Echo State Neural Network Based on an Improved Gray Wolf Algorithm Predicts Porosity through Logging Data

Area of Science:

• Petroleum Engineering
• Machine Learning
• Geophysics

Background:

• Accurate reservoir porosity prediction is crucial for oil exploration and development.
• Traditional porosity measurement methods are resource-intensive.
• Existing machine learning models for porosity prediction suffer from hyperparameter issues and suboptimal network architectures.

Purpose of the Study:

• To develop a more accurate and efficient method for predicting reservoir porosity using machine learning.
• To address the limitations of traditional machine learning models in porosity prediction.
• To optimize the Echo State Neural Network (ESN) using an enhanced meta-heuristic algorithm.

Main Methods:

• An improved Gray Wolf Optimization (IGWO) algorithm was developed by incorporating tent mapping and particle swarm optimization (PSO) principles.
• The IGWO algorithm was used to optimize the hyperparameters and network structure of the ESN model.
• A database was created using laboratory-measured porosity values and corresponding logging data (five logging curves as input).
• The performance of the optimized IGWO-ESN model was compared against standard ESN, BP neural network, least squares support vector machine, and linear regression models.

Main Results:

• The IGWO algorithm demonstrated superior performance in hyperparameter tuning compared to the standard Gray Wolf Optimization algorithm.
• The IGWO-ESN model achieved higher porosity prediction accuracy than all other evaluated machine learning models.
• The study confirmed the effectiveness of the enhanced meta-heuristic approach in improving ESN performance for logging data analysis.

Conclusions:

• The developed IGWO-ESN model offers a significant advancement in predicting reservoir porosity from logging data.
• This approach provides a more accurate and potentially cost-effective alternative to traditional methods.
• The findings highlight the potential of advanced meta-heuristic optimization techniques in enhancing machine learning applications within the oil and gas industry.