Recognition Of Flow Petterns And Research On Complex Characteristics Of The Horizontal Oil-water Two-phase Flow

Two-phase flow is a common phenomenon in nature and industrial applications.With the deepening of oil exploration,oil fields with high water content have generally increased worldwide.In order to excavate the remaining potential of oil fields,horizontal extraction technology has gradually developed,which marks the second revolution of oil.Therefore,it is vitality for researching the dynamic characteristics of the oil-water two-phase flow and recognizing the flow patterns.They are beneficial to optimize the injection-production plan and improve the efficiency.Based on multivariate entropy and neural network perspective,the complexity of the horizontal oil-water two-phase flow is characterized and the flow pattern is distinguished in this paper.Entropy is implemented to characterize the complexity of the system and has guiding significance for the analysis of time series.Due to oil-water two-phase is a typical nonlinear system.Therefore,this paper proposes to utilize multivariate multiscale permutation entropy to analyze the complex characteristics of the oil-water two-phase flow.We obtain the multivariate multi-scale permutation entropy of the4-element conductance fluctuation signal.According to analyzing the MMPE curve,we can know that the entropy value at different scales can characterize the complex characteristics of the two-phase flow well and describe the transition between flow patterns.We propose to employ the normalized entropy rate to further explore the distribution of various flow patterns under different working conditions.In order to obtain the intrinsic mode function(IMF)of the same mode of the multivariate signal,this paper adopts multivariate empirical mode decomposition(MEMD)to deal with the multi-channel conductance fluctuation signal.So as to verify the MEMD is more suitable for simultaneous decomposition of multi-channel signals than empirical mode decomposition(EMD)and ensemble empirical mode decomposition(EEMD),we perform the MEMD,EMD,and EEMD on the Lorenz system.We select the calculated normalized IMF as the input of the extreme learning machine(ELM)to identify the flow patterns.The experimental results indicate that the combination of MEMD and ELM to recognize flow patterns is not only fast,but also has high generalization performance,and achieve a good effect.From the perspective of multiple information fusion,this paper proposes to fuse the original conductance fluctuation signal,energy characteristic signal and power spectrum characteristic signal together to form the fused characteristic information.We combine deep learning theory and build a one-dimensional deep neural network,select appropriate network model parameters,implement the fused characteristic information as training set.After experimental investigation,the results show that this method identify the five typical flow patterns of horizontal oil-water two-phase flow can achieve great accuracy.The method proposed in this paper provides a new idea for studying the flow pattern of two-phase flow based on deep learning theory.