A real-time early warning method for downhole stuck drill failures in long water sections of shale gas

Abstract

The present invention discloses a real-time early warning method for underground drill stuck faults applicable to shale gas long water sections, comprising: step 100, constructing a drill stuck fault database; step 200, constructing a BP neural network intelligent algorithm based on particle swarm algorithm optimization; step 300, creating a drill stuck fault early warning model based on the BP neural network intelligent algorithm in step 200. In the present invention, firstly, the learning steps of the BP neural network are systematically analyzed, and the defects of the BP neural network are analyzed, and an algorithm based on particle swarm optimization of the BP neural network is proposed; then, the algorithm based on particle swarm optimization of the BP neural network is applied to the real-time early warning of drill stuck faults, thereby establishing a real-time early warning model for underground drill stuck faults, and realizing intelligent and real-time quantitative judgment of drill stuck faults. At the same time, a data transmission interface between comprehensive logging data and drill stuck fault early warning software is compiled, and real-time data transmission between comprehensive logging data and drill stuck fault early warning software is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of deep shale gas complex formation drilling, and in particular to a downhole pipe sticking fault real-time early warning method applicable to shale gas long water sections. Background Art

[0002] In recent years, Sinopec has accelerated the exploration and development of shale gas resources and made major breakthroughs in the Sichuan and Chongqing regions. Among them, the southern Sichuan work area (Weiyuan, Changning, Zigong, etc.) is one of the areas with the richest shale gas resources and the greatest development potential in my country. The high formation pressure coefficient and complex drilling geological conditions in this work area lead to complex drilling and frequent failures. Based on the analysis of well completion data in 2018, the drilling complexity and failure time reached 5029.79h, among which the stuck drill failure was the most prominent, accounting for 47.48% of the total complexity and failure time, which seriously restricted the safe and efficient development of shale gas. Therefore, ensuring the safety of the horizontal section of shale gas horizontal wells during drilling and completion is an important and urgent task for the work area to speed up and improve efficiency.

[0003] The existing stuck drill fault identification is mainly based on expert knowledge systems, or uses conventional logging data to analyze engineering anomalies, and only a small amount of measurement and calculation work is automatically completed by computers. Most of the analysis and judgment still rely on manual work. Due to differences in individual knowledge, experience, and sense of responsibility, it often leads to failure to timely discover and deal with drilling anomalies and complex situations, increasing risks. In fact, it is not realistic to require operators to concentrate on observing changes in monitoring data and quickly identify accidents. Therefore, the existing stuck drill identification technology has prominent problems such as poor comprehensive utilization of monitoring information, insufficient risk warning, and strong subjectivity. Summary of the invention

[0004] In view of this, the present invention provides a real-time early warning method for downhole drill stuck failures suitable for long water sections of shale gas. It can establish a real-time early warning model for downhole drill stuck failures based on an artificial neural network intelligent algorithm, thereby making the drill stuck failures predictable and controllable.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A real-time early warning method for downhole drill bit sticking failure applicable to shale gas long water section, comprising:

[0007] Step 100, building a drill stuck fault database;

[0008] Step 200: construct a BP neural network intelligent algorithm based on particle swarm optimization;

[0009] Step 300: Create a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200:

[0010] Step 310: Design BP neural network:

[0011] Step 311, design the input layer:

[0012] According to the drill stuck fault database in step 100, select a characterization parameter that is highly correlated with the drill stuck fault and set it as an input neuron;

[0013] Step 312, design the output layer:

[0014] Set two output neurons, which are the expected output vector q1 = (1, 0) when a drill stuck fault occurs and the expected output vector q2 = (0, 1) when a drill stuck fault does not occur;

[0015] Step 313, design the hidden layer:

[0016] The number of hidden layers is calculated by the classical formula, and the number of hidden layers ranges from 1 to 10;

[0017] Step 320: Design a particle swarm algorithm:

[0018] Step 321, calculating the connection weight and threshold of the BP neural network structure obtained in step 310, and then determining the particle length of the particle swarm algorithm according to the connection weight and threshold, and performing vector encoding on each particle;

[0019] Step 322: Create a fitness function:

[0020] The root mean square error of BP neural network training is used as the fitness function:

[0021]

[0022] Preferably, in step 311, based on the drill stuck fault database in step 100, characterization parameters with strong correlation with the drill stuck fault are selected and set as input neurons, and the characterization parameters include: hook load, vertical pressure, torque, mechanical drilling speed, rotary table speed and drilling pressure;

[0023] The number of hidden layers is calculated by the classical formula in step 313, and the number of hidden layers ranges from 1 to 10. Specifically, the number of hidden layers is set to 10;

[0024] The step 321 calculates the connection weight and threshold value according to the BP neural network structure obtained in step 310, determines the particle length of the particle swarm algorithm according to the connection weight and threshold value, and performs vector encoding on each particle, including:

[0025] According to the above, the BP neural network structure is 6-10-2, and the calculation shows that there are 6×10+10×2=80 weights, 10+2=12 thresholds, so the number of particle swarm optimization parameters is 92, and the length of the particle is 92;

[0026] Then particle i is encoded as:

[0027]

[0028] Preferably, the step 300 of creating a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200 further includes:

[0029] Step 330: pre-process the data of the drill stuck fault database in step 100:

[0030] The data in the stuck drill fault database in step 100 are normalized.

[0031] Preferably, the normalization of the data in the stuck drill fault database in step 100 includes:

[0032] The maximum-minimum method is used to normalize the data of the drill stuck fault database in step 100:

[0033]

[0034] In the formula, x min 、x max are the minimum and maximum values ​​in the sequence respectively.

[0035] Preferably, it also includes:

[0036] Step 400: training and tuning the drill stuck fault warning model in step 300.

[0037] Preferably, the training and tuning of the drill stuck fault warning model in step 300 in step 400 includes:

[0038] Using the drill stuck fault warning model constructed in step 300, combined with specific shale gas drilling field practices, comprehensive logging parameters of drilled wells with drill stuck faults are selected as training samples, and the established model is trained and model parameters are optimized.

[0039] Preferably, the step 200 of constructing a BP neural network intelligent algorithm based on particle swarm optimization includes:

[0040] Step 210, initialization parameters:

[0041] Determine the BP neural network topology, initialize its connection weights and thresholds; determine the particle swarm dimension D according to the number of connection weights and thresholds; set the particle swarm population size M and the number of BP neural network iterations N; perform real number encoding on the connection weights and thresholds to obtain the initial population of the particle swarm; and within the allowable range [s min ,smax ]、[z min ,z max ]Set the particle speed s i,d and the particle position z i,d ; Set learning factors c1, c2;

[0042] Step 220, taking the root mean square error obtained in each iteration of the BP neural network as the fitness function of the particle;

[0043] Step 230, according to the learning step of the particle swarm algorithm, solving the global optimal position of the particles;

[0044] Step 240: Check whether the iteration termination condition is met. If so, stop, output the optimal particle, and de-decode to obtain the optimal weight and threshold;

[0045] Step 250: Perform training and prediction according to the BP neural network.

[0046] Preferably, the learning step of the particle swarm algorithm in step 230 includes:

[0047] Step 231, initialize the particle swarm parameters, wherein the parameters include: population size M, particle swarm dimension D, iteration number N, particle speed s i,d and the particle position z i,d , learning factors c1, c2;

[0048] Step 232: Select a fitness function, calculate the fitness values ​​of all particles, and set the Pbest of the initial particles in the population. i,d , Gbest;

[0049] Step 233, updating the velocity and position of the particle;

[0050] Step 234, calculate the fitness value of the updated particle and update f(Pbest i,d ); otherwise Pbest i,d No update;

[0051] Step 235, check the end condition, if the requirement is met, the optimization ends; otherwise, go to step 233 to continue.

[0052] Preferably, it also includes:

[0053] Step 500: Establish a transmission interface between the comprehensive logging data and the stuck drill fault warning model.

[0054] Preferably, building a stuck drill fault database in step 100 includes:

[0055] Combined with on-site drilling practice and based on expert knowledge, the judgment methods corresponding to different types of drill sticking are sorted out and summarized; through the judgment method, the comprehensive logging parameters that can characterize the drill sticking failure are confirmed, and the comprehensive logging parameters are used as the characterization parameters of the drill sticking failure database.

[0056] It can be seen from the above technical solutions that the real-time early warning method for drill stuck fault in downhole wells in long water sections of shale gas provided by the present invention has the following beneficial effects: compared with the traditional drill stuck fault prediction method which is overly dependent on the subjective judgment of experts, the prediction results are mostly qualitative or semi-quantitative. The present invention proposes a real-time early warning method for drill stuck fault based on particle swarm algorithm to optimize BP neural network, and creates a drill stuck fault early warning model, which realizes intelligent and real-time quantitative judgment of drill stuck fault, and solves the problems of poor comprehensive utilization of monitoring information, insufficient risk warning, and strong subjectivity of traditional prediction methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A flowchart of the steps of a downhole drill stuck fault real-time early warning method provided by an embodiment of the present invention;

[0059] Figure 2 A flowchart of the steps of the BP neural network algorithm provided by an embodiment of the present invention;

[0060] Figure 3 A flow chart of the steps of a particle swarm algorithm provided by an embodiment of the present invention;

[0061] Figure 4 A flow chart of the steps of the particle swarm optimization BP neural network algorithm provided by an embodiment of the present invention;

[0062] Figure 5 A structural diagram of a BP neural network for early warning of a stuck drill fault provided by an embodiment of the present invention;

[0063] Figure 6 A BP neural network diagram for optimizing a stuck drill fault warning based on a particle swarm algorithm provided in an embodiment of the present invention;

[0064] Figure 7 The WITS transmission end software interface provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention discloses a real-time early warning method for underground drill stuck failures applicable to long water sections of shale gas. Artificial neural network is a relatively popular artificial intelligence theory at present. Its most prominent feature is its real-time learning ability, which overcomes the defects of traditional artificial intelligence based on logical symbols in processing unstructured information. These characteristics and advantages of artificial neural networks just meet the requirements of prediction and judgment of drilling engineering anomalies and risks. The present invention establishes a real-time early warning model for underground drill stuck failures based on artificial neural networks and other intelligent algorithms, and realizes the predictability and controllability of drill stuck failures. The research and application of the technology of the present invention are of great significance to improving the profitability of single wells in deep shale gas drilling in the southern Sichuan work area, reducing complexity and failure losses, and improving the competitiveness of the team in the work area.

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The embodiment of the present invention provides a real-time early warning method for downhole drill bit stuck faults applicable to shale gas long water sections, such as Figure 1 As shown, including:

[0068] Step 100, building a drill stuck fault database;

[0069] Step 200, construct a BP neural network intelligent algorithm based on particle swarm algorithm optimization; it should be noted that the BP neural network has its own defects, and it is difficult to obtain the optimal values ​​of the connection weights and thresholds, which ultimately causes a large error between the BP neural network prediction results and the actual values, affecting the accuracy of the results. Therefore, it is necessary to combine another optimal solution algorithm with the BP neural network to improve the accuracy and precision of the prediction. Preferably, the present invention uses the particle swarm algorithm's ability to search for the optimal solution globally in conjunction with the BP neural network, so as to ultimately improve the accuracy of the BP neural network prediction;

[0070] Step 300: Create a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200:

[0071] Step 310: Design BP neural network:

[0072] Step 311, design the input layer:

[0073] According to the drill stuck fault database in step 100, select a characterization parameter that is highly correlated with the drill stuck fault and set it as an input neuron;

[0074] Step 312, design the output layer:

[0075] Two output neurons are set, which are the expected output vector q1=(1,0) when a drill stuck fault occurs and the expected output vector q2=(0,1) when a drill stuck fault does not occur. It should be noted that since the present invention mainly solves the identification of a drill stuck fault in a well, only two judgments need to be made, namely, whether a drill stuck fault occurs, that is, whether a drill stuck fault occurs or not; at the same time, if the probability of occurrence is high, therefore, only two output neurons are set when designing the output layer.

[0076] Step 313, design the hidden layer:

[0077] The number of hidden layers is calculated by the classical formula, and the number of hidden layers ranges from 1 to 10;

[0078] Step 320: Design a particle swarm algorithm:

[0079] Step 321, calculating the connection weight and threshold of the BP neural network structure obtained in step 310, and then determining the particle length of the particle swarm algorithm according to the connection weight and threshold, and performing vector encoding on each particle;

[0080] Step 322: Create a fitness function:

[0081] The root mean square error of BP neural network training is used as the fitness function:

[0082]

[0083] It can be seen from the above technical solutions that the real-time early warning method for drill stuck faults in downhole wells suitable for long water sections of shale gas provided by the embodiment of the present invention has the following beneficial effects: Compared with the traditional drill stuck fault prediction method, which relies too much on the subjective judgment of experts, the prediction results are mostly qualitative or semi-quantitative. The present invention proposes a real-time early warning method for drill stuck faults based on BP neural network, and creates a drill stuck fault early warning model, which realizes intelligent and real-time quantitative judgment of drill stuck faults, and solves the problems of poor comprehensive utilization of monitoring information, insufficient risk warning, and strong subjectivity of traditional prediction methods. Furthermore, the present invention also systematically organizes and analyzes the learning steps of BP neural network, analyzes the defects of BP neural network, and proposes an algorithm based on particle swarm optimization of BP neural network.

[0084] In this solution, in step 311, according to the drill stuck fault database in step 100, characterization parameters with strong correlation with the drill stuck fault are selected as input neurons, and the characterization parameters include: hook load, vertical pressure, torque, mechanical drilling speed, rotary table speed and drilling pressure; that is, 6 input units are set when designing the input layer;

[0085] In step 313, the number of hidden layers is calculated by the classical formula, and the number of hidden layers is 1 to 10. Specifically, the number of hidden layers is set to 10. It should be noted that in this solution, the number of hidden layers is preferably 10 through research, and the result obtained is most consistent with the actual situation; that is, the BP neural network structure is finally established as 6-10-2, such as Figure 5 As shown;

[0086] The step 321 calculates the connection weight and threshold value according to the BP neural network structure obtained in step 310, determines the particle length of the particle swarm algorithm according to the connection weight and threshold value, and performs vector encoding on each particle, including:

[0087] According to the above, the BP neural network structure is 6-10-2, such as Figure 6 As shown, the calculation results show that there are 6×10+10×2=80 weights and 10+2=12 thresholds, so the number of particle swarm optimization parameters is 92 and the length of the particle is 92;

[0088] Then particle i is encoded as:

[0089]

[0090] Specifically, the step 300 of creating a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200 also includes:

[0091] Step 330: pre-process the data of the drill stuck fault database in step 100:

[0092] The data of the stuck pipe fault database in step 100 are normalized. By normalizing the data of the stuck pipe fault database (i.e., the logging data), the problem of increased error in the BP neural network prediction result caused by the large difference in the order of magnitude of the input and output data can be avoided, thereby ensuring the accuracy of the real-time warning based on the BP neural network.

[0093] In order to further optimize the above technical solution, the normalization process of the data in the stuck drill fault database in step 100 includes:

[0094] The maximum-minimum method is used to normalize the data of the drill stuck fault database in step 100:

[0095]

[0096] In the formula, x min 、x max are the minimum and maximum values ​​in the sequence respectively. The maximum-minimum method is used to normalize the data in order to obtain a good data preprocessing effect.

[0097] Specifically, Figure 4 As shown, the construction of the BP neural network intelligent algorithm based on particle swarm optimization in step 200 includes:

[0098] Step 210, initialization parameters:

[0099] Determine the BP neural network topology, initialize its connection weights and thresholds; determine the particle swarm dimension D according to the number of connection weights and thresholds; set the particle swarm population size M and the number of BP neural network iterations N; perform real number encoding on the connection weights and thresholds to obtain the initial population of the particle swarm; and within the allowable range [s min ,s max ]、[z min ,z max ]Set the particle speed s i,d and the particle position z i,d ; Set learning factors c1, c2;

[0100] Step 220, taking the root mean square error obtained in each iteration of the BP neural network as the fitness function of the particle;

[0101] Step 230, according to the learning steps of the particle swarm algorithm, solve the global optimal position of the particle; it should be noted that the global optimal position of the particle is mainly solved according to steps 232 to 234 in the learning steps of the particle swarm algorithm;

[0102] Step 240: Check whether the iteration termination condition is met. If so, stop, output the optimal particle, and de-decode to obtain the optimal weight and threshold;

[0103] Step 250: Perform training and prediction according to the BP neural network.

[0104] Regarding the principle and model of the particle swarm algorithm, it is necessary to explain that: the search for the optimal solution is set in a D-dimensional space, and M particles are initialized to form a "bird flock" T = {Z1, Z2, ... Z M}, i = 1, 2, ... M. Among them, z i =(z i1 ,z i2 ,...z iD ), i=1,2,...M represents the position vector of the ith particle in the D-dimensional space, s i =(s i1 ,s i2 ,...s iD ), i=1,2,...M represents the velocity vector of the ith particle in the D-dimensional space; the particle’s own flight motion experience is Pbest i =(Pbesti1 ,Pbest i2 ,...,Pbest iD ), the global optimal position is Gbest=(Gbest1,Gbest2,...Gbest D ). The recursive formula of the particle algorithm is:

[0105]

[0106] Individual cognition It reflects the particle's ability to remember its optimal position; at the same time, group cognition It reflects the information sharing between particles in the population.

[0107] Specifically, Figure 3 As shown, the learning steps of the particle swarm algorithm in step 230 include:

[0108] Step 231, initialize the particle swarm parameters, wherein the parameters include: population size M, particle swarm dimension D, iteration number N, particle speed s i,d and the particle position z i,d , learning factors c1, c2;

[0109] Step 232: Select a fitness function, calculate the fitness values ​​of all particles, and set the Pbest of the initial particles in the population. i,d , Gbest;

[0110] Step 233, updating the velocity and position of the particle;

[0111] Step 234, calculate the fitness value of the updated particle and update f(Pbest i,d ); otherwise Pbest i,d No update;

[0112] Step 235, check the end condition, if the requirement is met, the optimization ends; otherwise, go to step 233 to continue.

[0113] Accordingly, for the BP neural network intelligent algorithm, it is necessary to further explain:

[0114] BP neural network is a typical multi-layer feedforward network. The main advantage of BP neural network is its strong nonlinear mapping ability. BP neural network is a three-layer feedforward artificial neural network consisting of input layer, hidden layer and output layer. According to the calculation formula of neuron output signal, the output of each neuron in the hidden layer can be obtained as:

[0115]

[0116] Where: v ijis the connection weight between input layer neuron i and hidden layer neuron j; θ j is the threshold of the hidden layer neuron j; f() is the activation function of the neuron.

[0117] Similarly, the output signal of the output layer of the BP neural network is:

[0118]

[0119] Where: μ jk is the connection weight between hidden layer neuron j and output layer neuron k; β k is the threshold of neuron k in the output layer; f() is the activation function of the neuron.

[0120] like Figure 2 As shown, the BP neural network intelligent algorithm also has the following learning steps:

[0121] (1) Initialize the parameters of the neural network:

[0122] The connection weight v of each layer of neurons ij , μ jk Assign a random number between [-1,1]; set the learning rate η to a small number between 0 and 1; set the error E to 0; and the error threshold E min Set to a positive decimal; the number of hidden layer nodes m is obtained according to the empirical formula; the sample pattern counter p and the number of training times q are reset to 1; the number of training times is preset to M;

[0123] (2) Input training samples and calculate the output of each layer:

[0124] Select training samples (X p ,Q p ) assigns values ​​to the input parameters and uses the calculation formula of the neural network output signal to calculate the output O, Y;

[0125] (3) Calculate the output error of the neural network:

[0126] Assume that the total number of training samples is P, and the error obtained in each training is Total output error

[0127] (4) Calculate the error signal of each layer:

[0128] The error signal calculation formula of the output layer and the hidden layer is:

[0129]

[0130] The error signal calculation formula between the hidden layer and the input layer is:

[0131]

[0132] (5) Adjust the connection weights and thresholds of each layer:

[0133] The calculation formulas for the connection weights and thresholds of the output layer and the hidden layer are as follows:

[0134] Δμ jk =η(q k -y k )y k (1-y k ) j (9)

[0135] Δβ k =η(q k -y k )y k (1-y k ) (10)

[0136] The calculation formulas for the connection weights and thresholds of the hidden layer and the input layer are as follows:

[0137]

[0138]

[0139] (6) Determine whether all samples have been trained once:

[0140] If the sample pattern counter p < the total number of training samples P, the sample pattern counter p and the number of training times q are both increased by 1, and then the process returns to step (2); otherwise, the process goes to step (7);

[0141] (7) Check whether the total error of the neural network meets the error threshold:

[0142] If the total output error E RME <The error threshold E min , or the training number q> the preset training number M, then it ends; otherwise, the error E is reset to 0, the sample mode calculator p is set to 1, and then returns to step (2).

[0143] In order to further optimize the above technical solution, the downhole drill stuck fault real-time early warning method applicable to the shale gas long water section provided by the embodiment of the present invention also includes:

[0144] Step 500: Establish a transmission interface between the comprehensive logging data and the drill stuck fault warning model to facilitate the real-time dynamic warning of drill stuck faults. Accordingly, the computer system in the well transmits various field data to the computer system outside the well through the WITS communication format (Wellsite Information Transfer Standard). Specifically, the transmission of the comprehensive logging data from the field to the drill stuck fault warning software (drill stuck fault warning model) for data analysis must go through the following five steps:

[0145] Step 510: The logging company obtains real-time data on site through sensors on the tool;

[0146] Step 520: The collected data is transmitted to the ground via WITS for storage or other purposes;

[0147] Step 530, using WITS data transmission software, the data is transmitted to the software in real time in the form of wits files or TCP / IP protocols;

[0148] Step 540: Save the real-time transmitted data to a local database for other purposes;

[0149] Step 550: The drill stuck fault software directly reads the data in the local database for dynamic early warning of the drill stuck fault.

[0150] Furthermore, relevant interface software was compiled, and the interface is as follows Figure 7 As shown, real-time data transmission between comprehensive logging data and stuck drill fault warning software is realized.

[0151] According to the data to be transmitted, find the corresponding parameters in the WITS record: hook load, vertical pressure, torque, mechanical drilling speed, rotary table speed, drilling pressure, as shown in Table 1:

[0152]

[0153] Table 1

[0154] In order to further optimize the above technical solution, the construction of the stuck drill fault database in step 100 includes:

[0155] Combined with on-site drilling practice and based on expert knowledge, the judgment methods corresponding to different types of drill sticking are sorted out and summarized; through the judgment method, the comprehensive logging parameters that can characterize the drill sticking failure are confirmed, and the comprehensive logging parameters are used as the characterization parameters of the drill sticking failure database.

[0156] It should be noted that drill sticking can be divided into adhesion, collapse, sand bridge, diameter reduction, keyway, mud packing, and drill stuck by falling objects according to different causes. Since there are many types of drill stuck, the type of drill stuck should be first identified through various phenomena and various information obtained during the drilling process, and then corresponding solutions can be proposed for different types of drill stuck.

[0157] Analysis of the causes of drill sticking:

[0158] Pipe sticking is also called differential pressure sticking. There are two main reasons for the formation of pipe sticking: the presence of filter cake on the well wall and the pressure difference between the bottom hole pressure and the formation pressure. Pipe sticking occurs under static drilling conditions and is related to the drilling fluid properties, formation characteristics and the shape of the wellbore. When the drilling fluid density is too high, the viscosity is too high, and the downhole tool is stationary in the wellbore for too long, pipe sticking is very likely to occur.

[0159] Drilling collapse and stuck drill is a phenomenon of drilling stuck drill caused by instability of wellbore wall. The main reasons for drilling collapse and stuck drill include: geological reasons such as complex structure and special rock physical properties that lead to easy instability of wellbore wall; drilling process reasons such as the pressure of drilling fluid column cannot balance the collapse pressure of bottom well formation, which makes wellbore wall unstable and leads to collapse risk.

[0160] The properties of drill bit stuck by sand settling and drill bit stuck by reduced diameter are similar to drill bit stuck by collapse. The causes of the occurrence are also due to the formation properties and rheological properties of drilling fluid. When the drilling fluid viscosity is low and the shear force is small, the ability to carry the cuttings in the annulus is reduced. When the pump is stopped, the cuttings in the annulus gather or sink, and then bury the drill bit or drilling tools, resulting in drill bit stuck by sand settling. Drill bit stuck by reduced diameter often occurs in expansive formations with large porosity and good permeability, such as salt-gypsum layers; generally speaking, drill bit stuck by reduced diameter encounters resistance when being lifted up, but is normal when being lowered; severe drill bit stuck by reduced diameter will lead to the burial of drilling tools.

[0161] Keyway sticking occurs when drilling into hard formations and at wellbore locations where the azimuth or inclination angle changes significantly. Generally, keyway sticking occurs when drilling is started.

[0162] Mud packing and drilling sticking is a phenomenon in which filter cake and mud debris adhere to the centralizer or around the drill bit, causing it to get stuck when it is lifted or lowered. Mud packing and drilling sticking will produce a swabbing effect, which can seriously cause subsequent accidents such as blowouts.

[0163] Drill bit sticking occurs when the drilling operator accidentally drops the tool at the wellhead into the well, causing the drill bit to get stuck.

[0164] Determine the type of stuck drill based on expert knowledge:

[0165] Different types of stuck pipe correspond to different judgment methods. Combined with on-site drilling practice and based on expert knowledge, the methods for judging stuck pipe are summarized, as shown in Table 2.

[0166]

[0167] Table 2

[0168] Determine the logging characteristic parameters related to stuck pipe:

[0169] Through the expert knowledge judgment of the stuck drill fault in the previous step, it is determined that the comprehensive logging parameters that can characterize the stuck drill fault are: hook load, vertical pressure, torque, mechanical drilling speed, rotary table speed, and drilling pressure, as shown in Table 3.

[0170]

[0171]

[0172] Table 3

[0173] The real-time early warning method for downhole stuck drill failure applicable to shale gas long water section provided by the embodiment of the present invention also includes:

[0174] Step 400: training and tuning the pipe stuck fault warning model in step 300. The pipe stuck fault warning model is debugged to achieve a good initialization effect, so as to better serve the real-time warning of pipe stuck fault in the well.

[0175] In this solution, the training and tuning of the drill stuck fault warning model in step 300 in step 400 includes:

[0176] Using the drill stuck fault warning model constructed in step 300, combined with specific shale gas drilling field practices, comprehensive logging parameters of drilled wells with drill stuck faults are selected as training samples, and the established model is trained and model parameters are optimized.

[0177] Step 410: Set the control elements of the algorithm:

[0178] The control elements of the BP neural network and particle swarm algorithm in the drill stuck fault warning model are set as shown in Table 4:

[0179]

[0180] Table 4

[0181] Step 420, training and simulating the model:

[0182] Combined with the logging and well history data of 6 wells with stuck pipe failure in the southern Sichuan shale gas area (XX-1, XX-2, XX-3, XX-4, XX-5, and XX-6), 12 groups of working condition data were selected as samples, 6 groups of data corresponding to stuck pipe failures, and 6 groups of data corresponding to unstuck pipe conditions; the comprehensive logging parameters monitored in the domain before and after the stuck pipe failure were selected, and normalized according to the maximum and minimum method, and the results are shown in Table 5. The first 10 groups of samples were selected as training samples to train the constructed stuck pipe failure model, and the other two groups of samples were used as verification samples to verify the accuracy of the constructed model.

[0183]

[0184] Table 5 (BP neural network training simulation samples)

[0185] The training results of the stuck pipe fault model are shown in Table 6. Result analysis: The stuck pipe fault diagnosis results obtained through simulation are consistent with the actual downhole status, indicating that the model established through training can be applied to the early warning and identification of stuck pipe faults in this block and this layer.

[0186]

[0187]

[0188] Table 6 (Simulation results of neural network training for drill stuck fault)

[0189] In summary, the real-time early warning method for drill stuck fault in downhole wells suitable for long water sections of shale gas provided by the present invention first systematically analyzes the learning steps of the BP neural network, analyzes the defects of the BP neural network, and proposes an algorithm based on particle swarm optimization of the BP neural network; then, the algorithm based on the particle swarm optimization of the BP neural network is applied to the real-time early warning of drill stuck fault, and the intelligent and real-time quantitative judgment of the drill stuck fault is realized; at the same time, a data transmission interface between the comprehensive logging data and the drill stuck fault early warning software is compiled, and the real-time data transmission between the comprehensive logging data and the drill stuck fault early warning software is realized.

[0190] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0191] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A real-time early warning method for downhole drill bit failure in a long water section of shale gas, characterized in that: include: Step 100, building a drill stuck fault database; Step 200, constructing a BP neural network intelligent algorithm, including: constructing a BP neural network intelligent algorithm based on particle swarm algorithm optimization; Step 300: Create a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200: Step 310: Design BP neural network: Step 311, design the input layer: According to the drill stuck fault database in step 100, select a characterization parameter that is highly correlated with the drill stuck fault and set it as an input neuron; Step 312, design the output layer: Set two output neurons, which are the expected output vector q1 = (1, 0) when a drill stuck fault occurs and the expected output vector q2 = (0, 1) when a drill stuck fault does not occur; Step 313, design the hidden layer: The number of hidden layers is calculated by the classical formula, and the number of hidden layers ranges from 1 to 10; Step 320, design a particle swarm algorithm: Step 321, calculate the connection weight and threshold value of the BP neural network structure obtained in step 310, determine the particle length of the particle swarm algorithm according to the connection weight and threshold value, and perform vector encoding on each particle, including: According to the above, the BP neural network structure is 6-10-2, and the calculation shows that there are 6×10+10×2=80 weights, 10+2=12 thresholds, so the number of particle swarm optimization parameters is 92, and the length of the particle is 92; Then particle i is encoded as: Step 322: Create a fitness function: The root mean square error of BP neural network training is used as the fitness function:

2. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 1 is characterized in that: In step 311, according to the drill stuck fault database in step 100, characterization parameters with strong correlation with the drill stuck fault are selected and set as input neurons, wherein the characterization parameters include: hook load, vertical pressure, torque, mechanical drilling speed, rotary table speed and drilling pressure; In the step 313 , the number of hidden layers is calculated by the classical formula, and the number of hidden layers ranges from 1 to 10. Specifically, the number of hidden layers is set to 10.

3. The real-time early warning method for downhole stuck drill failure in a long water section of shale gas according to claim 1 is characterized in that: The step 200 of constructing a BP neural network intelligent algorithm based on particle swarm optimization includes: Step 210, initialization parameters: Determine the BP neural network topology, initialize its connection weights and thresholds; determine the particle swarm dimension D according to the number of connection weights and thresholds; set the particle swarm population size M and the number of BP neural network iterations N; perform real number encoding on the connection weights and thresholds to obtain the initial population of the particle swarm; and within the allowable range [s min ,s max ]、[z min ,z max ]Set the particle speed s i,d and the particle position z i,d ; Set learning factors c1, c2; Step 220, taking the root mean square error obtained in each iteration of the BP neural network as the fitness function of the particle; Step 230, according to the learning step of the particle swarm algorithm, solving the global optimal position of the particles; Step 240: Check whether the iteration termination condition is met. If so, stop, output the optimal particle, and de-decode to obtain the optimal weight and threshold; Step 250: Perform training and prediction according to the BP neural network.

4. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 3 is characterized in that: The learning steps of the particle swarm algorithm in step 230 include: Step 231, initialize the particle swarm parameters, wherein the parameters include: population size M, particle swarm dimension D, iteration number N, particle speed s i,d , particle position z i,d , learning factors c1, c2; Step 232: Select a fitness function, calculate the fitness values ​​of all particles, and set the Pbest of the initial particles in the population. i,d , Gbest; Step 233, updating the velocity and position of the particle; Step 234, calculate the fitness value of the updated particle and update f(Pbest i,d ); otherwise Pbest i,d No update; Step 235, check the end condition, if the requirement is met, the optimization ends; otherwise, go to step 233 to continue.

5. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 1 is characterized in that: The step 300 of creating a drill stuck fault warning model based on the BP neural network intelligent algorithm in step 200 also includes: Step 330: pre-process the data of the drill stuck fault database in step 100: The maximum-minimum method is used to normalize the data of the drill stuck fault database in step 100: In the formula, x min 、x max are the minimum and maximum values ​​in the sequence respectively.

6. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 1 is characterized in that: Also includes: Step 400: training and tuning the drill stuck fault warning model in step 300.

7. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 6 is characterized in that: The training and tuning of the drill stuck fault warning model in step 300 in step 400 includes: Using the drill stuck fault warning model constructed in step 300, combined with specific shale gas drilling field practices, comprehensive logging parameters of drilled wells with drill stuck faults are selected as training samples, and the established model is trained and model parameters are optimized.

8. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 1 is characterized in that: Also includes: Step 500: Establish a transmission interface between the comprehensive logging data and the stuck drill fault warning model.

9. The downhole drill bit stuck fault real-time early warning method applicable to shale gas long water section according to claim 1 is characterized in that: The step 100 of constructing a stuck drill fault database includes: Combined with on-site drilling practice and based on expert knowledge, the judgment methods corresponding to different types of drill sticking are sorted out and summarized; through the judgment method, the comprehensive logging parameters that can characterize the drill sticking failure are confirmed, and the comprehensive logging parameters are used as the characterization parameters of the drill sticking failure database.

Images