Equalization Optimizer-Based LSTM Application in Reservoir Identification

Abstract

In recent years, the use of long short-term memory (LSTM) has made significant contributions to various fields and the use of intelligent optimization algorithms combined with LSTM is also one of the best ways to improve model shortcomings and increase classification accuracy. Reservoir identification is a key and difficult point in the process of logging, so using LSTM to identify the reservoir is very important. To improve the logging reservoir identification accuracy of LSTM, an improved equalization optimizer algorithm (TAFEO) is proposed in this paper to optimize the number of neurons and various parameters of LSTM. The TAFEO algorithm mainly employs tent chaotic mapping to enhance the population diversity of the algorithm, convergence factor is introduced to better balance the local and global search, and then, a premature disturbance strategy is employed to overcome the shortcomings of local minima. The optimization performance of the TAFEO algorithm is tested with 16 benchmark test functions and Wilcoxon rank-sum test for optimization results. The improved algorithm is superior to many intelligent optimization algorithms in accuracy and convergence speed and has good robustness. The receiver operating characteristic (ROC) curve is used to evaluate the performance of the optimized LSTM model. Through the simulation and comparison of UCI datasets, the results show that the performance of the LSTM model based on TAFEO has been significantly improved, and the maximum area under the ROC curve value can get 99.43%. In practical logging applications, LSTM based on an equalization optimizer is effective in well-logging reservoir identification, the highest recognition accuracy can get 95.01%, and the accuracy of reservoir identification is better than other existing identification methods.