Universidad de Las Palmas de Gran Canaria
Published January 24, 2025 | Version v1
Model Open

ANN Model presented in "Surrogate model based on ANN for the evaluation of the fundamental frequency of offshore wind turbines supported on jackets"

Authors/Creators

  • 1. ROR icon Universidad de Las Palmas de Gran Canaria

Description

This repository contains the best of the models developed in the scientific article "Surrogate model based on ANN for the evaluation of the fundamental frequency of offshore wind turbines supported on jackets" by Román Quevedo-Reina, Guillermo M. Álamo, Luis A. Padrón, Juan J. Aznárez. (https://doi.org/10.1016/j.compstruc.2022.106917)

This is an enssemble model of 20 artificial neural networks with 22 neurons in the input layer, 4 hidden layers with 125 neurons per hidden layer, and 1 neuron in the output layer.

This model was trained using the dataset described in the paper above and available at: https://doi.org/10.5281/zenodo.14732047

Notes

The model is stored in the file "model.mat" and can be used by the included "class_model" class.
 
Model Contents
 
The "model.mat" file contains a Matlab variable named "model", which has the following properties:
  • dlnet: A collection of 20 artificial neural networks (ANNs) generated in the study.
  • name_input: Names of the input variables for the model (described in the scientific article).
  • name_output: Name of the output variable of the model (fundamental frequency).
  • mean_input: Mean values of each input variable, obtained from the training dataset. Used for normalization.
  • std_input: Standard deviation of each input variable, obtained from the training dataset. Used for normalization.
  • mean_output: Mean value of the output variable, obtained from the training dataset. Used for normalization.
  • std_output: Standard deviation of the output variable, obtained from the training dataset. Used for normalization.
 
How to Use the Model
 
Prerequisites
To use the model, you must have the Deep Learning Toolbox installed in Matlab.
 
Evaluation Method
The model can be evaluated using the "eval" method included in the "class_model". This method provides the following options:
  • output: Specifies how the individual ANN predictions should be aggregated. The options are:
    • "mean": Returns the mean of the predictions from the individual ANNs. (Default)
    • "std": Returns the standard deviation of the predictions from the individual ANNs.
    • "both": Returns both the mean and the standard deviation of the predictions from the individual ANNs, using the third dimension of the output matrix.
    • "all": Returns the predictions from all individual ANNs, using the third dimension of the output matrix.
  • IDnet: Specifies which ANNs to use for the evaluation. The options are:
    • "all": Uses all ANNs embedded in the model. (Default)
    • any number: Uses only the ANN at the specified position.
    • any vector: Uses only the ANNs at the positions specified in the vector.
 
Input Data Format
Prepare your input as a matrix where:
  • Each row represents a sample.
  • The matrix must have 22 columns, corresponding to the 22 input variables, arranged in the same order as specified in name_input.
 
Example
Below is an example of how to load the model and use the eval method:
 
% Load the model
load('model.mat');
 
% Prepare your input data (using case example)
load("case_example.mat");   % Load case example
inputData=example{:,:};     % Extract matrix of data
 
% Evaluate the model (default)
output = model.eval(inputData);
 
% Evaluate the model (specific options)
output = model.eval(inputData,'output','both','IDnet',1:10);

Files

Files (3.8 MB)

Name Size Download all
md5:66ca7ce20dfe44e6fef1e96f959f5fb7
2.0 kB Download
md5:e01a24076239625500d4ad06ae3746c8
2.6 kB Download
md5:e999a7bdcdc8af94fb9d49bc64e0e26c
3.7 MB Download

Additional details

Related works

Is described by
Journal article: 10.1016/j.compstruc.2022.106917 (DOI)

Funding

Agencia Estatal de Investigación
Design of offshore wind turbine supports including advanced models of dynamic soil-structure interaction and seismic action using neural nets (PID2020-120102RB-I00) MCIN/ AEI/ 10.13039/501100011033
Ministerio de Universidades
FPU research fellowship (FPU19/04170)