CN117872487A - Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field - Google Patents
Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field Download PDFInfo
- Publication number
- CN117872487A CN117872487A CN202410264266.0A CN202410264266A CN117872487A CN 117872487 A CN117872487 A CN 117872487A CN 202410264266 A CN202410264266 A CN 202410264266A CN 117872487 A CN117872487 A CN 117872487A
- Authority
- CN
- China
- Prior art keywords
- noise
- layer
- net
- data
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 90
- 230000005291 magnetic effect Effects 0.000 title claims abstract description 32
- 239000003921 oil Substances 0.000 title claims abstract description 19
- 239000010779 crude oil Substances 0.000 title claims abstract description 18
- 238000000034 method Methods 0.000 claims abstract description 88
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 50
- 230000008569 process Effects 0.000 claims abstract description 30
- 238000005481 NMR spectroscopy Methods 0.000 claims abstract description 25
- 238000009826 distribution Methods 0.000 claims abstract description 15
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000008030 elimination Effects 0.000 claims abstract description 11
- 238000003379 elimination reaction Methods 0.000 claims abstract description 11
- 230000005284 excitation Effects 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 51
- 238000012360 testing method Methods 0.000 claims description 30
- 230000004913 activation Effects 0.000 claims description 21
- 238000005303 weighing Methods 0.000 claims description 18
- 239000003673 groundwater Substances 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 13
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 claims description 12
- 238000004891 communication Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 125000004435 hydrogen atom Chemical group [H]* 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 6
- 230000001965 increasing effect Effects 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000001228 spectrum Methods 0.000 claims description 5
- 238000000685 Carr-Purcell-Meiboom-Gill pulse sequence Methods 0.000 claims description 3
- IIDJRNMFWXDHID-UHFFFAOYSA-N Risedronic acid Chemical compound OP(=O)(O)C(P(O)(O)=O)(O)CC1=CC=CN=C1 IIDJRNMFWXDHID-UHFFFAOYSA-N 0.000 claims description 3
- 238000013500 data storage Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- GNFTZDOKVXKIBK-UHFFFAOYSA-N 3-(2-methoxyethoxy)benzohydrazide Chemical compound COCCOC1=CC=CC(C(=O)NN)=C1 GNFTZDOKVXKIBK-UHFFFAOYSA-N 0.000 claims description 2
- 230000001939 inductive effect Effects 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 6
- 239000002957 persistent organic pollutant Substances 0.000 description 6
- 239000004215 Carbon black (E152) Substances 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 229930195733 hydrocarbon Natural products 0.000 description 3
- 150000002430 hydrocarbons Chemical class 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 239000000523 sample Substances 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- YTAHJIFKAKIKAV-XNMGPUDCSA-N [(1R)-3-morpholin-4-yl-1-phenylpropyl] N-[(3S)-2-oxo-5-phenyl-1,3-dihydro-1,4-benzodiazepin-3-yl]carbamate Chemical compound O=C1[C@H](N=C(C2=C(N1)C=CC=C2)C1=CC=CC=C1)NC(O[C@H](CCN1CCOCC1)C1=CC=CC=C1)=O YTAHJIFKAKIKAV-XNMGPUDCSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 239000013256 coordination polymer Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000002283 diesel fuel Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000005298 paramagnetic effect Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 239000003403 water pollutant Substances 0.000 description 1
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention relates to the field of geophysical method detection of underground water organic pollution, in particular to a magnetic resonance mobile detection method suitable for crude oil leakage detection of an oil field. The receiving coil is carried on the supporting framework and is pulled by the mobile detection vehicle, the mobile detection vehicle is used for step-by-step mobile scanning in the excitation area of the transmitting coil, the organic pollution of the underground water in the excitation area of the transmitting coil is detected, and the distribution condition of the organic pollution of the underground water is explained through inversion. In the data processing process, firstly, the collected actually measured noise-containing nuclear magnetic resonance mobile detection signals are arranged, expanded and screened to form an original noise-containing signal and a pure noise training set, then an expanded convolutional neural network module and a balance device are connected in parallel to form a novel convolutional neural network, a complex nonlinear mapping relation between the original noise-containing signal and noise data in the original noise-containing signal is constructed, the noise data in the original noise-containing signal is removed from the original noise-containing signal, and intelligent noise elimination with higher accuracy is realized.
Description
Technical Field
The invention relates to the field of geophysical method detection of underground water organic pollution, in particular to a magnetic resonance mobile detection method suitable for crude oil leakage detection of an oil field.
Background
Petroleum is called blood of national economy, is an important strategic resource of the country, and plays a strategic supporting role in national economy and safety. The method is widely applied to the fields of transportation, industrial production, chemical raw materials and the like, and solves the large energy demand of human beings. In order to ensure the safety of the ecological environment and accelerate the treatment, providing accurate information for the repair work of the organic pollution of the underground water and enhancing the detection level of the organic pollutant of the underground water, the method has become urgent. At present, the geophysical detection method for the organic pollution of the underground water is numerous, but the method belongs to indirect detection and has multiple resolvability. Because the groundwater and the organic pollution contain a large amount of hydrogen protons, the magnetic resonance technology utilizes the principle that the nuclear energy of paramagnetic hydrogen protons generates the magnetic resonance relaxation phenomenon, and provides an effective technical approach for directly detecting the groundwater organic matters. The ground magnetic resonance technology utilizes an artificial field to excite hydrogen protons in underground water and organic pollutants to generate larmor precession by applying alternating current to a transmitting coil paved on the ground from small to large, and then uses a receiving coil to measure precession signals, so that the distribution information of the organic pollutants in the underground water can be obtained from shallow to deep, but the distribution information is applied to detection of the organic pollutants in the underground water, but the problems that the magnetic resonance technology has few measurement points, long time consumption of each measuring point, low detection efficiency, serious dependence on artificial experience in data processing and the like are found in the actual field detection process, so that the technology capable of directly detecting the organic pollutants in the underground water is explored, and the space distribution of the organic pollutants is rapidly defined in a large range from the field scale, and the technology has important significance to the actual application.
In the aspect of a magnetic resonance detection method, a traditional detection method is a four-channel nuclear magnetic resonance signal full-wave acquisition system and an acquisition method disclosed by CN103955004A, a computer is connected with a power management module through a controller, a high-speed digital I/O card and the controller, wherein a preamplifier effectively resists saturation of the amplifier, Q_SWITCH shortens dead time, improves signal to noise ratio, noise signals are transmitted remotely by current, signal attenuation in the transmission process is restrained, the acquired nuclear magnetic resonance signals are subjected to data processing by a self-adaptive reference noise elimination algorithm, and the anti-interference capability of the instrument and the transverse resolution and accuracy of underground water distribution measurement are improved. However, the invention lays a plurality of coils, the wiring is complex, the workload is large, and the measuring work consumes long time and has low efficiency; the system is built with larger limitation of the topography of the detection area, and enough coils cannot be paved in the detection area with small area, so that the number of measurement points is small; reducing the size of the receiving coil without increasing the number of turns reduces the effective receiving area of the coil, resulting in reduced depth and accuracy of detection.
In terms of magnetic resonance detection data processing, a traditional processing method is disclosed as CN113655534A, namely a method for suppressing noise of a nuclear magnetic resonance FID signal based on multi-linear singular value tensor decomposition, wherein at each interval time delta t, one channel is enabled to start sampling, and a multi-channel signal is obtained; converting each channel signal into a Hankel matrix to form a third-order tensor A, performing Tucker decomposition on the A, and processing to obtain A new Will A new Restoring to multichannel signals, performing CP tensor decomposition processing on the second-order tensor X formed by the multichannel signals, and finally obtaining the high signal-to-noise ratio signal X after multi-communication signal fusion new The signal-to-noise ratio of the FID is improved, and the limitation of the current algorithm under the interference of strong noise is effectively overcome. However, the matrix related to the invention needs to set parameters manually, relies on manual priori experience and has low universality.
Aiming at the problem of organic pollution detection of groundwater, CN103852794A discloses a device and a method for detecting hydrocarbon pollution shallow groundwater by magnetic resonance, and the invention utilizes a low-field magnetic resonance instrument to detect the pollution sample prepared by mixing hydrocarbon substances represented by gasoline and diesel oil with water in a laboratory, thereby realizing non-invasive quantitative qualitative measurement and rapidly obtaining a test result on site. However, the invention has the advantages of limited sampling quantity, small measuring range, difficult comprehensive understanding of pollution distribution, and limited detection depth, and can only detect hydrocarbon pollution in 5 meters underground.
With the development of deep learning, in recent years, a neural network is gradually applied to the field of magnetic resonance detection data processing, li Bang is published in Jilin university journal (earth science edition), [2022, 52 (3), 775-784] and paper, "groundwater magnetic resonance data random noise suppression method based on convolutional neural network" is based on convolutional neural network framework, and a training mode of supervised learning is adopted to obtain a nonlinear mapping relation between a time spectrum of a noise-containing signal and a time spectrum of an original noise-free signal, so as to further realize suppression of magnetic resonance signal noise. However, the method is complicated in data pre-processing, long in training time, weak in generalization capability and strong in dependence on data.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a magnetic resonance mobile detection method suitable for crude oil leakage detection in an oil field, which is characterized in that a receiving coil is carried on a supporting framework, a mobile detection vehicle is used for traction, step-by-step mobile scanning is carried out in a transmitting coil, organic pollution of groundwater in an arrangement area of the transmitting coil is detected, and the distribution condition of the organic pollution of the groundwater is obtained by processing data. The method solves the problems of complex wiring and large workload in the detection process in the traditional detection method, and saves the detection time; and the receptive field is improved by introducing an expansion convolution layer in data processing, the attention degree of a model to noise characteristics is improved by introducing a weighting apparatus, a complex mapping relation between an original noise-containing signal and noise data in the original noise-containing signal is constructed, the noise data in the original noise-containing signal is removed from the original noise-containing signal, and a noise-removed signal is obtained, so that the problems that the traditional method depends on manual priori experience, the generalization capability of a common convolution neural network is weak, and the error of the noise characteristic learning result is large are solved.
The present invention has been achieved in such a way that,
a magnetic resonance mobile detection method suitable for crude oil leak detection in an oilfield, the method comprising:
a. setting up a magnetic resonance mobile detection system in a region;
b. according to the detection direction, the field shape and the position of the transmitting coil, the speed of the mobile detecting vehicle carrying the receiving coil is controlled, so that the mobile detecting vehicle pulls the supporting framework carrying the receiving coil to run on the measuring line at a constant speed and step-by-step mobile scanning is performed in the transmitting coil;
c. setting current parameters of a control system by using an upper computer; after the setting is finished, the transmitter is controlled by a control system, and a transmitting current with excitation pulse as a CPMG sequence and frequency as local Larmor frequency is introduced into a transmitting coil from small to large to excite spin directions of hydrogen protons in groundwater and oil to generate a wrenching angle;
d. turning off the emission current, recovering the spin of hydrogen proton to the initial direction, inducing by the receiving coil and collecting nuclear magnetic resonance signal by the receiver; the position of the receiving coil on the measuring line is used as a measuring point, and the position information of the receiving coil is recorded by a GPS (global positioning system) arranged on the receiving coil; after the signal acquisition is finished, the receiver transmits data to the upper computer through the wireless receiving communication module for data storage;
e. setting the emission current to beC-d, namely repeatedly changing the current value for multiple times to generate alternating current fields with different excitation pulse moments, carrying out layered scanning on underground organic pollution distribution, and repeating the steps to obtain a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals;
f. setting the emission current to beRepeating the steps c-d to obtain a plurality of groups of actually measured pure noise data;
g. and processing the acquired multiple groups of actually measured noise-containing nuclear magnetic resonance mobile detection signals and multiple groups of actually measured pure noise data to obtain the actual situation of underground organic pollution distribution.
Further, the step g specifically includes:
1) Amplitude scaling the acquired multiple groups of actually measured noise-containing nuclear magnetic resonance movement detection signals to expand the signals into data with the number ofLIs the original noise-containing signal of (a)YDividing the data into model training sets according to the ratio of the data number ratio of 4:1And model test set->WhereinpAndqrespectively represent model training setsY 1 And model test setY 2 Is used for the data amount of the data,p、qsatisfy->And->；/>Representative ofY 1 Is provided with a data set of a certain data,；/>representative ofY 2 Some data of (a), is +>The method comprises the steps of carrying out a first treatment on the surface of the Screening and model training set from multiple groups of collected actual measurement pure noise data according to the principle of corresponding sampling timeY 1 Pure noise data corresponding to the medium data to form a pure noise training set +.>Wherein->Represent the firstiNoise (S)>；
2) Constructing network parameters asθ 1 Is a convolutional neural network moduleNet 1 (θ 1 ) And network parameters areθ 2 Weight of (2)Net 2 (θ 2 ) Will beNet 1 (θ 1 ) And (3) withNet 2 (θ 2 ) Parallel connection, constructing network parameters asθNovel convolutional neural network model of (2)Net(θ)；
3) Training set of modelsY 1 The first of (3)iThe original noise-containing signalsInput to a novel convolutional neural network modelNet(θ) During training, the convolutional neural network module is expandedNet 1 (θ 1 ) Model training setY 1 Original noise-containing signal->Preliminary noise feature learning is carried out to obtain preliminary noise data +.>Weighing deviceNet 2 (θ 2 ) Training set for modelY 1 Original noise-containing signal->Corresponding weights are given to different characteristic information of the noise, wherein the preliminary noise data of the noise +.>Is given the highest level weight to make preliminary noise data +>Close to pure noise training setNPure noise data in the original noise-containing signal to obtain noise data +.>Establishing an original noise-containing signal->Noise data +.>Nonlinear mapping relation between ∈>Noise data in the original noise-containing signal is +.>From the original noisy signal->Removing the middle part and outputting the denoised signal +.>；
4) Test set of modelsY 2 The data in the model is input into a novel convolutional neural network model with complete trainingNet(θ) Testing; when the root mean square error of the novel convolutional neural network test set is less than 10 -2 During training, the novel convolutional neural network model is completedNet(θ) Otherwise, adjusting network parameters, and repeating the steps 2) -4) until reaching the standard;
5) Inputting a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals into a novel convolutional neural network model after the test reaches the standardNet(θ) The denoising operation is performed to obtain a denoising result; and carrying out inversion interpretation on the denoising result to obtain the actual situation of underground organic pollution distribution.
Further, the convolutional neural network module is expandedNet 1 (θ 1 ) The device comprises an input layer, an expansion operation layer and an output layer in sequence; the input layer comprises a convolution layer and a linear activation function, and a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals are read; the expansion operation layer comprises a plurality of alternating series arrangement of a first sub-module layer and a second sub-module layer, wherein the first sub-module layer sequentially comprises a convolution layer, a batch normalization layer and a linear activation function, and the second sub-module layer sequentially comprises an expansion convolution layer, a batch normalization layer and a linear activation function; the output layer comprises a rollLamination, reconstruct and output the noise information; weighing apparatusNet 2 (θ 2 ) Comprises a convolution layer, a dense module, a global average pooling layer, a convolution layer,sigmoidThe activation function is connected in series; wherein the dense module comprises four sub-module layers, the first three sub-module layers have the same structure and all comprise a convolution kernel with the size ofA batch normalization layer and a linear activation function, the last submodule layer comprising a convolution kernel of size +.>A batch normalization layer and a linear activation function; in a dense module, each sub-module layer is interconnected, i.e. each sub-module layer splices inputs of all sub-module layers arranged in front of it as additional inputs and delivers the output characteristic information to all sub-module layers arranged behind it.
Further, the training process specifically includes:
(1) By combining the original noise-containing signalsy i Input to an expanded convolutional neural network moduleNet 1 (θ 1 ) In (a) and (b); expanding convolutional neural network moduleNet 1 (θ 1 ) Middle (f)kReceptive field of layer convolution layerThe size is as follows:
，
wherein the firstkThe layer convolution layer is an expansion convolution neural network moduleNet 1 (θ 1 ) Is provided with a convolution layer of any one layer,is the firstk-Layer 1 convolutional layerSize of receptive field, ->Is the firstkConvolution kernel size of layer convolution layer, +.>Is the firstmStep size of layer convolution layer,/>The method comprises the steps of carrying out a first treatment on the surface of the Adding intervals into a basic convolution kernel by the expansion convolution layer, and increasing the size of the convolution kernel;
signal warp of the firstkLayer convolution layer operation to obtain outputThe method comprises the following steps: />，
In the method, in the process of the invention,is the firstkThe input of the layer convolution layer is convolution operation,/->For linear activation function +.>Is the firstkBias of layer convolution layers;
when the first iskLayer is expansion convolution neural network moduleNet 1 (θ 1 ) At the time of the last layer of (c),to expand convolutional neural network moduleNet 1 (θ 1 ) From original noisy signalsy i Preliminary noise data learned in +.>；
Then the original noise-containing signal is processedy i Input deviceTo a balance deviceNet 2 (θ 2 ) In the method, the signal is subjected to preliminary feature extraction through two convolution layers:
，
in the method, in the process of the invention,for the extracted features, < >>，/>The first layer convolution and the second layer convolution operation after the original noise-containing signal enters the weighing apparatus are respectively carried out;
deep extraction is carried out on the characteristics by the signal through the dense module, and the obtained result is:
，
in the method, in the process of the invention,deep extracted features for dense modules, +.>Is a dense module operation;
the signal is processed by global average pooling layer, two convolution layers andSigmoidactivating the function to obtain the weight coefficient of the noise:
，
in the method, in the process of the invention,is a weight coefficient>Is thatSigmoidFunction (F)>For the global average pooling operation,and->Performing first-layer convolution and second-layer convolution operation after global average pooling operation;
imparting noise from original signaly i Preliminary noise data learned in (1)The highest weighting factor->To increase the model's attention to noise, improve preliminary noise data +.>Make it more similar to pure noise training setNPure noise data in (a); weight coefficient->And preliminary noise data->Combining in a certain way to obtain noise data in the original noise-containing signal>：
，
From the input raw noisy signaly i Removing noise data from original noise-containing signalObtaining the denoised signalx i As a novel convolutional neural network modelNet(θ) Final of (2)And (3) outputting:
；
(2) Setting up a loss function optimization model to reduce noise data in an original noise-containing signalAnd pure noise training setNThe difference between the pure noise data in the model (a) and the loss function adopts a gradient descent algorithm:
，
in the method, in the process of the invention,，/>as a function of the losses of the different networks,ptraining set for modelY 1 Is used for the data amount of the (a),to input original noise-containing signalsy i The output of the different networks is then,y i and (3) withx i Is a group of corresponding original noise-containing signals and signals after noise elimination, < + >>Is F norm;
from expanding convolutional neural network modulesNet 1 (θ 1 ) Weighing deviceNet 2 (θ 2 ) Obtaining a final loss function from the loss function of (2)：
，
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) Is a loss function of->For weighing deviceNet 2 (θ 2 ) A loss function of (2);
through the final loss function, network parameters are continuously adjusted, network noise elimination performance is optimized, an ideal model is obtained through training, and the root mean square error of a test set is smaller than 10 at the moment -2 。
Further, the step of performing the test specifically includes:
model test setInput to a trained novel convolutional neural network modelNet(θ) In the method, the expanded convolutional neural network module is obtainedNet 1 (θ 1 ) Prediction output of +.>，/>：
，
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) From a model test setY 2 In the original noisy signalPreliminary noise data learned in (a);
and then the preliminary noise dataWeighing deviceNet 2 (θ 2 ) Prediction output->Combining to obtain predictive noise->：
，
Finally, the original noise-containing signalPrediction noise->Removing to obtain a novel convolutional neural network modelNet(θ) Output of +.>I.e. denoised signal->：
。
Further, inversion interpretation includes: inputting the noise elimination result into multi-index inversion software to obtain transverse relaxation timeT 2 Graph related to nuclear magnetic resonance signal amplitude and observation of transverse relaxation timeT 2 The method comprises the steps of carrying out a first treatment on the surface of the Recording transverse relaxation timeT 2 Obtaining the organic pollution of the groundwater of different material components; recording transverse relaxation timeT 2 The spectrum peak area of the underground water organic pollution content of different material components is obtained.
The invention has the following advantages and beneficial effects:
compared with the existing magnetic resonance detection technology, the magnetic resonance mobile detection method suitable for crude oil leakage detection in the oil field can reduce the coil laying times, so that the workload of building an exploration system is reduced, and the detection time is shortened. In the method, a novel convolutional neural network is built in the data processing process, the actually measured noise-containing nuclear magnetic resonance movement detection signal is taken as input, a network model is continuously optimized, the network generalization capability and the noise learning capability are enhanced, noise data are accurately removed from an original noise-containing signal, and therefore a noise-removed signal is obtained; compared with a common convolutional neural network, the method has the advantages that the expanded convolutional layer is introduced, the receptive field is enlarged, more information is learned from the input signal by the model, the expanded convolutional layer and the convolutional layer are used at intervals, loss of effective information is avoided, network feature extraction capacity is improved, meanwhile, the weight balancer is introduced, high weight is given to a noise part, the attention of the network to noise is increased, and accuracy of model to noise feature learning is improved.
Drawings
FIG. 1 is a schematic diagram of a field work layout of a system used in a magnetic resonance mobile detection method suitable for crude oil leak detection in an oilfield;
FIG. 2 is a block diagram of a noise cancellation model in a magnetic resonance mobile detection method suitable for crude oil leakage detection in an oil field;
the system comprises a first wireless transmitting communication module, a second wireless receiving communication module, a 3 upper computer, a 4 control system, a 5 transmitter, a 6 power supply, a 7 transmitting coil, a 8 receiver, a 9 receiving coil, a 10GPS (global positioning system), a 11 mobile probe vehicle, a 12 bearing framework, a 13 measuring line, 14 measuring points, a 15 convolution layer, a 16 linear activation function, a 17 batch normalization layer, a 18 expansion convolution layer, a 19 intensive module, a 20 global average pooling layer and a 21sigmoid activation function.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The magnetic resonance mobile detection method suitable for crude oil leakage detection in the oil field in the embodiment comprises the following steps:
referring to fig. 1 in combination with fig. 2, a magnetic resonance movement detection method suitable for crude oil leakage detection in an oil field in this embodiment includes:
a. referring to fig. 1, a magnetic resonance mobile detection system suitable for oil field crude oil leakage detection is built in a detection area, the system comprises a transmitter 5, a receiver 8, an upper computer 3, a control system 4, a transmitting coil 7, a receiving coil 9, a first wireless transmitting communication module 1, a second wireless receiving communication module 2 and a mobile detection platform, and power is supplied through a power supply 6, wherein the control system 4 controls the receiver 8 to be connected with the transmitter 5, the transmitter 5 is communicated with the upper computer 3 through the first wireless transmitting communication module 1, the receiver 8 is connected with the receiving coil and is communicated with the upper computer 3 through the second wireless receiving communication module 2, the mobile detection platform comprises a mobile detection vehicle 11, a movable supporting framework 12, the receiver 8 is mounted on the mobile detection vehicle 11, the receiving coil 9 is mounted on the supporting framework 12, and a GPS10 is mounted on the receiving coil 9;
b. arranging a measuring line 13 according to the size and shape of a polluted site and the position of the transmitting coil 7, setting a detection direction, controlling the speed of the mobile detection vehicle 11, enabling the supporting framework 12 to uniformly run on the measuring line 13, and performing step-by-step mobile scanning in the transmitting coil 7;
c. the upper computer 3 is used for setting current parameters of the control system 4, the frequency of the emission current is 2330Hz, the current size is 5A, then the emitter 5 is controlled by the control system 4, the emission current with excitation pulse as a CPMG sequence is fed into the emission coil 7, the emission time is 60ms, and the spin direction of hydrogen protons in groundwater and oil is excited to generate a wrenching angle;
d. after the emission current is turned off, the spin of hydrogen protons is restored to the initial direction, and the receiving coil 9 senses and the receiver 8 collects nuclear magnetic resonance signals; at this time, the position of the receiving coil 9 on the measuring line 13 is set as a measuring point 14, and the position information is recorded by the GPS10 attached to the receiving coil 9; after the signal acquisition is finished, the receiver 8 transmits data to the upper computer 3 for data storage through the wireless receiving communication module 2;
e. setting the emission current to beChanging the current value for 10 times from small to large to generate alternating current fields with different excitation pulse moments, carrying out layered scanning on underground organic pollution distribution, and repeating the c-d process for 10 times to obtain 10 groups of actually measured noise-containing nuclear magnetic resonance movement detection signals;
f. setting the emission current to beRepeating the c-d process for 10 times to obtain 10 groups of actual measurement pure noise data;
g. amplitude scaling transformation is carried out on the acquired actually-measured noise-containing nuclear magnetic resonance movement detection signal, and the amplitude scaling transformation is expanded into an original noise-containing signal with the data number of 10240YDividing the data into model training sets according to the ratio of the data number ratio of 4:1And model test set->WhereinpAndqrespectively representY 1 AndY 2 the amount of data in the data set is,p=8192，q=2048the method comprises the steps of carrying out a first treatment on the surface of the Screening and model training set from multiple groups of collected actual measurement pure noise data according to the principle of corresponding sampling timeY 1 Pure noise data corresponding to the medium data to form a pure noise training setWherein->Represent the firstiNoise (S)>；
h. Referring to fig. 2, the network parameters are constructed as followsθ 1 Is a convolutional neural network moduleNet 1 (θ 1 ) And network parameters areθ 2 Weight of (2)Net 2 (θ 2 ) Will beNet 1 (θ 1 ) And (3) withNet 2 (θ 2 ) Parallel connection, constructing network parameters asθNovel convolutional neural network model of (2)Net(θ) The method comprises the steps of carrying out a first treatment on the surface of the Initializing network parametersθSetting it to 0, the initial learning rate isThe training period is 50 rounds;
i. training the modelY 1 The first of (3)iThe original noise-containing signalsInput to a novel convolutional neural network modelNet(θ) During training, the convolutional neural network module is expandedNet 1 (θ 1 ) For the original noise-containing signal in the model training set +.>Preliminary noise feature learning is carried out to obtain preliminary noise data +.>Weighing deviceNet 2 (θ 2 ) As preliminary noise dataGiving a weight of 0.9 to make preliminary noise data +.>Close to pure noise training setNPure noise data in the original noise-containing signal to obtain noise data +.>Establishing an original noise-containing signal->Noise data +.>Nonlinear mapping relation between ∈>And noise data in the original noise-containing signal +.>From the original noisy signal->Removing the middle part and outputting the denoised signal +.>；
j. Model test setY 2 The data in the model is input into a novel convolutional neural network model with complete trainingNet(θ) Testing; when the root mean square error of the novel convolutional neural network test set is less than 10 -2 When the new convolutional neural network model after training reaches the standard, otherwise, adjusting network parameters, and repeating the steps h-j until the new convolutional neural network model reaches the standard;
k. inputting the actually measured noise-containing nuclear magnetic resonance movement detection signal into a novel convolutional neural network model with standard testNet(θ) The denoising operation is performed to obtain a denoising result; and carrying out inversion interpretation on the denoising result to obtain the actual situation of underground organic pollution distribution.
Referring to fig. 1, the transmitting coil 7 in the step a adopts a rectangular large coil; the receiving coil 9 adopts a multi-turn small coil, the multi-turn structure ensures the effective receiving area of the coil, and the problems of detection depth and precision reduction caused by reducing the size of the coil so as to make the measurement work more convenient and flexible are avoided; the supporting framework 12 in the mobile detection platform needs to keep a certain distance with the mobile detection vehicle 11, so that noise interference generated in the running process of the detection vehicle is avoided, and the maximum intensity of signals received by the coil is ensured; the supporting framework 12 is made of hard plastic, is light and has strong plasticity, and is made into a foldable small-size object block, so that the supporting framework is convenient to transport and carry.
In the step d, the distance between adjacent measuring points 14 along the direction of the measuring line 13 is 8m.
Referring to fig. 2, the expanded convolutional neural network module in the step hNet 1 (θ 1 ) The expansion operation layer is formed by an input layer, an expansion operation layer and an output layer; the input layer comprises a convolution layer 15 and a linear activation function 16, and is used for reading the actually measured noise-containing nuclear magnetic resonance movement detection signal; the expansion operation layer comprises twelve alternating serial arrangements of a first sub-module layer and a second sub-module layer, wherein the first sub-module layer comprises a convolution layer 15, a batch normalization layer 17 and 1 linear activation function 16, the second sub-module layer comprises an expansion convolution layer 18, a batch normalization layer 17 and a linear activation function 16, the use of the expansion convolution layer 18 can improve the receptive field, and meanwhile, the expansion convolution layer 18 in the second sub-module layer is arranged at intervals with the convolution layer 15 in the first sub-module layer, so that the problem of information loss caused by continuous use of a large number of expansion convolution layers 18 is avoided; the output layer comprises a convolution layer 15 for reconstructing and outputting the noise information. Weighing apparatusNet 2 (θ 2 ) Comprising four convolution layers 15, a dense module 19, a global averaging pooling layer 20 and asigmoidThe function 21 is activated, wherein the following are sequentially performed in the order of data processing: two convolution layers 15, a dense module 19, a global average pooling layer 20, two convolution layers 15 and onesigmoidActivating function 21; the dense module 19 comprises four sub-module layers, wherein the first three sub-module layers are identical in structure and each comprise a convolution kernel of sizeA batch normalization layer 17 and a linear activation function 16, the last submodule layer comprising a convolution kernel of size +.>A batch normalization layer 17 and a linear activation function 16; in the dense module 19, each sub-module layer is connected to each other, that is, each sub-module layer splices inputs of all sub-module layers arranged in front of the sub-module layer, and transmits the output characteristic information to all sub-module layers arranged behind the sub-module layer as an additional input, so as to ensure maximum information transmission between the sub-module layers.
In the step i, the model training step specifically includes:
(1) By combining the original noise-containing signalsy i Input to an expanded convolutional neural network moduleNet 1 (θ 1 ) In (a) and (b); expanding convolutional neural network moduleNet 1 (θ 1 ) Middle (f)kReceptive field of layer convolution layerThe size is as follows:
，
wherein the firstkThe layer convolution layer is an expansion convolution neural network moduleNet 1 (θ 1 ) Is provided with a convolution layer of any one layer,is the firstk-The receptive field size of the 1-layer convolution layer, < >>Is the firstkConvolution kernel size of layer convolution layer, +.>Is the firstmStep size of layer convolution layer,/>The method comprises the steps of carrying out a first treatment on the surface of the Expanding the convolution layer 18 adds space in the base convolution kernel increasing the convolution kernel sizeThereby improving the neural network receptive field, and as the signal passes through the dilated convolution layer 18, more detailed characteristic information is learned by the model;
signal warp of the firstkThe layer convolution layer 15 operates to obtain an output of：
，
In the method, in the process of the invention,is the firstkThe input of layer convolution layer 15 is convolution operation,/>In order to activate the function 16 in a linear fashion,is the firstkBias of layer convolution layer 15;
when the first iskLayer is expansion convolution neural network moduleNet 1 (θ 1 ) Is used for the first layer of the film,kwhen the value of the ratio is =14,to expand convolutional neural network moduleNet 1 (θ 1 ) From original noisy signalsy i Preliminary noise data learned in +.>；
Then the original noise-containing signal is processedy i Input to a balanceNet 2 (θ 2 ) In (3), the signal is subjected to preliminary feature extraction by the two convolution layers 15:
，
in the method, in the process of the invention,for the extracted features, < >>，/>The first layer convolution and the second layer convolution operation after the original noise-containing signal enters the weighing apparatus are respectively carried out;
next, the signal is subjected to deep extraction of features by the dense module 19:
，
in the method, in the process of the invention,deep extracted features for dense module 19, < >>Operate for dense module 19;
next, the signal is passed through a global averaging pooling layer 20, two convolution layers 15 andSigmoidactivating function 21 yields the weighting coefficients of the noise:
，
in the method, in the process of the invention,is a weight coefficient>Is thatSigmoidFunction (F)>For the global average pooling operation,and->Convolution operations of the 1 st convolution layer 15 and the 2 nd convolution layer 15 after the global average pooling operation;
to be from the original noise-containing signaly i Preliminary noise data learned in (1)Giving weight coefficient->To increase the attention of the model to noise characteristics, to improve preliminary noise data->Make it more similar to pure noise training setNPure noise data in (a); weight coefficient->And preliminary noise data->Combining in a certain way to obtain noise data in the original noise-containing signal>：
，
Finally, from the original noisy signal inputy i Removing noise data from original noise-containing signalObtaining the denoised signalx i As a novel convolutional neural network modelNet(θ) Output of (2):
；
(2) Setting up an effective loss function, optimizing a model, and reducing noise data in an original noise-containing signalAnd pure noise training setNThe difference between the pure noise data in the model (a) and the loss function adopts a gradient descent algorithm:
，
in the method, in the process of the invention,，/>as a function of the losses of the different networks,ptraining set for modelY 1 Is used for the data amount of the (a),to input original noise-containing signalsy i The output of the different networks is then,y i and (3) withx i Is a group of corresponding original noise-containing signals and signals after noise elimination, and is F norm ++>；
Comprehensive expansion convolutional neural network moduleNet 1 (θ 1 ) Weighing deviceNet 2 (θ 2 ) Obtaining a final loss function from the loss function of (2)：
，
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) Is a loss function of->For weighing deviceNet 2 (θ 2 ) A loss function of (2);
through the final loss function equation, network parameters are continuously adjusted, network noise elimination performance is optimized, an ideal model is obtained through training, and the root mean square error of a test set is smaller than 10 -2 。
In the step j, the model testing step specifically includes:
model test setInput to a trained novel convolutional neural network modelNet(θ) In the method, the expanded convolutional neural network module is obtainedNet 1 (θ 1 ) Prediction output of +.>，/>：
，
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) From a model test setY 2 Noise-containing signal->Preliminary noise data learned in (a);
and then the preliminary noise dataWeighing deviceNet 2 (θ 2 ) Prediction output of +.>Combining to obtain predictive noise->：
，
Finally, the original noise-containing signalPrediction noise->Removing to obtain a novel convolutional neural network modelNet(θ) Output of +.>I.e. denoised signal->：/>
，
The inversion interpretation in the k step includes: inputting the noise elimination result into multi-index inversion software to obtain transverse relaxation timeT 2 Graph related to nuclear magnetic resonance signal amplitude and observation of transverse relaxation timeT 2 The method comprises the steps of carrying out a first treatment on the surface of the Recording transverse relaxation timeT 2 Spectrum peak positions corresponding to groundwater organic pollution of different material components; recording transverse relaxation timeT 2 Corresponding to the content of organic pollution of groundwater of different material components.
Claims (6)
1. A magnetic resonance mobile detection method suitable for crude oil leakage detection in an oil field, which is characterized by comprising the following steps:
a. setting up a magnetic resonance mobile detection system in a region;
b. according to the detection direction, the field shape and the position of the transmitting coil, the speed of the mobile detecting vehicle carrying the receiving coil is controlled, so that the mobile detecting vehicle pulls the supporting framework carrying the receiving coil to run on the measuring line at a constant speed and step-by-step mobile scanning is performed in the transmitting coil;
c. setting current parameters of a control system by using an upper computer; after the setting is finished, the transmitter is controlled by a control system, and a transmitting current with excitation pulse as a CPMG sequence and frequency as local Larmor frequency is introduced into a transmitting coil from small to large to excite spin directions of hydrogen protons in groundwater and oil to generate a wrenching angle;
d. turning off the emission current, recovering the spin of hydrogen proton to the initial direction, inducing by the receiving coil and collecting nuclear magnetic resonance signal by the receiver; the position of the receiving coil on the measuring line is used as a measuring point, and the position information of the receiving coil is recorded by a GPS (global positioning system) arranged on the receiving coil; after the signal acquisition is finished, the receiver transmits data to the upper computer through the wireless receiving communication module for data storage;
e. setting the emission current to beC-d, namely repeatedly changing the current value for multiple times to generate alternating current fields with different excitation pulse moments, carrying out layered scanning on underground organic pollution distribution, and repeating the steps to obtain a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals;
f. setting the emission current to beRepeating the steps c-d to obtain a plurality of groups of actually measured pure noise data;
g. and processing the acquired multiple groups of actually measured noise-containing nuclear magnetic resonance mobile detection signals and multiple groups of actually measured pure noise data to obtain the actual situation of underground organic pollution distribution.
2. The method for magnetic resonance mobile detection of crude oil leak in oil field according to claim 1, wherein step g specifically comprises:
1) Amplitude scaling the acquired multiple groups of actually measured noise-containing nuclear magnetic resonance movement detection signals to expand the signals into data with the number ofLIs the original noise-containing signal of (a)YDividing the data into model training sets according to the ratio of the data number ratio of 4:1And model test set->WhereinpAndqrespectively represent model training setsY 1 And model test setY 2 Is used for the data amount of the data,p、qsatisfy->And->；/>Representative ofY 1 Is provided with a data set of a certain data,；/>representative ofY 2 Some data of (a), is +>The method comprises the steps of carrying out a first treatment on the surface of the Screening and model training set from multiple groups of collected actual measurement pure noise data according to the principle of corresponding sampling timeY 1 Pure noise data corresponding to the medium data to form a pure noise training set +.>Wherein->Represent the firstiNoise (S)>；
2) Constructing network parameters asθ 1 Is a convolutional neural network moduleNet 1 (θ 1 ) And network parameters areθ 2 Weight of (2)Net 2 (θ 2 ) Will beNet 1 (θ 1 ) And (3) withNet 2 (θ 2 ) Parallel connection, constructing network parameters asθNovel convolutional neural network model of (2)Net(θ)；
3) Training set of modelsY 1 The first of (3)iThe original noise-containing signalsInput to a novel convolutional neural network modelNet(θ) During training, the convolutional neural network module is expandedNet 1 (θ 1 ) Model training setY 1 Original noise-containing signal->Preliminary noise feature learning is carried out to obtain preliminary noise data +.>Weighing deviceNet 2 (θ 2 ) Training set for modelY 1 Original noise-containing signal->Corresponding weights are given to different characteristic information of the noise, wherein the preliminary noise data of the noise +.>Is given the highest level weight to make preliminary noise data +>Close to pure noise training setNPure noise data in the original noise-containing signal to obtain noise data +.>Establishing an original noise-containing signal->Noise data +.>Nonlinear mapping relation between ∈>Noise data in the original noise-containing signal is +.>From the original noisy signal->Removing the middle part and outputting the denoised signal +.>；
4) Test set of modelsY 2 The data in the model is input into a novel convolutional neural network model with complete trainingNet(θ) Testing; when the root mean square error of the novel convolutional neural network test set is less than 10 -2 During training, the novel convolutional neural network model is completedNet(θ) Otherwise, adjusting network parameters, and repeating the steps 2) -4) until reaching the standard;
5) Inputting a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals into a novel convolutional neural network model after the test reaches the standardNet(θ) The denoising operation is performed to obtain a denoising result; and carrying out inversion interpretation on the denoising result to obtain the actual situation of underground organic pollution distribution.
3. The method for magnetic resonance mobile detection for crude oil leak detection in oil field according to claim 2, wherein the expanding convolutional neural network moduleNet 1 (θ 1 ) The device comprises an input layer, an expansion operation layer and an output layer in sequence; the input layer comprises a convolution layer and a linear activation function, and a plurality of groups of actually measured noise-containing nuclear magnetic resonance movement detection signals are read; the expansion operation layer comprises a plurality of alternating series arrangement of a first sub-module layer and a second sub-module layer, wherein the first sub-module layer sequentially comprises a convolution layer, a batch normalization layer and a linear activation function, and the second sub-module layer sequentially comprises an expansion convolution layer, a batch normalization layer and a linear activation function; the output layer comprises a convolution layer and is used for reconstructing and outputting noise information; weighing apparatusNet 2 (θ 2 ) Comprises a convolution layer, a dense module, a global average pooling layer, a convolution layer,sigmoidThe activation function is connected in series; wherein the dense module comprises four sub-module layers, the first three sub-module layers have the same structure and all comprise a convolution kernel with the size ofA batch normalization layer and a linear activation function, the last submodule layer comprising a convolution kernel of size +.>A batch normalization layer and a linear activation function; in a dense module, each sub-module layer is interconnected, i.e. each sub-module layer splices inputs of all sub-module layers arranged in front of it as additional inputs and delivers the output characteristic information to all sub-module layers arranged behind it.
4. The method for detecting magnetic resonance movement suitable for crude oil leakage detection in oil fields according to claim 3, wherein the training process specifically comprises the following steps:
(1) By combining the original noise-containing signalsy i Input to an expanded convolutional neural network moduleNet 1 (θ 1 ) In (a) and (b); expanding convolutional neural network moduleNet 1 (θ 1 ) Middle (f)kReceptive field of layer convolution layerThe size is as follows: />,
Wherein the firstkThe layer convolution layer is an expansion convolution neural network moduleNet 1 (θ 1 ) Is provided with a convolution layer of any one layer,is the firstk-The receptive field size of the 1-layer convolution layer, < >>Is the firstkConvolution kernel size of layer convolution layer, +.>Is the firstmThe step size of the layer convolution layer,the method comprises the steps of carrying out a first treatment on the surface of the Adding intervals into a basic convolution kernel by the expansion convolution layer, and increasing the size of the convolution kernel;
signal warp of the firstkLayer convolution layer operation to obtain outputThe method comprises the following steps: />,
In the method, in the process of the invention,is the firstkThe input of the layer convolution layer is convolution operation,/->For linear activation function +.>Is the firstkBias of layer convolution layers;
when the first iskLayer is expansion convolution neural network moduleNet 1 (θ 1 ) At the time of the last layer of (c),to expand convolutional neural network moduleNet 1 (θ 1 ) From original noisy signalsy i Preliminary noise data learned in +.>；
Then the original noise-containing signal is processedy i Input to a balanceNet 2 (θ 2 ) In the method, the signal is subjected to preliminary feature extraction through two convolution layers:,
in the method, in the process of the invention,for the extracted features, < >>，/>Respectively, the first of the original noise-containing signals after entering the weighing apparatusOne layer convolution and a second layer convolution operation;
deep extraction is carried out on the characteristics by the signal through the dense module, and the obtained result is:,
in the method, in the process of the invention,deep extracted features for dense modules, +.>Is a dense module operation;
the signal is processed by global average pooling layer, two convolution layers andSigmoidactivating the function to obtain the weight coefficient of the noise:,
in the method, in the process of the invention,is a weight coefficient>Is thatSigmoidFunction (F)>For global average pooling operations,/->And->Performing first-layer convolution and second-layer convolution operation after global average pooling operation;
imparting noise from original signaly i Preliminary noise data learned in (1)The highest weight coefficient/>To increase the model's attention to noise, improve preliminary noise data +.>Make it more similar to pure noise training setNPure noise data in (a); weight coefficient->And preliminary noise data->Combining in a certain way to obtain noise data in the original noise-containing signal>：,
From the input raw noisy signaly i Removing noise data from original noise-containing signalObtaining the denoised signalx i As a novel convolutional neural network modelNet(θ) Is a final output of (2): />;
(2) Setting up a loss function optimization model to reduce noise data in an original noise-containing signalAnd pure noise training setNThe difference between the pure noise data in the model (a) and the loss function adopts a gradient descent algorithm: />,
In the method, in the process of the invention,，/>as a function of the losses of the different networks,ptraining set for modelY 1 Data volume of->To input original noise-containing signalsy i The output of the different networks is then,y i and (3) withx i Is a group of corresponding original noise-containing signals and signals after noise elimination, < + >>Is F norm;
from expanding convolutional neural network modulesNet 1 (θ 1 ) Weighing deviceNet 2 (θ 2 ) Obtaining a final loss function from the loss function of (2)：/>,
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) Is a loss function of->For weighing deviceNet 2 (θ 2 ) A loss function of (2);
through the final loss function, the network parameters are continuously adjusted, the network noise elimination performance is optimized, and training is carried out to obtainIdeal model, the root mean square error of the test set is less than 10 -2 。
5. The method for magnetic resonance mobile detection for crude oil leak detection in oil fields of claim 4, wherein the step of performing the test comprises:
model test setInput to a trained novel convolutional neural network modelNet(θ) In the method, the expanded convolutional neural network module is obtainedNet 1 (θ 1 ) Prediction output of +.>，/>：,
In the method, in the process of the invention,to expand convolutional neural network moduleNet 1 (θ 1 ) From a model test setY 2 Original noise-containing signal->Preliminary noise data learned in (a);
and then the preliminary noise dataWeighing deviceNet 2 (θ 2 ) Prediction output->Combining to obtain predictive noise->：
，
Finally, the original noise-containing signalPrediction noise->Removing to obtain a novel convolutional neural network modelNet(θ) Output of +.>I.e. denoised signal->：
。
6. The method for magnetic resonance mobile detection for crude oil leak detection in oil fields of claim 2, wherein the inversion interpretation comprises: inputting the noise elimination result into multi-index inversion software to obtain transverse relaxation timeT 2 Graph related to nuclear magnetic resonance signal amplitude and observation of transverse relaxation timeT 2 The method comprises the steps of carrying out a first treatment on the surface of the Recording transverse relaxation timeT 2 Obtaining the organic pollution of the groundwater of different material components; recording transverse relaxation timeT 2 The spectrum peak area of the underground water organic pollution content of different material components is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410264266.0A CN117872487B (en) | 2024-03-08 | 2024-03-08 | Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410264266.0A CN117872487B (en) | 2024-03-08 | 2024-03-08 | Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117872487A true CN117872487A (en) | 2024-04-12 |
| CN117872487B CN117872487B (en) | 2024-05-24 |
Family
ID=90594959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410264266.0A Active CN117872487B (en) | 2024-03-08 | 2024-03-08 | Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117872487B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014165050A1 (en) * | 2013-03-12 | 2014-10-09 | The General Hospital Corporation | Method for increasing signal-to-noise ratio in magnetic resonance imaging using per-voxel noise covariance regularization |
| CN105785455A (en) * | 2016-03-09 | 2016-07-20 | 吉林大学 | Two-dimensional ground nuclear magnetic resonance inversion method based on B spline interpolation |
| US20190120995A1 (en) * | 2017-10-20 | 2019-04-25 | Jilin University | Method for random noise reduction from mrs oscillating signal using joint algorithms of emd and tfpf |
| CN109765628A (en) * | 2019-01-24 | 2019-05-17 | 吉林大学 | Vehicular prepolarizing field magnetic resonance water source exploration device and detection method |
| CN109870729A (en) * | 2019-01-31 | 2019-06-11 | 吉林大学 | Deep neural network magnetic resonance signal noise-eliminating method based on discrete cosine transform |
| WO2020219915A1 (en) * | 2019-04-24 | 2020-10-29 | University Of Virginia Patent Foundation | Denoising magnetic resonance images using unsupervised deep convolutional neural networks |
| CN114091538A (en) * | 2021-11-23 | 2022-02-25 | 吉林大学 | Signal feature-based loss-discriminating convolutional neural network intelligent noise reduction method |
| CN116148935A (en) * | 2023-02-02 | 2023-05-23 | 吉林大学 | Magnetic resonance random noise suppression method based on self-adaptive self-encoder |
| CN116401513A (en) * | 2023-04-13 | 2023-07-07 | 吉林大学 | Magnetic resonance power frequency harmonic noise suppression method based on depth residual error network |
| CN116561515A (en) * | 2023-05-15 | 2023-08-08 | 吉林大学 | Power frequency noise suppression method based on cyclic neural network magnetic resonance signals |
-
2024
- 2024-03-08 CN CN202410264266.0A patent/CN117872487B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014165050A1 (en) * | 2013-03-12 | 2014-10-09 | The General Hospital Corporation | Method for increasing signal-to-noise ratio in magnetic resonance imaging using per-voxel noise covariance regularization |
| CN105785455A (en) * | 2016-03-09 | 2016-07-20 | 吉林大学 | Two-dimensional ground nuclear magnetic resonance inversion method based on B spline interpolation |
| US20190120995A1 (en) * | 2017-10-20 | 2019-04-25 | Jilin University | Method for random noise reduction from mrs oscillating signal using joint algorithms of emd and tfpf |
| CN109765628A (en) * | 2019-01-24 | 2019-05-17 | 吉林大学 | Vehicular prepolarizing field magnetic resonance water source exploration device and detection method |
| CN109870729A (en) * | 2019-01-31 | 2019-06-11 | 吉林大学 | Deep neural network magnetic resonance signal noise-eliminating method based on discrete cosine transform |
| WO2020219915A1 (en) * | 2019-04-24 | 2020-10-29 | University Of Virginia Patent Foundation | Denoising magnetic resonance images using unsupervised deep convolutional neural networks |
| CN114091538A (en) * | 2021-11-23 | 2022-02-25 | 吉林大学 | Signal feature-based loss-discriminating convolutional neural network intelligent noise reduction method |
| CN116148935A (en) * | 2023-02-02 | 2023-05-23 | 吉林大学 | Magnetic resonance random noise suppression method based on self-adaptive self-encoder |
| CN116401513A (en) * | 2023-04-13 | 2023-07-07 | 吉林大学 | Magnetic resonance power frequency harmonic noise suppression method based on depth residual error network |
| CN116561515A (en) * | 2023-05-15 | 2023-08-08 | 吉林大学 | Power frequency noise suppression method based on cyclic neural network magnetic resonance signals |
Non-Patent Citations (5)
| Title |
|---|
| 万玲;张扬;林君;蒋川东;林婷婷;: "基于能量运算的磁共振信号尖峰噪声抑制方法", 地球物理学报, no. 06, 15 June 2016 (2016-06-15) * |
| 林君;张扬;张思远;舒旭;杜文元;林婷婷;: "强电磁干扰下磁共振地下水探测噪声压制方法研究进展", 吉林大学学报(地球科学版), no. 04, 26 July 2016 (2016-07-26) * |
| 林婷婷;蒋川东;齐鑫;史文龙;段清明;林君;: "地面磁共振测深分布式探测方法与关键技术", 地球物理学报, no. 11, 15 November 2013 (2013-11-15) * |
| 陈亮;李凡;鲁恺;陈佳泓;张加刚;李振宇;: "全相位模型估计方法压制磁共振测深谐波噪声应用研究", 地质科技通报, no. 04 * |
| 陈亮;李凡;鲁恺;陈佳泓;张加刚;李振宇;: "全相位模型估计方法压制磁共振测深谐波噪声应用研究", 地质科技通报, no. 04, 25 May 2020 (2020-05-25) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117872487B (en) | 2024-05-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111639747B (en) | GNSS-R sea surface wind speed inversion method and system based on BP neural network | |
| CN109870729A (en) | Deep neural network magnetic resonance signal noise-eliminating method based on discrete cosine transform | |
| CN103353462A (en) | Rock heterogeneous quantitative evaluation method based on magnetic resonance imaging | |
| CN113568055B (en) | Aviation transient electromagnetic data inversion method based on LSTM network | |
| CN107271405A (en) | A kind of water quality parameter retrieving concentration method and apparatus of multi-model collaboration | |
| CN111898734B (en) | NMR relaxation time inversion method based on MLP | |
| CN113359212B (en) | Reservoir characteristic prediction method and model based on deep learning | |
| CN103675722B (en) | Rock T2-G experiment acquisition parameter automatic matching method | |
| CN113687433A (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
| CN115906656A (en) | Method for inverting soil water content based on meteorological and gravity satellite data and other data | |
| Jiang et al. | Magnetic resonance tomography for 3-D water-bearing structures using a loop array layout | |
| CN117872487B (en) | Magnetic resonance mobile detection method suitable for crude oil leakage detection in oil field | |
| Zhou et al. | Determination of pore size distribution in tight gas sandstones based on Bayesian regularization neural network with MICP, NMR and petrophysical logs | |
| CN106769734A (en) | A kind of focusing ultrasonic wave formula river load concentration On-line Measuring Method | |
| CN108120944B (en) | weighted iterative low-field nuclear magnetic resonance T2 spectrum inversion algorithm | |
| CN116662872A (en) | Automobile panel finishing line bearing mechanical fault diagnosis method | |
| CN110907996A (en) | Automatic dense gas reservoir identification method | |
| CN105486709A (en) | Measuring method and device for underground thick oil molecule chain length | |
| CN115356672A (en) | Multi-dimensional magnetic resonance imaging method, system and storage medium | |
| CN115329802A (en) | Nuclear magnetic resonance signal relaxation time distribution calculation method based on deep learning | |
| CN114330450A (en) | Method and system for detecting and identifying underwater vehicle by fusing multiple physical fields | |
| CN117765286B (en) | Method and device for representing water phase state of cement-based material | |
| CN112882022A (en) | Ground penetrating radar data time-frequency combination full waveform inversion method | |
| XIE et al. | A method for multiple echo trains jointing inversion of NMR relaxation measurements | |
| CN117708717B (en) | Magnetic resonance underground water detection signal high-precision extraction method based on random forest |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |