Prediction Of Water Block Damage In A Bedrock Gas Reservoir Through The Application Of Genetic Algorithm Optimized Grey Neural Network

Based on the damage mechanism of water block in bedrock gas reservoir,this paper chooses gas permeability,porosity,gas-water interfacial tension,clay mineral content,contact angle,average pore throat radius and water saturation as the important influencing factors on the degree of water block damage,and obtains experimental data by water block damage evaluation experiment on 23 cores from three blocks of a bedrock reservoir in Qinghai Oilfield.The experimental results show that the bedrock reservoir is an ultra-low porosity,ultra-low permeability,water wetting,fine pore throat,high clay mineral and illite content reservoir,which indicates that the bedrock reservoir has water block damage potential.Subsequently,the grey relational analysis method is proposed to analyze the correlation between the proposed seven factors and water block damage.The results show that the selected seven factors have significant influence on water block damage.They can be used as inputs of neural network to predict water block damage.Among them,contact angle,water saturation and porosity have relatively high correlation with water block damage.However,the correlation of initial permeability,clay mineral content and water block damage is relatively weak.Although the previous researchers used the combination of grey system and neural network model to predict the water block damage and obtained relatively high prediction accuracy,the BP neural network model did not perform ideally,the weights and bias of the neural network are still randomly initialized,and the prediction accuracy still has room for improving.Therefore,genetic algorithm is proposed to optimize the initial weights and bias of the grey neural network,which can help the network to obtain global optimal or global suboptimal solutions fast,and further improve the prediction accuracy of the grey neural network model.By programming with MATLAB software,77 collected sample data are grouped and predicted.The results show that the mean relative error of the testset for predicting water block damage by the genetic algorithm optimized grey neural network model is 5.1%,which is a better prediction accuracy compared to 6.1%predicted by grey neural network model itself,and 10.3% predicted by BP neural network.At the same time,the performance of the optimized network has been greatly improved.