| With the development of drilling measurement technology,a large amount of drilling data has been generated,but information islands have formed between different well data,making it difficult to effectively utilize the massive drilling data.With the maturation of big data and artificial intelligence technologies,a new path has been opened for digital and intelligent drilling engineering to solve the problem of information silos in drilling data.In this paper,a distributed cluster based on Hadoop technology is constructed,and the principal component analysis feature extraction method is used to achieve data dimensionality reduction.A drilling rate optimization model based on NCPSOoptimized BP neural network is established,and drilling parameter optimization is achieved based on the NCPSO-BP model.Through multi-attribute feature extraction of the distributed cluster,drilling data can be effectively utilized,solving the problem of information silos,and improving drilling efficiency and performance.The main research contents and results are as follows:(1)Based on the Hadoop ecosystem,a simulated multi-wellsite interconnected petroleum distributed cluster is constructed.Using the idea of machine learning,the heterogeneous data bodies of multiple well sites are integrated through data cleaning and attribute analysis,and the integrated data is distributed and stored in each well site server,forming a physical distribution,logical Unified distributed well site cluster;(2)Designed a feature extraction method based on principal component analysis(PCA)to achieve dimensionality reduction of drilling data.By applying PCA,feature extraction was performed on high-dimensional drilling data based on variance contribution rate,which reduced the input dimensions of the drilling optimization model,thereby reducing the training model overhead and improving the training efficiency.When the cumulative variance contribution rate was selected as 90% and95%,the drilling data was reduced from the original 18 dimensions to 8 dimensions and 10 dimensions,respectively.The use of low-dimensional data after feature extraction for training improved the training speed by an average of 38.7% to 39.7%.This method not only maximizes the retention of complex attribute relationships among drilling data and ensures data integrity,but also reduces the complexity of the model.(3)Constructed NCPSO-BP hybrid algorithm drilling rate optimization model.Using the Niche Chaotic Particle Swarm Optimization algotithm(NCPSO)to optimize the BP neural network structure,and established the NCPSO-BP drilling rate optimization model,which improved the convergence speed and accuracy of the model and alleviated the problem of the model easily falling into local optima when dealing with complex data.Compared the accuracy of the models combining PSO,GA,NCPSO algorithms with BP network at different input dimensions.Under the input of 8 and 10 dimensions,the accuracy of the NCPSO-BP drilling rate optimization model increased by an average of 59%,and the training speed increased by an average of 26.3%.(4)Constructed a drilling parameter optimization method based on NCPSO-BP.Based on the output of the NCPSO-BP drilling rate optimization model,the drilling parameter combination was optimized.The results showed that the optimized drilling parameters increased the average drilling rate by 28.35% and decreased the average cost per foot by 11.47%.The optimized drilling parameter combination effectively improved construction efficiency and controlled drilling costs.The research focuses on the optimization of drilling parameters through multiattribute feature extraction based on a distributed cluster.It maximizes the utilization of high-dimensional and complex attributes of drilling data,ensures the integrity of information,and improves the speed and accuracy of drilling parameter optimization.This provides a theoretical basis for achieving drilling parameter optimization in the petroleum field under the background of big data. |
PDF Full Text Request