Logging downhole instrument fault diagnosis method based on DBN-RVM and related equipment

Abstract

The invention discloses a fault diagnosis method for a logging downhole instrument based on DBN-RVM and related equipment, and belongs to the field of logging downhole instruments.The method comprises the steps that a DBN model and an RVM model are constructed through sample data; establishing a classification index according to the relationship between the estimated output of the DBN model to the instrument test sample and the actual output value, performing performance evaluation on the RVM model and outputting a final DBN-RVM model, and performing fault diagnosis on the to-be-tested logging downhole instrument by using the DBN-RVM model so as to judge the running state of the downhole instrument; according to the method, the relationship between instrument responses is comprehensively considered, the relationship between data is fully mined by using the excellent feature extraction capability of deep learning, and the estimated output of the instrument responses is obtained, so that the performance of the DBN-RVM model meets the expected requirements; by adopting the method, the influence of stratum factors on fault diagnosis can be effectively eliminated, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] The invention belongs to the field of well logging downhole instrument fault diagnosis, and in particular relates to a well logging downhole instrument fault diagnosis method based on DBN-RVM and related equipment. Background Art

[0002] As an important part of logging engineering, the normal operation of logging instruments is a prerequisite for reliable logging data. With the gradual improvement of the scientific and technological level of petroleum downhole measuring instruments. Logging instruments and equipment are becoming more and more diversified, complex, intelligent, integrated and precise. In addition, the working environment of the instruments is becoming more and more harsh and changeable, which leads to various failures of logging instruments. Although the reliability of the instruments is considered in the production process of the instruments, and the failures of the instruments are checked before and after logging. However, during the logging process, the petroleum logging instruments are in an environment of electromagnetic interference and many vibration sources. In addition, the internal components of the petroleum logging instruments are highly precise, the structure is complex, and the production process is complex. The normal operation of the instruments will inevitably be greatly affected by being in such a high-pressure, high-temperature, electromagnetic interference, and unstable working environment for a long time, resulting in extremely high damage and maintenance rates of the instruments. In the actual logging process, the logging data acceptance personnel generally judge whether the instrument is operating normally based on experience, comparison between logging response curves, and data from neighboring wells. For logging responses that do not match the formation conditions, when it is impossible to judge based on experience, the instrument is often replaced for repeated measurements. These methods require high experience and professional quality of the logging personnel. It can be seen that due to the complexity of the components of the logging instrument string and the large number of influencing factors, the diagnostic method based on the traditional methods and means and the experience of experts and the personal experience of the logging operator has great limitations, and is prone to misjudgment and omission, resulting in low logging efficiency and a large waste of human, material and financial resources.

[0003] According to the previous research of this project team, a method for diagnosing well logging instrument faults was invented: a method for diagnosing well logging instrument faults based on the t-test method. This method can measure instrument faults in a timely and accurate manner. However, due to the presence of a thick (about 10 meters) gas layer in the formation, the acoustic time difference curve has a sharp "fluctuating" frequency jump phenomenon, which reduces the gray correlation between it and the compensated neutron and compensated density curves, resulting in misjudgment. It can be seen that the formation factors will affect the judgment of faults. Therefore, this method has the possibility of misjudgment, that is, there is a situation where the geological gas-bearing layer information is misjudged as an instrument fault. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned technology, the present invention provides a DBN-RVM-based downhole logging instrument fault diagnosis method and related equipment, which can solve the technical problem that the existing diagnosis method is easily affected by formation factors, resulting in misdiagnosis of faults.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A DBN-RVM-based downhole logging instrument fault diagnosis method, comprising:

[0007] Obtain sample data of the downhole logging instrument in a normal state and sample data of the downhole logging instrument in a fault state;

[0008] Based on the sample data under normal conditions, a DBN model is constructed; based on the sample data under normal conditions and the sample data under fault conditions, an RVM model is constructed;

[0009] Use the DBN model to estimate the test sample and calculate the estimated output value;

[0010] Based on the estimated output value, the actual output value corresponding to the test sample and the RVM model, the DBN-RVM model is output;

[0011] The fault data of the downhole logging instrument to be tested is input into the DBN-RVM model, and the fault diagnosis results are output.

[0012] Furthermore, the sample data under normal conditions includes a data output vector and a corresponding input vector, and a DBN model is constructed using a "multi-input-single output" approach.

[0013] Furthermore, the specific construction steps of the RVM model are as follows:

[0014] S201: Selecting a normal output sample from the sample data in a normal state and a faulty output sample from the sample data in a faulty state;

[0015] S202: Calculate the grey correlation degree and average relative error between the normal output samples and the faulty output samples respectively;

[0016] S203: Use the grey correlation degree and the average relative error obtained in S202 as classification feature indicators to obtain the RVM model.

[0017] Furthermore, the specific construction steps of the DBN-RVM model are as follows:

[0018] S401: Calculate the grey correlation degree and average relative error between the estimated output value and the actual output value corresponding to the test sample respectively;

[0019] S402: input the estimated average relative error obtained in S401 into the DBN model, and output the estimated grey correlation degree;

[0020] S403: Based on the estimated grey correlation degree and the grey correlation degree in S401, the performance of the DBN model is evaluated and the DBN-RVM model is output.

[0021] Further, in S403, according to the estimated grey correlation degree and the grey correlation degree in S401, the number of samples corresponding to the four categories of A, B, C, and D are obtained by statistics, namely a, b, c, and d respectively; the classification accuracy is calculated according to the number of samples corresponding to the four categories; and the performance of the DBN model is evaluated according to the classification accuracy;

[0022] Among them, category A is: actually normal data, marked as correct by the DBN model;

[0023] Category B: data that is actually faulty but marked as correct by the DBN model;

[0024] Category C: data that is actually normal but marked as faulty by the DBN model;

[0025] Category D: data that is actually faulty and is marked as faulty by the DBN model.

[0026] Furthermore, before constructing the DBN model, the sample data under normal conditions is normalized.

[0027] Furthermore, before the test samples are input into the DBN model, the test samples are normalized; and the calculated estimated output values ​​are denormalized.

[0028] A DBN-RVM-based downhole logging instrument fault diagnosis system is used to implement the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method, including:

[0029] The sample acquisition module is used to acquire sample data of the downhole logging instrument in a normal state and sample data in a fault state;

[0030] A model building module is used to build a DBN model based on sample data in a normal state; and to build an RVM model by combining sample data in a normal state and sample data in a fault state;

[0031] The estimated output module is used to estimate the test sample using the DBN model and calculate the estimated output value;

[0032] The performance evaluation module is used to output the DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model;

[0033] The fault diagnosis module is used to input the fault data of the downhole logging instrument to be tested into the DBN-RVM model and output the fault diagnosis result.

[0034] A device comprising:

[0035] Memory for storing computer programs;

[0036] The processor is used to implement the steps of the above-mentioned DBN-RVM-based downhole logging instrument fault diagnosis method when executing the computer program.

[0037] A computer-readable storage medium stores a computer program, which is used to implement the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method when executed by a processor.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention also provides a DBN-RVM-based downhole logging instrument fault diagnosis method, which constructs a DBN model and an RVM model respectively through sample data; a classification index is established according to the relationship between the estimated output of the instrument test sample and the actual output value of the DBN model, and the performance of the RVM model is evaluated and the final DBN-RVM model is output, and the DBN-RVM model is used to perform fault diagnosis on the downhole logging instrument to be tested, so as to judge the operating status of the downhole instrument; the method comprehensively considers the relationship between instrument responses, uses the excellent feature extraction ability of deep learning, fully mines the connection between data, obtains the estimated output of the instrument response, and makes the DBN-RVM model performance meet the expected requirements; the use of this method can effectively eliminate the influence of formation factors on fault diagnosis, and improve the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of a contrast divergence algorithm provided by an embodiment of the present invention;

[0041] Figure 2 A graph showing the convergence of the number of iterations provided by an embodiment of the present invention;

[0042] Figure 3 A diagram showing the change of the fault diagnosis model of the compensating acoustic wave instrument provided by the embodiment of the present invention as the number of network layers increases; wherein (a) is a network with 3 layers; (b) is a network with 4 layers; (c) is a network with 5 layers; (d) is a network with 6 layers;

[0043] Figure 4Comparison diagram of DBN model training results with different numbers of hidden layer nodes provided by an embodiment of the present invention; wherein (a) is a comparison between the predicted value of random number 1 and the actual value (left figure) and its relative error (right figure); (b) is a comparison between the predicted value of random number 2 and the actual value (left figure) and its relative error (right figure); (c) is a comparison between the predicted value of random number 3 and the actual value (left figure) and its relative error (right figure);

[0044] Figure 5 Results of test samples provided for embodiments of the invention (left figure) and their relative errors (right figure);

[0045] Figure 6 The classification results of the RVM model provided by the embodiment of the invention;

[0046] Figure 7 A flow chart of a DBN-RVM-based downhole logging instrument fault diagnosis method provided by the present invention;

[0047] Figure 8 A schematic structural diagram of a DBN-RVM-based downhole logging instrument fault diagnosis system provided by the present invention. DETAILED DESCRIPTION

[0048] The present invention provides a DBN-RVM-based downhole logging instrument fault diagnosis method, such as Figure 7 As shown, the following steps are included:

[0049] S1: Obtain sample data of the logging downhole instrument in normal state and sample data in fault state; wherein, the sample data in normal state includes data output vector and corresponding input vector, and the DBN model is constructed by adopting the "multi-input-single output" method.

[0050] S2: Based on the sample data under normal conditions, a DBN model is constructed; and the RVM model is constructed by combining the sample data under normal conditions and the sample data under fault conditions.

[0051] Specifically, the specific steps for building the RVM model are as follows:

[0052] S201: Selecting a normal output sample from the sample data in a normal state and a faulty output sample from the sample data in a faulty state;

[0053] S202: Calculate the grey correlation degree and average relative error between the normal output samples and the faulty output samples respectively;

[0054] S203: Use the grey correlation degree and the average relative error obtained in S202 as classification feature indicators to obtain the RVM model.

[0055] S3: Use the DBN model to estimate the test sample and calculate the estimated output value;

[0056] S4: Based on the estimated output value, the actual output value corresponding to the test sample and the RVM model, output the DBN-RVM model.

[0057] Specifically, the specific steps for building the DBN-RVM model are as follows:

[0058] S401: Calculate the grey correlation degree and average relative error between the estimated output value and the actual output value corresponding to the test sample respectively;

[0059] S402: input the estimated average relative error obtained in S401 into the DBN model, and output the estimated grey correlation degree;

[0060] S403: Based on the estimated grey correlation degree and the grey correlation degree in S401, the performance of the DBN model is evaluated and the DBN-RVM model is output.

[0061] In S403, according to the estimated grey correlation degree and the grey correlation degree in S401, the number of samples corresponding to the four categories of A, B, C, and D are obtained by statistics, namely a, b, c, and d respectively; the classification accuracy is calculated according to the number of samples corresponding to the four categories; and the performance of the DBN model is evaluated according to the classification accuracy;

[0062] Among them, category A is: actually normal data, marked as correct by the DBN model;

[0063] Category B: data that is actually faulty but marked as correct by the DBN model;

[0064] Category C: data that is actually normal but marked as faulty by the DBN model;

[0065] Category D: data that is actually faulty and is marked as faulty by the DBN model.

[0066] S5: Input the fault data of the downhole logging instrument to be tested into the DBN-RVM model and output the fault diagnosis result.

[0067] Here, before constructing the DBN model, the sample data in the normal state is normalized. At the same time, before the test sample is input into the DBN model, the test sample is also normalized; and the calculated estimated output value is denormalized.

[0068] like Figure 8As shown, the present invention also provides a DBN-RVM-based well logging downhole instrument fault diagnosis system, including: a sample acquisition module, used to obtain sample data of the well logging downhole instrument in a normal state and sample data in a fault state; a model construction module, used to construct a DBN model based on the sample data in the normal state; construct an RVM model in combination with the sample data in the normal state and the sample data in the fault state; an estimated output module, used to use the DBN model to estimate the test sample and calculate the estimated output value; a performance evaluation module, used to output the DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model; a fault diagnosis module, used to input the fault data of the well logging downhole instrument to be tested into the DBN-RVM model, and output the fault diagnosis result.

[0069] The present invention also provides a device, comprising: a memory for storing a computer program; and a processor for implementing the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method when executing the computer program.

[0070] When the processor executes the computer program, the above-mentioned steps of DBN-RVM-based downhole logging instrument fault diagnosis are implemented, for example: obtaining sample data of the downhole logging instrument in a normal state and sample data in a faulty state; constructing a DBN model based on the sample data in the normal state; constructing an RVM model in combination with the sample data in the normal state and the sample data in the faulty state; using the DBN model to estimate the test sample and calculate the estimated output value; outputting the DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model; inputting the fault data of the downhole logging instrument to be tested into the DBN-RVM model, and outputting the fault diagnosis result.

[0071] Alternatively, the processor implements the functions of each module in the above system when executing the computer program, for example: a sample acquisition module, used to obtain sample data of the logging downhole instrument in a normal state and sample data in a fault state; a model construction module, used to construct a DBN model based on the sample data in a normal state; a RVM model is constructed by combining the sample data in a normal state and the sample data in a fault state; an estimated output module, used to use the DBN model to estimate the test sample and calculate the estimated output value; a performance evaluation module, used to output a DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model; a fault diagnosis module, used to input the fault data of the logging downhole instrument to be tested into the DBN-RVM model and output the fault diagnosis result.

[0072] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing preset functions, and the instruction segments are used to describe the execution process of the computer program in the DBN-RVM-based downhole logging instrument fault diagnosis device. For example, the computer program can be divided into a sample acquisition module, a model building module, an estimated output module, a performance evaluation module and a fault diagnosis module; the specific functions of each module are as follows: a sample acquisition module, used to obtain sample data of a well logging downhole instrument in a normal state and sample data in a fault state; a model building module, used to build a DBN model based on the sample data in a normal state; a RVM model is built by combining the sample data in a normal state and the sample data in a fault state; an estimated output module, used to use the DBN model to estimate the test sample and calculate the estimated output value; a performance evaluation module, used to output a DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model; a fault diagnosis module, used to input the fault data of the well logging downhole instrument to be tested into the DBN-RVM model and output the fault diagnosis result.

[0073] The DBN-RVM-based downhole logging instrument fault diagnosis device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The DBN-RVM-based downhole logging instrument fault diagnosis device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above is an example of a DBN-RVM-based downhole logging instrument fault diagnosis device, and does not constitute a limitation on the DBN-RVM-based downhole logging instrument fault diagnosis device, and can include more components than the above, or a combination of certain components, or different components. For example, the DBN-RVM-based downhole logging instrument fault diagnosis device can also include input and output devices, network access devices, buses, etc.

[0074] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the DBN-RVM-based well logging downhole instrument fault diagnosis, and uses various interfaces and lines to connect various parts of the entire DBN-RVM-based well logging downhole instrument fault diagnosis device.

[0075] The memory can be used to store the computer program and / or module, and the processor implements various functions of the DBN-RVM-based downhole logging instrument fault diagnosis equipment by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0076] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0077] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method are implemented.

[0078] If the integrated module / unit of the DBN-RVM-based downhole logging instrument fault diagnosis system is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0079] Based on such understanding, the present invention implements all or part of the processes in the above-mentioned DBN-RVM-based downhole logging instrument fault diagnosis method, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned DBN-RVM-based downhole logging instrument fault diagnosis method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0080] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0081] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0082] As mentioned in the background technology: Although the current method can measure instrument failures in a timely and accurate manner, due to the presence of a relatively thick (about 10 meters) gas layer in the stratum, the acoustic time difference curve shows a drastic "fluctuating" frequency jump phenomenon, which reduces the gray correlation between it and the compensated neutron and compensated density curves and causes misjudgment. It can be seen that stratum factors will affect the judgment of faults. Therefore, this method has the possibility of misjudgment, that is, there is a situation where the geological gas-bearing layer information is misjudged as an instrument failure.

[0083] In order to solve the above problems, this embodiment provides a DBN-RVM-based logging downhole instrument fault diagnosis method, the DBN-model is a deep belief network prediction network model, and the RVM model is a relevance vector machine classifier; the core idea of ​​the present invention is to use the excellent feature extraction ability of deep learning to fully explore the relationship between data and obtain the estimated output of the instrument response. Then, according to the relationship between the output of the prediction model and the measured value, a classification index is established to construct a DBN-RVM model; finally, the fault state and normal operation state of the logging instrument are judged.

[0084] The specific steps are as follows:

[0085] (I) Establishing DBN estimation model

[0086] 1. Determination of DBN model parameters

[0087] For the deep confidence learning model, the network parameter setting has a great impact on the final prediction effect. The most important influencing parameters are the number of iterations, the number of hidden layers, and the number of hidden layer nodes. The following is an analysis of the above parameters to establish a suitable fault diagnosis network model.

[0088] 1.1 Iterations

[0089] When an input vector is given, the features of the input data are first mapped from the visible layer v of the RBM to its hidden layer h according to formula (1.1) to obtain the activation state of the hidden layer h. After that, the visible layer v is reconstructed according to the activation state of the hidden layer nodes and formula (1.2), and the error between the visible layer input data and the reconstructed data is calculated at the same time. The parameters between the visible layer and the hidden layer are adjusted according to the size of the error, and then the calculation is performed until the error between the two is minimized. When the reconstruction error is minimized, it is considered that the number of iterations has reached a reasonable state.

[0090] When the states of all neurons in the visible layer are given, the probability of activation of a neuron in the hidden layer is p(h k =1 / v), and its calculation formula is shown in (1.1):

[0091]

[0092] When the states of all neurons in the hidden layer are given, the probability p(v k =1 / h), and its calculation formula is shown in 1.2.

[0093]

[0094] In formulas (1.1) and (1.2), is a nonlinear S-shaped function; v i and h j Represent the state values ​​of the i-th neuron node in the visible layer and the j-th neuron node in the hidden layer respectively; w ij is the connection weight between the i-th neuron node in the visible layer and the j-th neuron node in the hidden layer; a i and b j are the biases of the i-th neuron node in the visible layer and the j-th neuron node in the hidden layer, respectively.

[0095] The reconstruction error calculation formula is shown in (1.3):

[0096]

[0097] 1.2 Number of hidden layers

[0098] The goal of the model training process is to maximize the probability that the trained RBM model conforms to the distribution of the input sample data. That is, under the condition of given input sample data, by continuously adjusting the internal parameters of the restricted Boltzmann machine, P θ(v) The value of reaches the maximum, achieving the purpose of training. From formula 1.4, we can see that if we want P θ(v) maximum, then we need to adjust a, b, w to reduce the value of the energy function so that P θ(v) maximum.

[0099]

[0100]

[0101] The method for adjusting a, b, and w is the contrast divergence method, and its algorithm flow is as shown in the attached figure. Figure 1 The calculation steps are:

[0102] (1) Starting from Gibbs sampling, the input sample data is used as the input vector of the visible layer unit to determine the visible layer state v (0) ;

[0103] (2) According to v (0) And formula (1.1) to calculate the state h of the hidden layer unit (0) ;

[0104] (3) According to h (0) The visible layer unit is reconstructed using formula (1.2) to obtain the reconstructed state v of the visible layer unit (1) ;

[0105] (4) According to step (2) and step (3), the calculation is repeated until the maximum number of iterations is reached.

[0106] (5) Use formula (1.5) to adjust the parameters of the restricted Boltzmann machine (RBM model);

[0107] (6) The parameter update is completed and the algorithm stops.

[0108] Δw ij =ε( <v i h j >0- <v i h j > k )

[0109] Δb i =ε( <h j >0- <h j > k )

[0110] Δa j =ε( <v i >0- <v i > k ) (1.5)

[0111] 1.3 Number of hidden layer nodes

[0112] There is no fixed formula to calculate, usually based on experiments and experience, the present invention refers to the hidden node calculation formula of BP neural network: the empirical formula of the traditional 3-layer neural network is shown in (1.6) and (1.7):

[0113]

[0114]

[0115] From the network structure of DBN, we know that the output of the previous layer is the input of the next layer. Therefore, the number of hidden nodes in the kth layer is the number of nodes in the k-1th layer. If we iterate the above traditional formula to the nth hidden layer, we get:

[0116]

[0117]

[0118] (II) Relevance Vector Machine (RVM)

[0119] Relevant vector machines are mostly used to deal with binary classification problems. Suppose the sample data set is (x n ,t n )(n=1,2,...,N,x∈R d ,t∈{0,1}where R d is a d-dimensional real number space, and t is the category label. When dealing with binary classification, t is the target vector, which can only be 0 or 1. The function expression of the related vector classification machine is:

[0120]

[0121] In the above formula (1.10), k is called the kernel function, w i is called weight, and w0 is called the weight corresponding to the value of weight i when it is 0. In this embodiment, the Gaussian kernel function is selected as the kernel function of the correlation vector machine, and its expression is shown in (1.11):

[0122]

[0123] In the formula, σ is a scale parameter. When σ is greater than 0 and approaches 1, it is easy to cause over-learning. For general problems such as binary classification, the value is generally taken as 0.5.

[0124] (III) DBN-RVM fault diagnosis model

[0125] Since there is a certain correlation between the response data of each downhole instrument, when establishing the depth confidence model, the response of a logging instrument to be diagnosed is used as the output target vector, and the parameters directly and indirectly related to it are used as input data. After determining the sample data, due to the different units of each sample parameter, in order to eliminate the influence of different unit dimensions of the input sample parameters, the method of normalizing the original sample data is adopted. The normalization method adopts the extreme difference method. Assume that the normalized input data is X n ={x1,x2,...,x m}, after DBN model learning, the estimated output vector is Y. Then:

[0126] Y=F(X n ) (1.12)

[0127] Where n = 1, 2, .... m, represents the input vector related to the output, F represents an estimation model, and the present invention adopts the DBN model. Since the input sample data is normalized, it is still necessary to normalize the test sample data when testing it (using the extreme difference method), and the sample data after the DBN model output is also denormalized, that is, the model output result is remapped back to the original data dimension state. The denormalization formula obtained according to the normalization formula is as follows:

[0128]

[0129] in Represents the normalized data. Finally, the classification index is extracted based on the relationship between the estimated output and the measured output. Assuming that the classification index data is learned through the RVM model to obtain the classification output Z, then:

[0130] Z=Φ(e) (1.14)

[0131] Wherein, Φ represents a classifier. In this embodiment, RVM is used to construct a classifier, and e is a feature of the data to be classified.

[0132] 4. Evaluation indicators of model performance

[0133] 4.1 Evaluation of DBN Estimation Model Performance

[0134] In the performance evaluation of DBN prediction model, the present invention uses relative error (RE) and mean relative error (MRE) to evaluate the prediction effect. Relative error can better reflect the deviation between the training result or test result of the prediction model and the actual sample value, reflecting the generalization ability of the model; while the mean relative error reflects the prediction accuracy of the prediction model as a whole.

[0135]

[0136]

[0137] In the above evaluation index formulas, y i Indicates the actual response value of the logging instrument. Represents the observation value of the DBN model.

[0138] 4.2 Performance Evaluation of Relevance Vector Machine Classifier

[0139] For the performance evaluation of the classifier, the accuracy rate is used to evaluate the performance of the classifier. After classification, the results are shown in Table 1 below.

[0140] Table 1 Possible results of classification

[0141]

[0142] The accuracy rate is the ratio of the number of samples correctly classified by the classifier to the total number of samples. Therefore, according to Table 1, its calculation method is as shown in the following formula.

[0143]

[0144] (V) Fault diagnosis method and processing steps based on DBN-RVM logging tool

[0145] 5.1 Establishing DBN Estimation Model

[0146] In order to obtain the prediction model of a downhole instrument, the data output vector of the instrument under normal operation and its related parameters are used as input vectors to construct a "multi-input-single output pair", and then the DBN model is trained with sample data to obtain the DBN prediction model of the instrument. In actual logging, the response of the instrument is estimated online through the trained DBN prediction model.

[0147] 5.2 Extraction of classification indicators

[0148] The present invention uses the gray correlation value between the estimated output of the DBN model and the measured value and the average relative error between the two as classification feature indicators. The estimated output obtained by the downhole instrument DBN model constructed by 5.1 is equivalent to the output response value when the instrument is not faulty. In addition, the output response value when the instrument fails can be obtained by simulating the instrument fault distortion data model. The present invention uses the gray correlation value between the output response value of the instrument estimation model and the measured output response value and the average relative error between the two as classification indicator features. The average relative error formula is shown in formula 1.16, and the gray correlation formula is shown in formula (1.18).

[0149] The grey relational degree formula is as follows:

[0150]

[0151] Among them, γ i Represents the grey relational data. i (k) is the correlation coefficient, and the calculation formula is as follows:

[0152]

[0153] in, Indicates the minimum difference between the two poles; represents the maximum difference between the two poles; ρ is the resolution coefficient, and its value range is (0, 1);

[0154] Δ i (k)=|x'0(k)-x' i (k)| represents the relationship between each point on the reference sequence X'0 curve and each comparison sequence X' i The absolute value difference of each point on the curve; k = 1, 2...n, n is a natural number, representing the number of correlation coefficients.

[0155] The specific steps to construct classification indicators are as follows:

[0156] 1. Select the output sample under normal operating condition of the instrument and record it as Y ZC

[0157] 2. Instrument Fault Distortion Data Model Obtain the instrument fault data and record it as Y GZ

[0158] 3. Let the output of DBN estimation model be Y DBN

[0159] 4. The sample used for testing is recorded as Y T

[0160] 5. Calculate Y according to the gray correlation formula ZC With Y GZ The grey relational degree is denoted as γ train , calculate Y ZC With Y GZ The average relative error is denoted by e train .

[0161] 6. Calculate Y DBN With Y T The grey relational degree is denoted as γ test , calculate Y DBN With Y T The average relative error is denoted by e test .

[0162] 7. With γ train and e train As the classification index, the relevant vector machine classification model is established. According to formula 1.10, γ train and e train Substituting into formula 1.10, we can get the model of the relevant vector machine (RVM model).

[0163] 8. And e test Substitute it into the relevant vector machine model to calculate the classification index feature r of the model output.

[0164] 9. Based on the calculated r and the actual γ test , the performance of the relevant vector machine model is judged according to formula 1.17.

[0165] The classification indexes of the instrument when it is operating normally and when it is faulty are significantly different, and the values ​​of the classification indexes are shown in Table 2. It should be noted here that, regardless of whether the instrument is operating normally or not, the calculated gray correlation degree is not exactly equal to 0 or 1. Therefore, the present invention sets 0.3 as an acceptable interval range, that is, the calculated gray correlation degree is close to 0 or 1 within 0.3, then its value can be simplified to 0 or 1.

[0166] Table 2 Classification characteristic index values

[0167]

[0168] 5.3 Classifier Construction

[0169] The present invention establishes a binary classifier, which is denoted as RVM1. The structure of the classifier is shown in Table 3 below.

[0170] Table 3 Construction of classifier

[0171]

[0172] The present invention will be further described below in conjunction with the embodiments and drawings:

[0173] Example

[0174] 1. Establishing DBN Estimation Model

[0175] The data in this embodiment are all from the logging surface system of Sichuan Qing Drilling and Well Logging Company. First, the input sample data is normalized using the range method. This embodiment uses the response of the acoustic logging instrument as the output of the DBN model, selects 8 logging instrument response curves with good gray correlation with the acoustic logging response as the input of the DBN model, and constructs the acoustic logging instrument DBN model. That is, the input dimension of the DBN model is 8, and the output dimension is 1.

[0176] (1) Number of iterations

[0177] When studying the number of iterations, other parameters of the network are set to be fixed. The number of nodes in the visible layer is the dimension of the input sample, 8. The number of nodes in the hidden layer can be randomly set within an appropriate range. The number of nodes in the hidden layer is set to 20, the learning rate ε is set to 0.1, and the learning momentum m is set to 0.5. 4000 sets of data are selected as sample data for training, and the number of iterations ranges from 0 to 150. The specific results are as follows. Figure 2 As shown in the result diagram, it can be clearly seen that when the number of iterations reaches nearly 100, the reconstruction error basically tends to a stable state.

[0178] (2) Number of network layers

[0179] For the fault diagnosis model of logging instruments, 4000 sets of data are selected as sample data. The number of nodes in the input layer of the DBN network model is 8, and the number of nodes in the output layer is 1. The number of nodes in the hidden layer is uniformly set to 20, the learning rate ε is set to 0.1, and the learning momentum m is set to 0.5. According to the conclusions in Section 8.1.1, the number of iterations of the network is set to 100. When the training data is input into the network model, under the condition that the number of nodes in the input layer and the number of nodes in the output layer do not change, as the number of layers of the network model increases (i.e. 3 layers: 8-20-1, 4 layers: 8-20-20-1, 5 layers: 8-20-20-20-1, 6 layers: 8-20-20-20-20-1), the comparison results between the predicted results and the actual measured values ​​are shown in the attached figure. Figure 3 As shown, in order to see the results clearly, only the results of 120 sampling points are shown in the figure.

[0180] The evaluation results corresponding to the predicted values ​​and true values ​​of the above different network layers are shown in Table 4.

[0181] Table 4 Comparison of evaluation results of DBN models with different network layers

[0182]

[0183] From the attached Figure 3 The predicted value and true value graphs and the average relative error evaluation results in Table 2 can be obtained. For details, refer to Figure 3 (a)— Figure 3 (d) The average relative error is the smallest when the number of network layers is 5 (i.e., the number of hidden layers is 3), indicating that the prediction result has a higher accuracy. As the number of network layers increases, the network training time increases. Therefore, the deep belief network of this embodiment is set to a 5-layer network structure. When the number of network layers is determined, the optimal number of nodes in the hidden layer is found to make the network model optimal.

[0184] (3) Number of hidden layer nodes

[0185] After experimental analysis and comparison, a=8 was finally selected, and the network structures were 8-12-13-13-1 and 8-8-8-8-1. The above two network structures were analyzed with three random network structures. The training results of DBN models with different numbers of hidden layer nodes are shown in Table 5. The comparison between the predicted value and the actual value and their relative error are shown in the attached figure. Figure 4 shown.

[0186] Table 5 Comparison of DBN model training time with different numbers of hidden layer nodes

[0187]

[0188] From Table 5, we can see that the fewer the total number of network nodes, the shorter the training time. Figure 4 As shown, specific reference Figure 4 (a) Figure 4 (b) and Figure 4 (c) The relative error of the training results of the network model obtained by the improved empirical formula is concentrated within 0.05. Among them, the relative error of the obtained network is smaller, within 0.035, and the prediction effect is better. The relative errors of the three groups of random network structures compared with it are all large. This shows that when the number of hidden layer nodes is close to the dimension of the input sample, the network performance is better; when the number of hidden layer nodes is less than or much larger than the dimension of the input sample, the network performance is poor. Based on the above analysis, considering the training time and model performance, the optimal structure of the deep belief network prediction model is 8-12-13-13-1.

[0189] After the training effect of the network model meets certain requirements, in order to further verify the effectiveness of the model, test samples are used to test the performance of the model. 200 groups of test samples are randomly selected and input into the above-trained network (8-12-13-13-1). The test results of the network model are as follows: Figure 5 shown.

[0190] As can be seen from the figure, the relative error between the actual output of the test sample and the predicted output is not all within 0.035, that is, the generalization ability of the network structure (8-12-13-13-1) in the test sample process is worse than that in the training process. Since the average relative error can reflect the prediction accuracy of the model as a whole, the average relative error is calculated to be 0.01696 (1.70%). Therefore, although the results of the test sample are slightly worse than the results during training, according to the calculation results of the average relative error, it is believed that the structure of the network model meets the prediction requirements.

[0191] 2. Analysis of RVM Classification Results

[0192] Using the above 4000 sets of logging data under normal instrument operation conditions, we first construct the distortion data when the acoustic instrument fails, then construct the classification index, and finally construct the relevance vector machine classifier RVM.

[0193] We use the first classification index as the horizontal axis and the second classification index as the vertical axis to plot the classification results. Figure 6 shown.

[0194] In the figure, the normal operation state of the compensation acoustic wave instrument is represented by gray solid dots, and the faulty operation state is represented by black solid dots. Figure 6 We can see that the relevance vector classifier RVM can correctly distinguish the normal and faulty operating states of the instrument.

[0195] It can be seen that when the grey correlation degree and the mean relative error are used as the classification indicators of the classifier and the kernel function is a Gaussian function, the constructed RVM binary classifier has good classification ability and can correctly distinguish the normal and faulty operating states of the logging instrument.

[0196] In summary, this embodiment provides a downhole logging instrument fault diagnosis method based on DBN-RVM, which has the following advantages over the existing diagnosis methods:

[0197] This method constructs a DBN model and a RVM model respectively through sample data; a classification index is established according to the relationship between the estimated output of the instrument test sample and the actual output value of the DBN model, and the performance of the RVM model is evaluated accordingly and the final DBN-RVM model is output. The DBN-RVM model is used to diagnose the fault of the downhole logging instrument to be tested, so as to judge the operating status of the downhole instrument; this method comprehensively considers the relationship between instrument responses, uses the excellent feature extraction ability of deep learning, fully mines the connection between data, obtains the estimated output of instrument response, and makes the performance of the DBN-RVM model meet the expected requirements; this method can effectively eliminate the influence of formation factors on fault diagnosis and improve the accuracy of diagnosis.

[0198] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.

Claims

1. A DBN-RVM-based downhole logging instrument fault diagnosis method, characterized in that: include: Obtain sample data of the downhole logging instrument in a normal state and sample data of the downhole logging instrument in a fault state; Based on the sample data under normal conditions, a DBN model is constructed; The RVM model is constructed by combining the sample data under normal conditions and the sample data under fault conditions; Use the DBN model to estimate the test sample and calculate the estimated output value; Based on the estimated output value, the actual output value corresponding to the test sample and the RVM model, the DBN-RVM model is output; The fault data of the downhole logging instrument to be tested is input into the DBN-RVM model, and the fault diagnosis results are output.

2. A DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 1, characterized in that: in, The sample data under normal conditions includes a data output vector and a corresponding input vector. The DBN model is constructed using the "multiple inputs - single output" approach.

3. The DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 1, characterized in that: The specific steps for building the RVM model are as follows: S201: Selecting a normal output sample from the sample data in a normal state and a faulty output sample from the sample data in a faulty state; S202: Calculate the grey correlation degree and average relative error between the normal output samples and the faulty output samples respectively; S203: Use the grey correlation degree and the average relative error obtained in S202 as classification feature indicators to obtain the RVM model.

4. The DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 2, characterized in that: The specific steps of building the DBN-RVM model are as follows: S401: Calculate the grey correlation degree and average relative error between the estimated output value and the actual output value corresponding to the test sample respectively; S402: input the estimated average relative error obtained in S401 into the DBN model, and output the estimated grey correlation degree; S403: Based on the estimated grey correlation degree and the grey correlation degree in S401, the performance of the DBN model is evaluated and the DBN-RVM model is output.

5. A DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 4, characterized in that: In S403, according to the estimated grey correlation degree and the grey correlation degree in S401, the number of samples corresponding to the four categories of A, B, C, and D are obtained by statistics, namely a, b, c, and d respectively; the classification accuracy is calculated according to the number of samples corresponding to the four categories; and the performance of the DBN model is evaluated according to the classification accuracy; Among them, category A is: actually normal data, marked as correct by the DBN model; Category B: data that is actually faulty but marked as correct by the DBN model; Category C: data that is actually normal but marked as faulty by the DBN model; Category D: data that is actually faulty and is marked as faulty by the DBN model.

6. The DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 1, characterized in that: Before building the DBN model, the sample data in the normal state is normalized.

7. The DBN-RVM-based downhole logging instrument fault diagnosis method according to claim 5, characterized in that: Before inputting the test samples into the DBN model, the test samples are normalized; and the calculated estimated output values ​​are denormalized.

8. A DBN-RVM-based downhole logging instrument fault diagnosis system, used to implement the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method according to any one of claims 1 to 7, characterized in that: include: The sample acquisition module is used to acquire sample data of the downhole logging instrument in a normal state and sample data in a fault state; Model building module, used to build DBN model based on sample data in normal state; The RVM model is constructed by combining the sample data under normal conditions and the sample data under fault conditions; The estimated output module is used to estimate the test sample using the DBN model and calculate the estimated output value; The performance evaluation module is used to output the DBN-RVM model based on the estimated output value, the actual output value corresponding to the test sample and the RVM model; The fault diagnosis module is used to input the fault data of the downhole logging instrument to be tested into the DBN-RVM model and output the fault diagnosis result.

9. A device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it is used to implement the steps of the DBN-RVM-based downhole logging instrument fault diagnosis method as described in any one of claims 1 to 7.

Images