Research On Early Warning Model Of Oilfield Abnormal Wells Based On Support Vector Regression

Currently, most of oil fields have been mined in the late stages. In the mining process,pump broken off, line perforation, electrical box failure and other accidents are often occurred due to oil shut shutdown. At last, oil production tends to be unstable. To solve these problems,we develop oil abnormal well warning system. It gives a timely warning information at the beginning of exception occurs in order to advance to take effective preventive measures to avoid accidents that caused unnecessary losses and ensure stable production.Most of oilfield early warning methods that are commonly used have the discrimination method of independent index threshold and the single warning mode, so they bring inaccurate warning results or they alarm instead of early warning when an abnormal event occurs. In order to settle these problems, puts forward the early warning model of oilfield abnormal wells based on support vector regression in this paper. The model that uses a combination of early warning methods is divided into three parts. The first, it selects and process wells exception parameters, then the data of resulting parameters is preprocessed. Next, it makes use of improved k- means algorithm to clustering preprocessed data to forming three kinds of well conditions model: normal, warning and abnormal. Last, support vector regression parameters optimized by the cross validation based on grid search, SVR machine defines and deals with the existence of edge data after clustering. Early warning model of oilfield abnormal wells based on support vector regression is set up to achieve oil wells abnormal warning analysis.On the basis of the model, we develop oil well abnormal warning system and the involved methods of system development process are detailed designed. To legacy data of a production plant sample, the results demonstrate the model has higher warning accuracy rate and the system based on also has great value.