Prediction Of Reservoir Parameters Based On Logging Data

Reservoir parameter is one of the important parameters to characterize the oil storage capacity of rock,and it is also an important physical parameter for reservoir evaluation.The traditional calculation methods of reservoir parameters are based on linear equations,such as plate method,Archie formula and regression analysis,but these traditional methods can only consider the simple linear relationship between variables.In practical application,the reservoir situation is very complex,with obvious heterogeneity and anisotropy characteristics.The accuracy of the results obtained by traditional prediction methods is low,The error is large and time-consuming.Therefore,it is particularly important to find a method that can better reflect the mapping relationship between logging curves and reservoir parameters.In view of the above problems,this paper combines depth learning algorithm with reservoir parameter prediction,and designs a reservoir parameter prediction model based on depth learning algorithm to predict porosity parameters and permeability parameters respectively.The main work of this paper is as follows.At present,there is no method to directly detect the formation permeability.The common permeability calculation methods rely on the empirical model of porosity parameters and saturation parameters,such as Timur formula.The correlation between logging curves and permeability parameters is low.If logging curves are directly used as input characteristics to build the model,the accuracy of the model is low and the mapping relationship between logging curves and permeability cannot be fully reflected.In view of the above problems,this paper proposes a multi-source information fusion algorithm,which reconstructs multiple logging curves to obtain data with high correlation with permeability parameters as the input data of the model,so as to improve the accuracy and stability of the model.In this paper,convolution long-term and short-term neural network model and convolution gated neural network model are designed,which are respectively combined with convolution neural network,long-term and short-term neural network and gated recurrent neural network.Both models have the characteristics of local sensing and long-term memory.While considering the linear relationship between logging data and reservoir parameters,the convolution long-term and short-term neural network model and convolution gated recurrent neural network model have the characteristics of local sensing and long-term memory,The nonlinear mapping relationship between logging data and reservoir parameters,and the relationship between logging data trend with depth and reservoir parameters are also considered.The two models are compared and analyzed,and a better predictive model is obtained.In this paper,particle swarm optimization algorithm with global optimization ability is used to optimize the super parameters of neural network.The learning rate of neural network and the number of hidden layer neurons are regarded as the attributes of search particles in space,and the output error is used as the objective function.In order to improve the accuracy of model prediction,reduce the time wasted by the complexity of model parameter adjustment,and reduce the operation difficulty of the operator,reduce the difficulty for its application in practical work.