Research On Analysis And Prediction Of Oilfield Development Data Based On Artificial Neural Network

The prediction of oilfield development index is an important way to realize scientific and reasonable oilfield development and ensure stable oilfield production.The commonly used methods for predicting oilfield development indicators mainly include reservoir numerical simulation,material balance method,classical formula method,etc.These methods are usually prediction models based on a small number of samples,which can be used to quickly predict development indexes after correction.However,when the sample data increases,the fitting effect often decreases,resulting in the decline of prediction accuracy.Because there is an unclear nonlinear relationship between oilfield development indicators and their influencing factors,the conventional prediction methods have some disadvantages,such as difficult model establishment,high error,large amount of calculation and so on.Aiming at this problem,the artificial neural network can be used to infinitely approximate the function of any nonlinear function,and the accurate prediction of oilfield development indicators can be realized.In this paper,the k-nearest neighbor algorithm is selected to preprocess the field data of the oilfield,analyze the oilfield development data from both qualitative and quantitative aspects,sort the contribution of the factors affecting the oil production of a single well by using the random forest algorithm,and determine the main influencing factors as the input variables of the subsequent genetic algorithm to optimize the BP neural network prediction model.By analyzing genetic algorithm and BP neural network,the prediction model of BP neural network optimized by genetic algorithm(GA)is established.Using this model,the prediction of water cut and cumulative oil production index of single well is realized,and the controlled reserves of single well are calculated by fitting,so as to predict the recovery of single well.According to the recovery prediction results and combined with the cumulative deviation coefficient method,two potential wells are selected from four typical oil wells,the feasibility and accuracy of genetic algorithm optimized BP(GA-BP)neural network prediction model are indirectly verified by Tong's chart.The GA-BP neural network prediction model is extended to predict 50 wells in the same block,and 15 potential wells are classified and selected.In order to improve the productivity of screened out potential wells and select fracturing stimulation measures,the factors affecting fracturing were analyzed from two aspects of well selection and layer selection,and the main factors affecting fracturing were determined.Aiming at the prediction problem of fracturing effect(oil increase amount,oil increase validity period),the whale optimization algorithm was introduced to optimize Elman neural network,and a four-layer structure WOAElman neural network prediction model was established.The prediction shows that after the implementation of fracturing measures,the potential well The production capacity has been greatly improved.The feasibility and effectiveness of the prediction model have been verified through mathematical methods(average relative error,root mean square error),and the WOA Elman neural network prediction model is extended to predict the fracturing effect(oil increase and effective period)of 15 potential wells in the same block.