Research On The Early Warning Technology Of Petroleum Drilling Engineering Lost Circulation Accidents

The whole structure of petroleum drilling engineering warning system is complex, harsh operating condition and working condition is changeable, currently, the main research direction of drilling engineering early warning system are real-time control and fault diagnosis towards each condition. The core technologies of fault detection and diagnosis technology include the sample, process, transmission of real-time data, the reject of abnormal data and the classification of different condition. In this paper, based on kernel principal component analysis method, on the purpose of realize the automation warning of petroleum drilling engineering system, we establish multiple kernel principal component analysis models to realize automatic obtain of drilling process information, dimensionality reduction and fault diagnosis.In this paper, directing at the complex multiple conditions of petroleum drilling, we assume the fault point of “lost circulation accident” is known and in-depth study the fault detection and diagnosis technology of the Early Warning system. The main research contents are as following:(1) Data processing: The original operation data obtained from the drilling field exists large amounts of process variables, such as SPP, and Flow out, existing strong correlation between each other, so these data must be pretreatment before analysis. The standardization of sample data is the foundation of data processing, the algorithm of abnormal data reject is mainly used to remove isolated abnormal data and continuous abnormal data, verified the reliability of basis on turning point of eliminating rate for the determine of adaptive sliding window length. In addition, we also comprehensive analysis the correlation between each variable, we extract some important characteristic variables(short-term variance, long-term variance, deviation) to describe the tendency of variable.(2) The classification of drilling condition: Directing at the early warning system of petroleum drilling engineering, this paper propose a new threshold classification method which can correctly classify each condition of drilling engineering. this method doesn’t need complicated calculation, only needs the data of drilling process and accurately preset the threshold parameter and the reference value can realize correct classification of each steady state condition. The drilling process is complicated, there exist strong correlation between variables and the fault type also presents diversity, if we use the commonly k-means clustering method to classify the sample data. the stability factor, classification index and membership degree can’t be accurately calculated just according to the drilling data. Threshold classification method doesn’t need to calculate the quantity, which can classify working condition by preset the threshold value of process variables.(3) Instance simulation: Because the research object is a nonlinear process, we consider to extend the single kernel principal component analysis model of fault detection method to the multiple kernel principal component analysis model, which can be applied to drilling engineering. The multiple kernel principal component analysis method of drilling engineering warning system can not only construct single kernel principal component analysis model throughout the whole drilling process, but also component multiple kernel principal component analysis model, which corresponds to different conditions. If there is an accident, then the multiple kernel principal component analysis model fault detection method based on threshold method can rapidly isolates the failure condition and introduces the corresponding fault detection module, therefore, the process control can be realized in different variable space and the corresponding fault detection statistic chart. Although they can’t directly determine the reason of the failure, they can show whether the process variable surpass the normal control limit through the statistical figure, then combine the test results with experience, we can finally determine fault types and causes of failure, realizing accurate and sensitive fault detection.