Evaluation of Dynamic Tensions of Single Point Mooring System under Random Waves with Artificial Neural Network

Peng Li, Conglin Jin, Gang Ma, Jie Yang, Liping SunJournal of Marine Science and Engineering 2022 · DOI

TL;DR

This paper introduces a Long Short-Term Memory (LSTM) neural network model for real-time prediction of mooring line tensions on Floating Production Storage and Offloading (FPSO) systems. By training on simulated data, the LSTM model demonstrates high accuracy in predicting dynamic tensions under various environmental conditions, offering a faster alternative to traditional numerical simulations.

What Problem Does This Paper Address?

Ensuring the operational safety of floating offshore structures like FPSOs is critical, especially in the harsh and complex marine environments of regions like the South China Sea. A key component of this safety is the mooring system, which keeps the platform in position. Monitoring the tension in these mooring lines is vital to prevent potential failures. However, traditional methods for calculating these tensions, such as finite element analysis software (like AQWA), are computationally intensive and too slow for real-time monitoring. Furthermore, physical monitoring equipment can be prone to failure and expensive to maintain. This paper addresses the need for a fast, accurate, and reliable method to predict mooring line tensions in real-time.

Methods

The core of this research is the application of a Long Short-Term Memory (LSTM) artificial neural network. LSTMs are a type of recurrent neural network particularly well-suited for processing sequential data and capturing long-term dependencies, making them ideal for time-series prediction tasks like mooring tension.

The methodology involves:

  1. Hydrodynamic and Mooring Model Establishment: A specific FPSO and its single-point mooring system were modeled using AQWA software. This forms the basis for generating simulation data.
  2. Data Generation: Time-domain simulations were performed using AQWA under various combinations of wind, wave, and current conditions. These simulations generated the training, validation, and test datasets. The input to the model consists of the FPSO's 5-DOF (degrees of freedom) motions (surge, sway, heave, roll, pitch), and the output is the tension of the mooring lines.
  3. LSTM Model Training: The generated simulation data was used to train a multi-layer LSTM network. The paper details the process of determining optimal hyperparameters for the LSTM model, including learning rate, batch size, number of epochs, and time steps.
  4. Prediction and Validation: The trained LSTM model was then used to predict mooring line tensions for scenarios not encountered during training. These predictions were compared against AQWA simulation results to evaluate the model's accuracy and generalization capabilities.
  5. Normalization: A Min-Max scaling normalization technique was employed to standardize the input and output data for the LSTM model, ensuring better performance.
  6. Error Metrics: Standard metrics like Root Mean Square Error (RMSN), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) were used to quantify the prediction accuracy.

Datasets

The study utilized datasets generated from numerical simulations using AQWA software. These datasets represent the time-domain responses of a specific FPSO under various environmental load conditions (wind, wave, current).

  • Training Data: Generated from AQWA simulations for a specific working condition (Case 1, condition 10,003) and multiple working conditions (Case 2).
  • Validation Data: Also generated from AQWA simulations, used to tune hyperparameters and assess model performance during training.
  • Test Data: Generated from AQWA simulations for both single (Case 1) and multiple (Case 2) environmental conditions, including scenarios with different wind, wave, and current directions. These datasets were used to evaluate the model's predictive capabilities on unseen data.

The paper does not introduce new benchmark datasets but rather uses simulated data derived from a specific FPSO model.

Key Insights and Ideas

  • LSTM's Efficacy for Mooring Tension: The research strongly validates the suitability of LSTM networks for predicting dynamic mooring line tensions. Their ability to capture complex temporal dependencies in sequential data is a significant advantage over simpler models.
  • Real-time Prediction Potential: The primary benefit highlighted is the significant speedup offered by the LSTM model compared to traditional numerical simulations. This opens the door for real-time monitoring and early warning systems for mooring integrity.
  • Generalization Across Conditions: The LSTM model demonstrated good performance not only under the specific conditions it was trained on but also under various combinations of wind, wave, and current directions, indicating a degree of generalization.
  • Amplitude Prediction Challenges: While the LSTM model accurately captures the low-frequency trends and average values, it shows slightly lower accuracy in predicting peak tension amplitudes. This is attributed to the inherent difficulty in learning extreme values from limited training data.
  • Directional Influence: The study suggests that the model's performance might be slightly more robust when wind, wave, and current are aligned, though it still performs reasonably well with different directional combinations.

Key Results

  • The LSTM model achieved high accuracy in predicting mooring line tensions, with errors generally within acceptable engineering ranges. For instance, in Case 1, the maximum error for mooring line tension prediction was reported as 5.25%.
  • The LSTM model's predictions showed good agreement with AQWA simulation results, particularly in capturing the low-frequency variation trends of the mooring line tension.
  • The LSTM model consistently predicted maximum values slightly lower than AQWA and minimum values slightly higher than AQWA, while the average values were very close. This indicates a tendency to smooth out extreme peaks.
  • The study successfully demonstrated the model's ability to predict tensions under multiple environmental conditions with varying wind, wave, and current directions, showcasing its generalization capability.

Limitations and Future Work

  • Amplitude Prediction Accuracy: The paper acknowledges that the LSTM model's prediction of peak tension amplitudes is less accurate than its prediction of average values. This is a common challenge in time-series prediction, especially for extreme events.
  • Data Dependency: The model's performance is directly tied to the quality and representativeness of the training data generated by AQWA.
  • Real-world Data Validation: While validated against simulation data, the paper does not present validation against actual measured mooring line tensions from a real-world deployment.

Future Work suggested by the paper:

  • Further investigation into improving the prediction accuracy of extreme tension values.
  • Exploring the application of this LSTM model to predict mooring line failures or damage detection.
  • Potentially incorporating real-world sensor data for training and validation to enhance the model's robustness.
  • Investigating the impact of other environmental factors or more complex mooring system configurations.

Summary updated

Abstract

Real-time monitoring of the mooring safety of floating structures is of great significance to their production operations. A deep learning model is proposed here, based on the long short-term memory (LSTM) artificial neural network. Firstly, the numerical simulation is carried out with the single-point mooring system of a Floating Production Storage and Offloading (FPSO) as the training data of LSTM. Then the proposed LSTM is performed. Finally, taking the motion of FPSO which is not encountered by LSTM neural network model as input, we predict the mooring line tension with this model. Here, one FPSO in the South China Sea is taken as a research case, hydrodynamic and mooring models are established, and the network structure and hyper-parameters of the LSTM model are determined. The prediction results of the LSTM under different combinations of wind, wave, and current are compared with the calculation results of AQWA software. The model constructed here can well predict the mooring line tension of different combinations of wind, wave and current.
BibTeX
@article{li2022evaluation,
  author = {Peng Li and Conglin Jin and Gang Ma and Jie Yang and Liping Sun},
  title = {{Evaluation of Dynamic Tensions of Single Point Mooring System under Random Waves with Artificial Neural Network}},
  journal = {Journal of Marine Science and Engineering},
  year = {2022},
  doi = {10.3390/jmse10050666},
}
Figure