of dangerous points at the top of flexible pipe cable based on floating body motion is established. Through an example, the effects of the neuron number and layer number of neural network on the accuracy of prediction are studied. The effects of delay time and time distribution of training samples on the prediction results are discussed. The results indicate that the trained LSTM model can accurately predict the response of flexible risers and cables.