基于PSO算法优化GRU神经网络的孔隙度预测

Optimization of Porosity Prediction by GRU Neural Network Based on PSO Algorithm

孔隙度参数是表征岩石存储石油能力的重要参数之一,也是储层评价的重要物性参数。传统的孔隙度计算方式基于线性方程,预测精度不高并费时费力。针对其问题,提出了基于粒子群优化(Particle Swarm Optimization,PSO)算法优化门控循环单元(Gated Recurrent Unit,GRU)神经网络的孔隙度参数预测模型。该模型可以很好地体现出孔隙度参数与测井曲线之间的非线性关系。首先构建GRU神经网络预测模型,然后利用具有全局优化能力的,更易收敛,鲁棒性较好的粒子群优化算法对GRU神经网络预测模型的超参数进行优化,有效提高模型的预测精度并减少交叉验证的时间。对探区实际测井数据进行相关性分析,挑选出与孔隙度参数相关度较高的测井数据,然后对PSO-GRU神经网络孔隙度参数预测模型进行训练和预测试验,并与传统的GRU神经网络预测模型以及反向传播(Back Propagation,BP)神经网络预测模型的结果进行比较分析,结果显示,PSO-GRU模型在孔隙度预测上具有较好的准确度。

Porosity parameter is one of the important parameters to characterize the oil storage capacity of rocks and is also an important physical parameter for reservoir evaluation.The traditional method of porosity calculation is based on linear equation,and the prediction accuracy is not high and time consuming.To solve the problem,a porosity parameter prediction model based on parti⁃cle swarm optimization(PSO)algorithm to optimize the gated recycling unit(GRU)neural network is proposed.This model can well reflect the nonlinear relationship between porosity parameters and logging curves.Firstly,the prediction model of GRU neural network is constructed,and then the super parameters of the prediction model of GRU neural network are optimized by the particle swarm optimization algorithm with global optimization ability,easier convergence and better robustness,which can effectively im⁃prove the prediction accuracy of the model and reduce the time of cross validation.Correlation analysis is carried out on the acreage of the actual logging data,sort out the logging data,and the porosity parameters so as to set then the PSO-porosity GRU helps neu⁃ral network training and parameter prediction model test,and compared with the traditional GRU helped neural network forecasting model and back propagation(BP)back propagation,the results of the neural network prediction model for comparative analysis,the results show that PSO-GRU helped on porosity prediction model has better accuracy.