Research On Geological Identification Based On Micro Inertial Measurement While Drilling System

With the rapid economic development in our country,the demand for energy has risen sharply and higher requirements have also been set for energy extraction technologies.Geological identification is a very important part of energy extraction technology.In the process of drilling,real-time identification of geological layer information of the position of drill bit can timely adjust the drilling program and improve the mining efficiency.Therefore,it is very important to study the method of geological identification.This paper studies the geological identification method based on the micro inertial drilling measurement system.It utilizes the advantages of micro-electromechanical systems(MEMS)to resist vibrations,high reliability and the advantages of the MWD system.By measuring the drill bit vibration during drilling signal and its processing,to identify the drill bit in which the geological layer.The main steps of geological identification methods are: denoising of geological signals,feature extraction of geological signals and recognition of geological signals.First of all,in order to reduce the impact of noise on the signal during the measurement while drilling,aiming at the problem that the empirical mode decomposition(EMD)denoising method is rough and the wavelet packet denoising method lacks self-adaptation,put forward a kind of MWD signal denoising method based on EMD-wavelet packet,through experimental comparison and verification of the method can well remove the noise while measuring the signal while drilling,and improve the signal to noise ratio of the signal.Secondly,according to the characteristics of the geological signal while drilling,the energy features of different geological signals are extracted by EMD decomposition.The non-negative matrix factorization(NMF)is used to reduce the dimension of the feature vector.Finally,the least squares support vector machine(LS-SVM),which is optimized by PSO algorithm,is used to classify the extracted eigenvectors.The result shows that the feature vectors after dimensionality reduction have a higher recognition rate.