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

In oil explorationand development, many reservoir parameters arevery essential for reservoir description, especially porosity. Theporosity obtained by indoor experiments is reliable, but human andmaterial resources will be greatly invested. Experts have introducedmachine learning into the field of porosity prediction but with theshortcomings of traditional machine learning models, such as hyperparameterabuse and poor network structure. In this paper, a meta-heuristicalgorithm (Gray Wolf Optimization algorithm) is introduced to optimizethe ESN (echo state neural) network for logging porosity prediction.Tent mapping, a nonlinear control parameter strategy, and PSO (particleswarm optimization) thought are introduced to optimize the Gray WolfOptimization algorithm to improve the global search accuracy and avoidlocal optimal solutions. The database is constructed by using loggingdata and porosity values measured in the laboratory. Five loggingcurves are used as model input parameters, and porosity is used asthe model output parameter. At the same time, three other predictionmodels (BP neural network, least squares support vector machine, andlinear regression) are introduced to compare with the optimized models.The research results show that the improved Gray Wolf Optimizationalgorithm has more advantages than the ordinary Gray Wolf Optimizationalgorithm in terms of super parameter adjustment. The IGWO-ESN neuralnetwork is better than all machine learning models mentioned in thispaper (GWO-ESN, ESN, BP neural network, least squares support vectormachine, and linear regression) in terms of porosity prediction accuracy.

Automated summary

In oil exploration and development, porosity is a crucial parameter for reservoir description. Traditional methods for obtaining porosity through indoor experiments require significant resources. This study introduces the Gray Wolf Optimization algorithm to optimize porosity prediction using the ESN network. The improved algorithm shows superior performance in parameter adjustment, and the IGWO-ESN neural network outperforms other machine learning models in terms of porosity prediction accuracy.