Research On Oil Well Parameter Analysis And Classification Optimization Algorithm

With the advent of big data age,the knowledge of data mining can be found everywhere in our daily life,and also has some good effect on the process to informatization in the oil field.The paper based on knowledge of clustering and lasso,and at the same time,considering parameters related to oil well efficiency or production in Tianjin Dagang oil field,gives analysis on the classification and working condition of the oil wells.The algorithm makes the management of the oil wells easier and helps improving the production of the oil wells by giving regulating advice to the worker about the working condition of wells and trying to avoid bad working condition.The paper has accomplished the following parts:At first,according to the situation of oilfield management problem and inefficient production,clustering methods are proposed to classify oil wells based on parameters which relates to efficiency or yield of the oil wells,so as to facilitate the management of similar wells.If the oil well is in bad working condition,some adjustment of parameters must be done on the oil well.Three different clustering methods K-means,DBSCAN and hierarchical clustering method are selected depending on the number of wells in the block and the similarity between wells,and three methods are respectively used to analyze multi-well,wells in one block and similar wells.At the same time,a voting-K-means based on voting principle and K-means is proposed which improves the accuracy through voting to find the similarity between wells and at the same time uses K-means clustering to ensure the computational speed,and thus obtains a more generalized algorithm.Secondly,because the parameters selected are very complicated and the information may overlap with redundancy,the calculation process will be more difficult and will also result in calculation costs increase.Therefore,the idea of dimension reduction using principal component analysis(PCA)to reduce the dimension of the data before clustering is proposed.This method keeps most of the information and at the same time greatly improves the calculation speed.Thirdly,by using clustering method,the similarity between oil wells has been analyzed to facilitate the classification management,and this is a kind of horizontal analysis.At the same time,longitudinal analysis of the oil well is proposed.The parameters are predicted based on Lasso regression method,which provides a kind of early-warning to the system.Instead of predicting single parameter,a method to predict all the nine parameters at one time is proposed.By considering the relationship between the parameters,it can reach high accuracy and this conclusion has been proven.At last,based on Python simulation platform,the algorithms of clustering with PCA dimensionality reduction data processing and parameter prediction using Lasso regression are completed.By comparing the classification result with the working condition Dagang oilfield provides,it shows that the clustering algorithm can reach the accuracy of ninety percent and voting-K-means method has higher F1 value.The algorithms in this paper have shown high accuracy and speed,and have been used in the platform of oil well.It has some auxiliary decision-making effect on the oil well parameter optimization and production optimization,which makes contribution to improvement of the economic benefits of the oilfield.