Research On Kick Intelligent Recognition Method Based On Multi-source Information Fusion Technology

Blowout is a major safety accident in the exploitation of oil and natural gas,and kick is a precursor to blowout.If the kick is discovered in time and well control measures are taken,occurrence of blowout will be avoided.The manual sitting is generally used to monitor the kick in the drilling site which is subjective and error-prone.The kick cannot be identified by using current single kick monitoring method and direct application of multiple monitoring methods may produce inconsistent monitoring results which makes the kick comprehensive judgment less reliable.To solve the above problems,this thesis proposed a kick comprehensive discriminant method based on SVM and improved D-S evidence theory,combining with monitoring data provided by drilling fluid outlet flow,logging and PWD to monitor the kick.On this basis,the completed work is listed as follows.Firstly,for the lack of kick accidents,the field measurement data is limited which makes the accuracy of the kick intelligent identification method low.In order to solve this problem,this thesis established numerical simulation models which were used for simulating kick detection parameters in normal drilling,pulling out of hole and running in hole conditions based on analyzing the relationship between kick detection parameters and drilling design parameters,geological engineering parameters,drilling conditions.The model could be used to simulate the kick sample data under specific conditions which provided a data foundation for the application of the intelligent kick identification method.The comparison between the simulation data and historical measured data could initially verify that the numerical simulation models had certain availability.Secondly,due to the methods for manual monitoring are subjective and error-prone,support vector machine(SVM)was proposed as a kick identification method by analysis and comparison.And the discriminant strategy of SVM was improved.The probability identification results of each method could be obtained by identifying the variation characteristics of various monitoring parameters separately.The experimental results showed that the accuracy of the SVM recognition method was improved compared with the traditional expert experience method,but it still could not meet the requirements of accurately identifying the kick.It is necessary to combine multiple means to monitor the kick.Thirdly,the direct application of multiple monitoring methods may produce inconsistent results which makes the kick comprehensive judgment less reliable.For this problem,a kick comprehensive discriminant method based on improved D-S evidence theory was proposed.By increasing the methods such as grading abnormal decision rejection and decision weighting,the contradictions and conflicts between the kick recognition results of different means were solved.The results of kick discriminant experiments showed that the accuracy of the method proposed in this paper had been greatly improved compared with single SVM recognition method.Finally,in order to verify the feasibility and effectiveness of the kick monitoring methods,this thesis developed an intelligent kick recognition software platform.The simulation experiments were carried on collecting or simulating kick monitoring parameters and real-time warning of kick by using the platform "simulation" feature.The results showed that the kick monitoring methods proposed in this thesis could realize the collecting or simulating of kick monitoring parameters,and accurately identifying the kick in real time,with certain feasibility and effectiveness.