A Smart Control Method for Fine Injection and Production Between Fractured Sections in Multi-Stage Fracturing Horizontal Wells

Abstract

This invention discloses a precise intelligent control method for injection and production between fractured sections in multi-stage fracturing horizontal wells, relating to the field of oil and gas field development technology. Based on the particle swarm optimization method, this invention optimizes the combination of communication parameters between the wellhead system and the downhole system, determining the globally optimal combination of communication parameters to enhance the data signal transmission performance between the two systems. Then, based on the globally optimal communication parameters, the downhole system is controlled to intelligently control the injection and production sections of the multi-stage fracturing horizontal well. During the injection and production process, real-time downhole raw measurement data is collected to establish a pressure database. A pressure early warning model is established and trained and validated using the pressure database before being embedded in the wellhead system. The collected data from the downhole system is input into the pressure early warning model in real time for pressure trend extrapolation. Based on the pressure early warning trigger mechanism, well site safety control measures are immediately executed when overpressure occurs, achieving precise control and overpressure early warning of the injection and production process in multi-stage fracturing horizontal wells.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, specifically to a method for precise injection and production control between fractured sections in multi-stage fracturing horizontal wells. Background Technology

[0002] In oil and gas field development, inter-fracture injection-production control is a key means to improve reservoir utilization and development effectiveness, often used to optimize fluid distribution and pressure maintenance within the fracture network. During multi-stage fracturing horizontal well injection-production, different injection-production events have varying impacts on the reservoir. For example, uneven injection pressure can lead to reservoir damage, unreasonable flow distribution between fracture segments can affect sweep efficiency, and abnormal pressure fluctuations can induce engineering accidents. Therefore, timely control and rapid feedback of different injection-production events are crucial for the efficient development of oil and gas fields.

[0003] Currently, conventional inter-fracture injection-production control methods mainly rely on manual experience to set fixed parameters, which is time-consuming, inefficient, and easily affected by human factors, lacking intelligence and adaptive capabilities. Therefore, there is an urgent need to propose a precise intelligent control method for inter-fracture injection-production in multi-stage fracturing horizontal wells to achieve accurate control of the inter-fracture injection-production process in multi-stage fracturing horizontal wells. Summary of the Invention

[0004] This invention aims to solve the above problems and proposes a precise intelligent control method for injection and production between fractured sections of multi-stage fractured horizontal wells. This method realizes pressure early warning for the injection and production process between fractured sections of multi-stage fractured horizontal wells and accurately controls the injection and production process of multi-stage fractured horizontal wells based on the predicted pressure, providing a new solution for the intelligent development and control of complex oil and gas reservoirs.

[0005] The present invention adopts the following technical solution:

[0006] A method for precise injection and production control between fractured sections in a multi-stage fracturing horizontal well includes the following steps:

[0007] Step 1: Obtain the communication parameter combination between the wellhead system and the downhole system of the multi-stage fracturing horizontal well. Based on the particle swarm optimization method, perform global optimization on each communication parameter in the communication parameter combination to determine the globally optimal communication parameter combination and enhance the data signal transmission performance between the wellhead system and the downhole system.

[0008] Step 2: Based on the global optimal communication parameters, control the downhole system and use the downhole system to issue injection and production scheduling commands to intelligently control the injection and production sections of the multi-stage fracturing horizontal well;

[0009] Step 3: During the injection and production process of multi-stage fracturing horizontal wells, raw measurement data is collected in real time using the downhole system and uploaded to the wellhead system to obtain a historical dataset. The measurement data in the historical dataset are preprocessed to establish a pressure database including training and test sets.

[0010] Step 4: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure. Use the training set in the pressure database to train the pressure early warning model to predict the pressure at future times based on flow rate, temperature and pressure data, and issue an early warning according to the preset pressure early warning trigger mechanism to obtain the trained pressure early warning model. Then use the test set to verify the trained pressure early warning model, and embed the verified pressure early warning model into the wellhead system of the multi-stage fracturing horizontal well.

[0011] Step 5: Input the flow rate, temperature and pressure data collected by the downhole system into the pressure early warning model in the wellhead system of the multi-stage fracturing horizontal well in real time. Use the pressure early warning model to perform pressure trend prediction, and trigger the pressure alarm device immediately when an overpressure condition occurs according to the pressure early warning trigger mechanism. Automatically execute the preset well site safety control measures and stop the injection and production of the multi-stage fracturing horizontal well.

[0012] Preferably, step 1 includes the following sub-steps:

[0013] Step 1.1, the communication parameter combination includes multiple communication parameters, namely carrier frequency. Modulation index Channel coding rate and transmission power Based on the particle swarm optimization method, the combination of communication parameters is used as the individual to be optimized. The maximum number of iterations is preset, and the optimization objective is set to maximize the transmission rate of the communication parameter combination while minimizing the bit error rate and the power consumption of the communication parameter combination. The fitness function is constructed as follows:

[0014] ;

[0015] In the formula, The fitness function; , , All are weighting coefficients. ; Bit error rate; Effective bit rate; For power efficiency; This represents the maximum effective bit rate. This represents the maximum power efficiency.

[0016] Step 1.2: Generate an initial population based on the Logistic chaotic mapping, wherein the formula for the Logistic chaotic mapping is:

[0017] ;

[0018] In the formula, , The first The, the A chaotic sequence value; These are the control parameters for the Logistic mapping;

[0019] Using the initial population as the current population, the initial values ​​of each individual in the current population in each dimension are mapped from the chaotic sequence to the feasible region of the communication parameters, resulting in:

[0020] ;

[0021] In the formula, For the first The individual in the first Initial values ​​for each dimension; , Corresponding to the first The upper and lower limits of the values ​​for each communication parameter; It is a chaotic number;

[0022] Step 1.3: Using the roulette wheel selection method, select parent individuals based on the fitness of each individual in the current population. and parental individuals Perform arithmetic crossover to generate offspring individuals. and offspring individuals ,get:

[0023] ;

[0024] ;

[0025] In the formula, These are the cross-weighting coefficients;

[0026] Randomly perturb and mutate the generated offspring individuals to obtain:

[0027] ;

[0028] In the formula, For the offspring individuals after the mutation in the first month The values ​​that can be taken in each dimension; For offspring individuals before mutation, in the first The values ​​that can be taken in each dimension; For the first The Gaussian random numbers corresponding to each dimension, wherein the mean of the Gaussian random numbers is 0 and the variance is . ;

[0029] Introduce chaotic perturbations to the best individual in the current population. In the process, new exploration points are generated. ,get:

[0030] ;

[0031] In the formula, The amplitude of the disturbance; A vector generated from a chaotic sequence;

[0032] Step 1.4: Determine if the current iteration count has reached the preset maximum iteration count. If not, return to Step 1.3 to continue optimization; otherwise, terminate the optimization, output and execute the globally optimal communication parameter combination. , ,in, For the optimal carrier frequency, This is the optimal modulation index. For the optimal channel coding rate, To achieve the optimal transmission power, This is the transpose of the matrix.

[0033] Preferably, in step 2, for the injection section of a multi-stage fractured horizontal well, the intelligent injector of the injection section is controlled based on globally optimal communication parameters to realize segmented flow regulation and status feedback of the multi-stage fractured horizontal well, including the following sub-steps:

[0034] Step 2.1.1: The wellhead system of the multi-stage fracturing horizontal well encodes the instruction containing the opening command, target fracture location, preset injection volume, and preset cumulative injection volume as a downlink wavecode pulse based on the preset water injection scheme and the global optimal communication parameters. This pulse is transmitted to the downhole system through the annular fluid medium of the multi-stage fracturing horizontal well. After receiving the downlink wavecode pulse, the downhole system controls the intelligent injectors of each injection segment. This allows each intelligent injector to continuously monitor and demodulate the fluid wavecode in the annular fluid medium and perform position comparison while in a dormant state. When the intelligent injector at the target fracture location recognizes an instruction that matches its own fracture location, the intelligent injector at the target fracture location immediately wakes up the main control chip and peripherals and enters the working state. The intelligent injectors at other non-target fracture locations continue to remain in a dormant state.

[0035] Step 2.1.2: The intelligent injection device at the target suture segment uses a built-in electromagnetic flowmeter to measure the instantaneous flow rate of the target suture segment in real time. and the preset dispensing volume Compare the results and calculate the flow deviation rate. , ;

[0036] Step 2.1.3: Use the main control chip of the intelligent dispensing unit to determine the flow deviation rate. Does it exceed the preset error threshold? If the flow deviation rate The error threshold was not exceeded. If the flow rate is within acceptable limits, the downhole system will maintain the nozzle opening unchanged; otherwise, the intelligent dispenser will control a micro-motor to rotate the nozzle valve core via a reducer, adjusting the nozzle opening. At that time, increase the opening of the water tap. At that time, reduce the opening of the water tap until the flow deviation rate is reached. Not exceeding the preset error threshold ;

[0037] Step 2.1.4: When the intelligent injector at the target fracture segment detects that the instantaneous flow rate is stable and the cumulative injection volume has reached the preset cumulative injection volume, the intelligent injector uses its wavecode generator to modulate the injection completion signal into an upward wavecode pulse and feed it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends a termination pulse command to the downhole system. The downhole system controls the intelligent injector at the target fracture segment to reset and close the water nozzle and cut off the power supply to the external devices, so that the intelligent injector at the target fracture segment returns to the dormant state and completes the water injection operation for the target fracture segment.

[0038] Preferably, in step 2, for the production section of a multi-stage fractured horizontal well, the intelligent opening and segmented oil production of the production section are controlled based on globally optimal communication parameters, including the following sub-steps:

[0039] Step 2.2.1: The intelligent switching valves of each production section of the multi-stage fractured horizontal well are in a dormant state. Based on the globally optimal communication parameters, the instruction containing the opening command, target fracture position, initial opening degree, and preset cumulative production is encoded as a downlink wavecode pulse and sent to the downhole system through the annular fluid medium of the multi-stage fractured horizontal well. After receiving the downlink wavecode pulse, the downhole system controls the intelligent switching valves of each production section. The decoder in each intelligent switching valve is used to verify and compare the position. When the intelligent switching valve at the target fracture position recognizes the instruction that matches its own fracture position, the intelligent switching valve at the target fracture position is immediately activated and drives the motor to rotate the valve core to the initial opening degree, connecting the production layer and the wellbore to establish a production channel. The intelligent switching valves at other non-target fracture positions are in a closed state, and the opening and closing status of each intelligent switching valve is fed back to the wellhead system through the uplink wavecode pulse.

[0040] Step 2.2.2: The flow monitoring device in the intelligent switch valve at the target fracture segment is used to monitor the cumulative production of the target fracture segment in real time and compare it with the preset cumulative production. If the preset cumulative production is not reached, the intelligent switch valve at the target fracture segment maintains the current opening and continues production. Otherwise, the wavecode generator of the intelligent switch valve modulates the production completion signal into an upward wavecode pulse and feeds it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends an end pulse command to the downhole system.

[0041] Step 2.2.3: After receiving the end pulse command, the intelligent switch valve at the target fracture section of the downhole system immediately drives the motor of the intelligent switch valve to rotate in the opposite direction to close the valve core of the intelligent switch valve and cut off the production channel. At this time, the intelligent switch valve at the target fracture section cuts off the power supply to the external device and enters the sleep state.

[0042] Preferably, step 3 includes the following sub-steps:

[0043] Step 3.1: Obtain the raw measurement data collected by the downhole system and construct a flow rate sequence. Pressure sequence and temperature sequence The historical dataset, the traffic sequence is The pressure sequence is The temperature sequence is ,in, For the time of collection, , , Corresponding to Flow data, pressure data, and temperature data are collected in real time;

[0044] Step 3.2: After smoothing and normalizing the measurement data within each sequence of the historical dataset, the data is then processed according to the pressure sequence. Obtain the pressure change characteristics of each pressure data, including moving average, rate of change and volatility. Set pressure stage labels for each pressure data in the pressure series according to the pressure change characteristics, so as to mark whether the pressure data is in the normal pressure stage, pressure rise warning stage, high pressure danger stage or overpressure emergency stage.

[0045] Step 3.3: Set up a sliding window, with the direction of increasing sampling time as the sliding window movement direction, and sample data in the flow, pressure and temperature sequences respectively. Randomly allocate the collected sample data to the training set and test set according to a preset ratio to establish a pressure database.

[0046] Preferably, step 4 includes the following sub-steps:

[0047] Step 4.1: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure, and set the pressure early warning triggering mechanism of the pressure early warning model;

[0048] Step 4.2: Set the training batch, optimizer, learning rate, and loss function for the stress warning model;

[0049] Step 4.3: Randomly select multiple sample data from the training set according to the training batch and input them into the pressure early warning model. Train the pressure early warning model to predict the pressure at future times based on the flow rate, temperature and pressure of the sample data. Calculate the loss function value of the pressure early warning model during each training process.

[0050] Step 4.4: Compare the calculated loss function value with the preset loss function value. If the calculated loss function value is greater than the preset loss function value, adjust the hyperparameters of the stress warning model and return to step 4.3 to continue training the stress warning model. Otherwise, stop training the stress warning model and proceed to step 4.5.

[0051] Step 4.5: Randomly select multiple sample data from the test set and input them into the trained stress early warning model to predict stress. Obtain the loss function value of the stress early warning model. If the loss function value of the stress early warning model is greater than the preset loss function value, return to step 4.3; otherwise, proceed to step 4.6.

[0052] Step 4.6: Stop the verification of the pressure early warning model and obtain the verified pressure early warning model.

[0053] Preferably, in step 4, the pressure early warning model includes an input layer, an LSTM encoding layer, a Transformer encoding layer, and an output layer;

[0054] The LSTM encoding layer is built on an LSTM neural network and includes an input gate, a forget gate, and an output gate. It is used to obtain candidate cell state values ​​and update cell states. The calculation formula is as follows:

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] ;

[0060] In the formula, for The output of the input gate is always being input; It is the sigmoid function; Here is the weight matrix of the input gate; This is the bias term for the input gate; for The hidden state of the LSTM at time step; for The hidden state of the LSTM at time step; for Input feature data at any given time; for The output of the forget gate; Here is the weight matrix for the forget gate; For the bias term of the forget gate; for Candidate values ​​for cell state at time step; This is the weight matrix for candidate cell states; The bias term for candidate cell state values; It is the hyperbolic tangent function; for Cellular state at any given moment for Cellular state at any given moment; This indicates element-wise multiplication; for The output of the output gate is always being output; This is the weight matrix of the output gate; This is the bias term for the output gate;

[0061] The Transformer coding layer is used for self-attention mechanism calculation, multi-head attention calculation, and feedforward network calculation. The calculation formula is as follows:

[0062] ;

[0063] ;

[0064] ;

[0065] In the formula, For attention output; , , All of them are matrices obtained by mapping input feature data; It is the transpose matrix; matrix elements The dimension; This is a function that normalizes the similarity score into attention weights; For multi-headed attention output; For the first The output of an independent self-attention head; A function that combines the outputs of multiple attention heads; The weight matrix for multi-head attention output; This is the output of the feedforward network; This serves as the input to the feedforward network; It is the ReLU activation function; , These are all weight matrices of the feedforward network; , These are all bias terms of the feedforward network;

[0066] The output layer is used to output pressure prediction results, including single-step pressure prediction results and multi-step pressure prediction results, calculated using the following formula:

[0067] ;

[0068] ;

[0069] In the formula, for The pressure value predicted by the pressure early warning model at all times is used to represent the single-step pressure prediction result; This is the weight matrix of the output layer; This is the function used to perform the flattening operation; The output features of the Transformer encoding layer; For the bias term of the output layer; From Time's up The pressure value is constantly predicted using the pressure early warning model; This represents the multi-step pressure prediction results of the pressure early warning model; From Time's up The sequence of input feature data at each time step; The length of the input feature data sequence.

[0070] Preferably, the pressure early warning triggering mechanism sets multi-level dynamic early warning thresholds based on the pressure sequence in the historical dataset, including a normal state threshold, a first-level early warning threshold, and a second-level early warning threshold, calculated using the following formula:

[0071] ;

[0072] ;

[0073] ;

[0074] in,

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] In the formula, This is the threshold for the normal state. The threshold for Level 1 early warning; The threshold is set at level two. This is the baseline threshold for normal operation. The threshold for Level 1 early warning; The threshold is the baseline threshold for Level II early warning; This represents the historical average pressure. The standard deviation of historical stress; For dynamic adjustment factors; To adjust the weighting coefficients; for Rate of change of pressure data over time; This represents the highest historical pressure value. This represents the historical minimum pressure.

[0080] Multi-level early warnings are issued based on the pressure prediction results from the pressure early warning model and the pressure early warning triggering mechanism. The triggering conditions for the multi-level early warnings are set as follows:

[0081] ;

[0082] In the formula, for The pressure value is predicted by the pressure early warning model at all times.

[0083] Preferably, the loss function is set as follows:

[0084] ;

[0085] in,

[0086] ;

[0087] ;

[0088] In the formula, The loss function; To reduce prediction accuracy loss; This is the regularization loss; For serial numbers; The total number of sample data points involved in the calculation; For the first The true stress value of each sample data point; For the first Each sample data point uses pressure values ​​predicted by a pressure early warning model; This is the mean square error term; This is the mean absolute error term; The weighting coefficient for the mean absolute error term; This is the set of hyperparameters for the pressure early warning model; For L1 regularization; For L2 regularization; These are the weighting coefficients for the L2 regularization term; The weight coefficients are the L1 regularization terms.

[0089] Preferably, in step 5, when the pressure prediction value exceeds the secondary warning threshold during pressure trend extrapolation by the pressure early warning model, an overpressure situation is determined to have occurred. At this time, the multi-stage fracturing horizontal wellhead system triggers the pressure alarm device to issue an alarm, automatically executes the shutdown of injection and production, and shuts down the blowout preventer.

[0090] The present invention has the following beneficial effects:

[0091] (1) This invention proposes a fine-grained intelligent control method for inter-segment injection and production in multi-stage fracturing horizontal wells. It combines the pseudo-random characteristics of chaotic mapping with the global search capability of evolutionary algorithms, which significantly improves the transmission performance of wavecode communication in complex downhole environments. By generating spreading code sequences with good correlation and randomness through Logistic chaotic mapping, the anti-interference capability and confidentiality of the communication system are effectively improved. At the same time, the chaotic sequences are optimized and screened through selection, crossover and mutation operations, so that the generated wavecodes not only maintain the excellent characteristics of chaotic sequences, but also meet the requirements of the communication system for inter-symbol interference and bit error rate.

[0092] (2) This invention proposes a fine injection and production intelligent control method for inter-segment injection and production in multi-stage fractured horizontal wells, which realizes closed-loop fine control and collaborative feedback of the injection and production process between inter-segment injection and production in multi-stage fractured horizontal wells. It adaptively adjusts based on formation dynamic changes, and timely controls the precise water injection and automatic dormancy of the injection segment of the multi-stage fractured horizontal well, as well as the intelligent start-up and shutdown and segmented oil production of the production segment. It effectively solves the problems of lag and low accuracy of traditional manual multi-stage fractured horizontal well injection and production schemes, significantly improves the reservoir utilization and water injection qualification rate, and realizes fine injection and production intelligent control of multi-stage fractured horizontal wells.

[0093] (3) This invention proposes a fine-grained intelligent control method for injection and production between fractured sections of multi-stage fracturing horizontal wells. It combines machine learning methods with downhole pressure early warning and uses an LSTM-Transformer hybrid network structure to establish a pressure early warning model. This method retains the ability of LSTM to capture long-term temporal dependencies and leverages the advantages of Transformer's multi-head attention mechanism in extracting multi-parameter correlation features. It provides early warning by accurately predicting pressure change trends and couples the injection and production control of multi-stage fracturing horizontal wells with safety interlocking, thus realizing intelligent management of the entire process from detection and prediction to early warning of pressure anomalies. Attached Figure Description

[0094] Figure 1 This is a schematic diagram of a method for precise injection and production control between fractured sections in a multi-stage fracturing horizontal well, according to the present invention.

[0095] Figure 2 The curve showing the change of the fitness function during the global optimization of communication parameters.

[0096] Figure 3 This is a comparison chart of the predicted values ​​and actual pressure values ​​of the pressure early warning model of this invention.

[0097] Figure 4 This is a comparison chart showing the utilization of various injection and extraction control methods.

[0098] Figure 5 This is a comparison chart of the injection pass rates for various injection control methods. Detailed Implementation

[0099] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0100] This invention proposes a method for precise injection and production control between fractured sections in multi-stage fracturing horizontal wells, such as... Figure 1 As shown, the specific steps include:

[0101] Step 1: Obtain the communication parameter combination between the wellhead system and the downhole system of the multi-stage fractured horizontal well. Based on the particle swarm optimization method, globally optimize each communication parameter in the communication parameter combination to determine the globally optimal communication parameter combination. Then, use the globally optimal communication parameter combination to control the downhole system, enhancing the data signal transmission performance between the wellhead system and the downhole system. This includes the following sub-steps:

[0102] Step 1.1, the communication parameter combination includes multiple communication parameters, namely carrier frequency. Modulation index Channel coding rate and transmission power Based on the particle swarm optimization method, the combination of communication parameters is used as the individual to be optimized. The maximum number of iterations is preset, and the optimization objective is set to maximize the transmission rate of the communication parameter combination while minimizing the bit error rate and the power consumption of the communication parameter combination. A fitness function is then constructed.

[0103] The fitness function is:

[0104] ;

[0105] In the formula, The fitness function; , , All are weighting coefficients. ; Bit error rate; Effective bit rate; For power efficiency; This represents the maximum effective bit rate. This represents the maximum power efficiency.

[0106] Step 1.2: Generate an initial population based on the Logistic chaotic mapping, wherein the formula for the Logistic chaotic mapping is:

[0107] ;

[0108] In the formula, For the first A chaotic sequence value; For the first A chaotic sequence value; These are the control parameters for the Logistic mapping.

[0109] Using the initial population as the current population, the initial values ​​of each individual in the current population in each dimension are mapped from the chaotic sequence to the feasible region of the communication parameters, resulting in:

[0110] ;

[0111] In the formula, For the first The individual in the first Initial values ​​for each dimension; , Corresponding to the first The upper and lower limits of the values ​​for each communication parameter; is a chaotic number, generated by the Logistic chaotic mapping.

[0112] Step 1.3: Using the roulette wheel selection method, select parent individuals based on the fitness of each individual in the current population. and parental individuals Perform arithmetic crossover to generate offspring individuals. and offspring individuals ,get:

[0113] ;

[0114] ;

[0115] In the formula, The cross-weighting coefficients are as follows: Random numbers.

[0116] To maintain population diversity, random perturbation mutations are applied to the generated offspring individuals, resulting in:

[0117] ;

[0118] In the formula, For the offspring individuals after the mutation in the first month The values ​​that can be taken in each dimension; For offspring individuals before mutation, in the first The values ​​that can be taken in each dimension; For the first The Gaussian random numbers corresponding to each dimension, wherein the mean of the Gaussian random numbers is 0 and the variance is . .

[0119] To avoid getting trapped in local optima, chaotic perturbations are introduced into the current best individual in the population. In the process, new exploration points are generated. ,get:

[0120] ;

[0121] In the formula, The amplitude of the disturbance; is a vector generated from a chaotic sequence.

[0122] Step 1.4: Determine if the current iteration count has reached the preset maximum iteration count. If not, return to Step 1.3 to continue optimization; otherwise, terminate the optimization, output and execute the globally optimal communication parameter combination. , ,in, For the optimal carrier frequency, This is the optimal modulation index. For the optimal channel coding rate, To achieve the optimal transmission power, This is the transpose of the matrix.

[0123] In this embodiment, the change of the fitness function during the global optimization of communication parameters is as follows: Figure 2 As shown, by Figure 2It can be seen that the fitness function converges after 800 iterations, at which point the optimal fitness, average fitness, and worst fitness all reach stability, thus determining the optimal combination of communication parameters. The multi-stage fracturing horizontal wellhead system utilizes a globally optimal combination of communication parameters to control the downhole system, employing the optimal carrier frequency. and optimal transmit power Send a string of optimal modulation indexes The encoded wake-up pulse command wakes up the downhole system. After being woken up, the downhole system replies with an acknowledgment signal using the same parameter set, thus completing the link synchronization between the wellhead system and the downhole system.

[0124] Step 2: Based on globally optimal communication parameters, control the downhole system. Utilize the downhole system to issue injection-production scheduling commands to intelligently control the injection and production sections of the multi-stage fractured horizontal well. Specifically, for the injection section of the multi-stage fractured horizontal well, control the intelligent injector based on globally optimal communication parameters to achieve segmented flow regulation and status feedback. For the production section of the multi-stage fractured horizontal well, control the intelligent opening and segmented oil production of the production section based on globally optimal communication parameters. This includes the following sub-steps:

[0125] Step 2.1.1: The wellhead system of the multi-stage fractured horizontal well, based on the preset water injection scheme and global optimal communication parameters, encodes the command containing the opening command, target fracture location, preset injection volume, and preset cumulative injection volume as a downlink wavecode pulse. This pulse is transmitted to the downhole system through the annular fluid medium of the multi-stage fractured horizontal well. Upon receiving the downlink wavecode pulse, the downhole system controls the intelligent injectors of each injection segment. This allows each intelligent injector to continuously monitor and demodulate the fluid wavecode in the annular fluid medium and perform position comparison while in a dormant state. When the intelligent injector at the target fracture location identifies a command that matches its own fracture location, it immediately wakes up the main control chip and peripherals and enters the working state. The intelligent injectors at other non-target fracture locations remain in a dormant state.

[0126] Step 2.1.2: The intelligent injection device at the target suture segment uses a built-in electromagnetic flowmeter to measure the instantaneous flow rate of the target suture segment in real time. and the preset dispensing volume Compare the results and calculate the flow deviation rate. , and compared with the preset error threshold Compare them.

[0127] Step 2.1.3: Use the main control chip of the intelligent dispensing unit to determine the flow deviation rate. Does it exceed the preset error threshold? In this embodiment, the error threshold The injection volume is set according to the injection volume, specifically 10% to 15% of the injection volume value; if the flow deviation rate... The error threshold was not exceeded. If the flow rate is within acceptable limits, the downhole system will maintain the nozzle opening unchanged; otherwise, the intelligent dispenser will control a micro-motor to rotate the nozzle valve core via a reducer, adjusting the nozzle opening. At that time, increase the opening of the water tap. At that time, reduce the opening of the water tap until the flow deviation rate is reached. Not exceeding the preset error threshold .

[0128] Step 2.1.4: When the intelligent injector at the target fracture segment detects that the instantaneous flow rate is stable and the cumulative injection volume has reached the preset cumulative injection volume, the intelligent injector uses its wavecode generator to modulate the injection completion signal into an upward wavecode pulse and feed it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends a termination pulse command to the downhole system. The downhole system controls the intelligent injector at the target fracture segment to reset and close the water nozzle and cut off the power supply to the external devices, so that the intelligent injector at the target fracture segment returns to the dormant state and completes the water injection operation for the target fracture segment.

[0129] Step 2.2.1: The intelligent switching valves of each production section in the multi-stage fractured horizontal well are in a dormant state. Based on the globally optimal communication parameters, the instruction containing the opening command, target fracture location, initial opening degree, and preset cumulative production is encoded as a downlink ripple pulse and sent to the downhole system through the annular fluid medium of the multi-stage fractured horizontal well. After receiving the downlink ripple pulse, the downhole system controls the intelligent switching valves of each production section. The decoder in each intelligent switching valve is used to verify and compare the position. When the intelligent switching valve at the target fracture location recognizes the instruction that matches its own fracture location, the intelligent switching valve at the target fracture location is immediately activated and drives the motor to rotate the valve core to the initial opening degree, connecting the production formation and the wellbore to establish a production channel. The intelligent switching valves at other non-target fracture locations are in a closed state, and the opening and closing status of each intelligent switching valve is fed back to the wellhead system through the uplink ripple pulse.

[0130] Step 2.2.2: The flow monitoring device inside the intelligent switch valve at the target fracture segment is used to monitor the cumulative production of the target fracture segment in real time and compare it with the preset cumulative production. If the preset cumulative production is not reached, the intelligent switch valve at the target fracture segment maintains the current opening and continues production. Otherwise, the wavecode generator of the intelligent switch valve modulates the production completion signal into an upward wavecode pulse and feeds it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends a termination pulse command to the downhole system.

[0131] Meanwhile, when the flow monitoring device inside the intelligent switching valve detects abnormal fluctuations in the production volume or approaches the preset cumulative production volume, it feeds back to the wellhead system in real time through wavecode signals and uses the downhole system to receive the opening adjustment command issued by the wellhead system to dynamically optimize the oil production parameters.

[0132] Step 2.2.3: After receiving the end pulse command, the intelligent switch valve at the target fracture section of the downhole system immediately drives the motor of the intelligent switch valve to rotate in the opposite direction to close the valve core of the intelligent switch valve and cut off the production channel. At this time, the intelligent switch valve at the target fracture section cuts off the power supply to the external device and enters the sleep state, waiting for the opening command in the next downlink wave code pulse.

[0133] Step 3: During the injection and production process of a multi-stage fracturing horizontal well, raw measurement data is collected in real time using the downhole system and uploaded to the wellhead system to obtain a historical dataset. Preprocessing is performed on each measurement data point in the historical dataset to establish a pressure database including training and testing sets. This specifically includes the following sub-steps:

[0134] Step 3.1: Obtain the raw measurement data collected by the downhole system and construct a flow rate sequence. Pressure sequence and temperature sequence The historical dataset, the traffic sequence is The pressure sequence is The temperature sequence is ,in, For the time of collection, , , Corresponding to Flow data, pressure data, and temperature data are collected in real time.

[0135] Step 3.2: A mean filter is used to smooth the measurement data within each sequence of the historical dataset, reducing data noise. Then, the measurement data within each sequence of the historical dataset is normalized to eliminate the adverse effects caused by outlier data within the sequence. Finally, based on the pressure sequence... Obtain the moving average, rate of change, and volatility of each pressure data point. Set pressure stage labels for each pressure data point in the pressure series to indicate whether the pressure data is in the normal pressure stage, the pressure rise warning stage, the high pressure danger stage, or the overpressure emergency stage.

[0136] In this embodiment, the moving average is used to reflect the short-term stability of pressure data and serves as the basis for determining whether the current pressure is in a stable or slowly changing state. Its calculation formula is as follows:

[0137] ;

[0138] In the formula, for Moving average of stress data over time; For serial numbers; To calculate the total number of moments within the period; for Stress data at any given moment.

[0139] The rate of change is used to directly quantify the rate of change of pressure over time and is the basis for distinguishing between slow and rapid pressure changes. Its calculation formula is as follows:

[0140] ;

[0141] In the formula, for Rate of change of pressure data over time; for Stress data at any given moment; This represents the time difference between two adjacent pressure data points.

[0142] The volatility is used to reflect the amplitude of pressure data fluctuations within a short-term window and serves as the basis for identifying whether pressure has entered a dangerous phase of abnormal fluctuations. Its calculation formula is as follows:

[0143] ;

[0144] In the formula, for Volatility of pressure data at any given time.

[0145] Specifically, in this embodiment, the pressure data in the pressure sequence are divided into stages based on the pressure change characteristics to determine the pressure stage in which the pressure data is located. The pressure stage division rule is as follows:

[0146] When the rate of change of pressure data is at a low level, short-term volatility is stable, and the pressure value falls within the normal range of historical statistics, the current pressure stage is determined to be a normal pressure stage. When the rate of change of pressure data shows an upward trend, short-term volatility increases, and the pressure value approaches the warning threshold range of historical statistics, the current pressure stage is determined to be a pressure rise warning stage. When the rate of change of pressure data rises rapidly, short-term volatility increases significantly, and the pressure value enters the danger range of historical statistics, the current pressure stage is determined to be a high-pressure danger stage. When the rate of change of pressure data breaks through the historical high, short-term volatility fluctuates violently, or the pressure value exceeds the safe upper limit of historical statistics, the current pressure stage is determined to be an overpressure emergency stage.

[0147] Step 3.3: Set up a sliding window, with the direction of increasing sampling time as the sliding window movement direction, and sample data in the flow, pressure and temperature sequences respectively. Randomly allocate the collected sample data to the training set and test set according to a preset ratio to establish a pressure database.

[0148] Step 4: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure. Use the training set in the pressure database to train the pressure early warning model to predict future pressure based on flow rate, temperature, and pressure data. Issue early warnings according to a preset pressure early warning trigger mechanism to obtain the trained pressure early warning model. Then, use a test set to validate the trained pressure early warning model to obtain the validated pressure early warning model, which is then embedded into the multi-stage fracturing horizontal wellhead system. This specifically includes the following sub-steps:

[0149] Step 4.1: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure, and set the pressure early warning triggering mechanism of the pressure early warning model.

[0150] In this embodiment, the pressure early warning model includes an input layer, an LSTM encoding layer, a Transformer encoding layer, and an output layer. The LSTM encoding layer, built upon an LSTM neural network, includes an input gate, a forget gate, and an output gate, used to obtain candidate cell state values ​​and update cell states. The calculation formula is as follows:

[0151] ;

[0152] ;

[0153] ;

[0154] ;

[0155] ;

[0156] In the formula, for The output of the input gate is always being input; It is the sigmoid function; Here is the weight matrix of the input gate; This is the bias term for the input gate; for The hidden state of the LSTM at time step; for The hidden state of the LSTM at time step; for Input feature data at any given time; for The output of the forget gate; Here is the weight matrix for the forget gate; For the bias term of the forget gate; for Candidate values ​​for cell state at time step; This is the weight matrix for candidate cell states; The bias term for candidate cell state values; It is the hyperbolic tangent function; for Cellular state at any given moment for Cellular state at any given moment; This indicates element-wise multiplication; for The output of the output gate is always being output; This is the weight matrix of the output gate; This is the bias term for the output gate.

[0157] The Transformer coding layer is used for self-attention mechanism calculation, multi-head attention calculation, and feedforward network calculation. The calculation formula is as follows:

[0158] ;

[0159] ;

[0160] ;

[0161] In the formula, For attention output; , , All of them are matrices obtained by mapping input feature data; It is the transpose matrix; matrix elements The dimension; This is a function that normalizes the similarity score into attention weights; For multi-headed attention output; For the first The output of an independent self-attention head; A function that combines the outputs of multiple attention heads; The weight matrix for multi-head attention output; This is the output of the feedforward network; This serves as the input to the feedforward network; The ReLU activation function is used to introduce nonlinearity. , These are all weight matrices of the feedforward network; , These are all bias terms of the feedforward network.

[0162] The output layer is used to output pressure prediction results, including single-step pressure prediction results and multi-step pressure prediction results, calculated using the following formula:

[0163] ;

[0164] ;

[0165] In the formula, for The pressure value predicted by the pressure early warning model at all times is used to represent the single-step pressure prediction result; This is the weight matrix of the output layer; This is the function used to perform the flattening operation; The output features of the Transformer encoding layer; For the bias term of the output layer; From Time's up The pressure value is constantly predicted using the pressure early warning model; This represents the multi-step pressure prediction results of the pressure early warning model; From Time's up The sequence of input feature data at each time step; The length of the input feature data sequence.

[0166] Specifically, in this embodiment, the pressure warning triggering mechanism sets multi-level dynamic warning thresholds based on the pressure sequence in the historical dataset, including a normal state threshold, a first-level warning threshold, and a second-level warning threshold. The calculation formula is as follows:

[0167] ;

[0168] ;

[0169] ;

[0170] in,

[0171] ;

[0172] ;

[0173] ;

[0174] ;

[0175] In the formula, This is the threshold for the normal state. The threshold for Level 1 early warning; The threshold is set at level two. This is the baseline threshold for normal operation. The threshold for Level 1 early warning; The threshold is the baseline threshold for Level II early warning; This represents the historical average pressure. The standard deviation of historical stress; For dynamic adjustment factors; To adjust the weighting coefficients; for Rate of change of pressure data over time; This represents the highest historical pressure value. This represents the minimum historical pressure.

[0176] Multi-level early warnings are issued based on the pressure prediction results from the pressure early warning model and the pressure early warning triggering mechanism. The triggering conditions for the multi-level early warnings are set as follows:

[0177] ;

[0178] In the formula, for The pressure value is predicted by the pressure early warning model at all times.

[0179] Step 4.2 sets the maximum number of iterations for the stress warning model to 300, the training batch size to 50, the optimizer to the AdamW optimizer, the learning rate to 0.001, and the loss function, wherein the loss function is set as follows:

[0180] ;

[0181] in,

[0182] ;

[0183] ;

[0184] In the formula, The loss function; To reduce prediction accuracy loss; This is the regularization loss; For serial numbers; The total number of sample data points involved in the calculation; For the first The true stress value of each sample data point; For the first Each sample data point uses pressure values ​​predicted by a pressure early warning model; This is the mean square error term; This is the mean absolute error term; The weighting coefficient for the mean absolute error term; This is the set of hyperparameters for the pressure early warning model; For L1 regularization; For L2 regularization; These are the weighting coefficients for the L2 regularization term; The weight coefficients are the L1 regularization terms.

[0185] Step 4.3: Randomly select multiple sample data from the training set according to the training batch and input them into the pressure early warning model. Train the pressure early warning model to predict the pressure at future times based on the flow rate, temperature and pressure of the sample data. Calculate the loss function value of the pressure early warning model during each training process.

[0186] Step 4.4: Compare the calculated loss function value with the preset loss function value. If the calculated loss function value is greater than the preset loss function value, adjust the hyperparameters of the pressure warning model and return to step 4.3 to continue training the pressure warning model. Otherwise, stop training the pressure warning model and proceed to step 4.5.

[0187] Step 4.5: Randomly select multiple sample data from the test set and input them into the trained stress early warning model to predict stress. Obtain the loss function value of the stress early warning model. If the loss function value of the stress early warning model is greater than the preset loss function value, return to step 4.3; otherwise, proceed to step 4.6.

[0188] Step 4.6: Stop the verification of the pressure early warning model and obtain the verified pressure early warning model.

[0189] In this embodiment, a validated pressure early warning model is used to predict pressure, and the prediction is compared with actual measured pressure data. Figure 3 As shown in the figure, the comparison revealed that the validated pressure early warning model can accurately predict pressure.

[0190] Step 5: Input the flow rate, temperature, and pressure data collected by the downhole system into the pressure early warning model in the wellhead system of the multi-stage fracturing horizontal well in real time. Use the pressure early warning model to perform pressure trend prediction, and trigger the pressure alarm device immediately when an overpressure condition occurs according to the pressure early warning trigger mechanism. At this time, the wellhead system immediately triggers the overpressure alarm and starts the emergency response program, issuing an emergency cut-off command to the downhole system. The downhole system controls the intelligent injectors of each injection section and the intelligent switching valves of each production section. After receiving the command, the intelligent injectors of each injection section and the intelligent switching valves of each production section immediately drive the micro motor to quickly reset the valve core to the fully closed state and cut off the power supply to the external devices to enter the dormant state. At the same time, the wellhead system simultaneously cuts off the power supply of the pumping unit, forcing it to stop reciprocating oil production operations, and links the hydraulic control system to quickly close the blowout preventer gate to physically isolate the wellbore from the outside world and prevent high-pressure fluid from being ejected out of control. After each device completes its emergency action, it sends the execution status back to the wellhead system via an uplink ping code. The wellhead system then verifies in real time the status information of valve closure, pumping unit shutdown, and blowout preventer closure at each joint to ensure that all safety measures are executed accurately.

[0191] The method of this invention was applied to a multi-stage fractured horizontal well. The intelligent control method for fine-grained injection and production between fracture segments in the multi-stage fractured horizontal well, proposed in this invention, was used to control the injection and production process of the well. The method was compared with general water injection methods, conventional segmented injection methods, and wavecode communication segmented injection methods. By comparing the utilization levels of general water injection methods, conventional segmented injection methods, wavecode communication segmented injection methods, and the method of this invention, such as… Figure 4 As shown, the method of this invention has the highest utilization rate, at 76.2%. Further comparison of the betting pass rates of the manual betting method, the wavecode communication betting method, and the method of this invention reveals the following: Figure 5 As shown, the intelligent control method for injection sampling achieved the highest injection qualification rate after optimizing communication parameters using the method of the present invention, thus verifying the feasibility and superiority of the method of the present invention compared with the manual measurement and adjustment injection method and the wavecode communication injection method.

[0192] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for precise injection and production control between fractured sections in a multi-stage fracturing horizontal well, characterized in that, Includes the following steps: Step 1: Obtain the communication parameter combination between the wellhead system and the downhole system of the multi-stage fracturing horizontal well. Based on the particle swarm optimization method, perform global optimization on each communication parameter in the communication parameter combination to determine the globally optimal communication parameter combination and enhance the data signal transmission performance between the wellhead system and the downhole system. Step 2: Based on the global optimal communication parameters, control the downhole system and use the downhole system to issue injection and production scheduling commands to intelligently control the injection and production sections of the multi-stage fracturing horizontal well; Step 3: During the injection and production process of multi-stage fracturing horizontal wells, raw measurement data is collected in real time using the downhole system and uploaded to the wellhead system to obtain a historical dataset. The measurement data in the historical dataset are preprocessed to establish a pressure database including training and test sets. Step 4: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure. Use the training set in the pressure database to train the pressure early warning model to predict the pressure at future times based on flow rate, temperature and pressure data, and issue an early warning according to the preset pressure early warning trigger mechanism to obtain the trained pressure early warning model. Then use the test set to verify the trained pressure early warning model, and embed the verified pressure early warning model into the wellhead system of the multi-stage fracturing horizontal well. Step 5: Input the flow rate, temperature and pressure data collected by the downhole system into the pressure early warning model in the wellhead system of the multi-stage fracturing horizontal well in real time. Use the pressure early warning model to perform pressure trend prediction, and trigger the pressure alarm device immediately when an overpressure condition occurs according to the pressure early warning trigger mechanism. Automatically execute the preset well site safety control measures and stop the injection and production of the multi-stage fracturing horizontal well.

2. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 1, is characterized in that... Step 1 includes the following sub-steps: Step 1.1, the communication parameter combination includes multiple communication parameters, namely carrier frequency. Modulation index Channel coding rate and transmission power Based on the particle swarm optimization method, the combination of communication parameters is used as the individual to be optimized. The maximum number of iterations is preset, and the optimization objective is set to maximize the transmission rate of the communication parameter combination while minimizing the bit error rate and the power consumption of the communication parameter combination. The fitness function is constructed as follows: ； In the formula, The fitness function; , , All are weighting coefficients. ; Bit error rate; Effective bit rate; For power efficiency; This represents the maximum effective bit rate. This represents the maximum power efficiency. Step 1.2: Generate an initial population based on the Logistic chaotic mapping, wherein the formula for the Logistic chaotic mapping is: ； In the formula, , The first The, the A chaotic sequence value; These are the control parameters for the Logistic mapping; Using the initial population as the current population, the initial values ​​of each individual in the current population in each dimension are mapped from the chaotic sequence to the feasible region of the communication parameters, resulting in: ； In the formula, For the first The individual in the first Initial values ​​for each dimension; , Corresponding to the first The upper and lower limits of the values ​​for each communication parameter; It is a chaotic number; Step 1.3: Using the roulette wheel selection method, select parent individuals based on the fitness of each individual in the current population. and parental individuals Perform arithmetic crossover to generate offspring individuals. and offspring individuals ,get: ； ； In the formula, These are the cross-weighting coefficients; Randomly perturb and mutate the generated offspring individuals to obtain: ； In the formula, For the offspring individuals after the mutation in the first month The values ​​that can be taken in each dimension; For offspring individuals before mutation in the first The values ​​that can be taken in each dimension; For the first The Gaussian random numbers corresponding to each dimension, wherein the mean of the Gaussian random numbers is 0 and the variance is . ; Introduce chaotic perturbations to the best individual in the current population. In the process, new exploration points are generated. ,get: ； In the formula, The amplitude of the disturbance; A vector generated from a chaotic sequence; Step 1.4: Determine if the current iteration count has reached the preset maximum iteration count. If not, return to Step 1.3 to continue optimization; otherwise, terminate the optimization, output and execute the globally optimal communication parameter combination. , ,in, For the optimal carrier frequency, This is the optimal modulation index. For the optimal channel coding rate, To achieve the optimal transmission power, This is the transpose of the matrix.

3. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 1, is characterized in that... In step 2, for the injection section of a multi-stage fractured horizontal well, the intelligent injector of the injection section is controlled based on globally optimal communication parameters to realize segmented flow regulation and status feedback of the multi-stage fractured horizontal well, including the following sub-steps: Step 2.1.1: The wellhead system of the multi-stage fracturing horizontal well encodes the instruction containing the opening command, target fracture location, preset injection volume, and preset cumulative injection volume as a downlink wavecode pulse based on the preset water injection scheme and the global optimal communication parameters. This pulse is transmitted to the downhole system through the annular fluid medium of the multi-stage fracturing horizontal well. After receiving the downlink wavecode pulse, the downhole system controls the intelligent injectors of each injection segment. This allows each intelligent injector to continuously monitor and demodulate the fluid wavecode in the annular fluid medium and perform position comparison while in a dormant state. When the intelligent injector at the target fracture location recognizes an instruction that matches its own fracture location, the intelligent injector at the target fracture location immediately wakes up the main control chip and peripherals and enters the working state. The intelligent injectors at other non-target fracture locations continue to remain in a dormant state. Step 2.1.2: The intelligent injection device at the target suture segment uses a built-in electromagnetic flowmeter to measure the instantaneous flow rate of the target suture segment in real time. and the preset dispensing volume Compare the results and calculate the flow deviation rate. , ; Step 2.1.3: Use the main control chip of the intelligent dispensing unit to determine the flow deviation rate. Does it exceed the preset error threshold? If the flow deviation rate The error threshold was not exceeded. If the current flow rate is deemed acceptable, the downhole system will be used to maintain the current nozzle opening. Otherwise, the intelligent dispenser controls a micro motor to drive the water nozzle valve core to rotate via a reducer, adjusting the water nozzle opening. At that time, increase the opening of the water tap. At that time, reduce the opening of the water tap until the flow deviation rate is reached. Not exceeding the preset error threshold ; Step 2.1.4: When the intelligent injector at the target fracture segment detects that the instantaneous flow rate is stable and the cumulative injection volume has reached the preset cumulative injection volume, the intelligent injector uses its wavecode generator to modulate the injection completion signal into an upward wavecode pulse and feed it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends a termination pulse command to the downhole system. The downhole system controls the intelligent injector at the target fracture segment to reset and close the water nozzle and cut off the power supply to the external devices, so that the intelligent injector at the target fracture segment returns to the dormant state and completes the water injection operation for the target fracture segment.

4. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 1, is characterized in that... In step 2, for the production section of a multi-stage fractured horizontal well, the intelligent opening and segmented oil production of the production section are controlled based on globally optimal communication parameters, including the following sub-steps: Step 2.2.1: The intelligent switching valves of each production section of the multi-stage fractured horizontal well are in a dormant state. Based on the globally optimal communication parameters, the instruction containing the opening command, target fracture position, initial opening degree, and preset cumulative production is encoded as a downlink wavecode pulse and sent to the downhole system through the annular fluid medium of the multi-stage fractured horizontal well. After receiving the downlink wavecode pulse, the downhole system controls the intelligent switching valves of each production section. The decoder in each intelligent switching valve is used to verify and compare the position. When the intelligent switching valve at the target fracture position recognizes the instruction that matches its own fracture position, the intelligent switching valve at the target fracture position is immediately activated and drives the motor to rotate the valve core to the initial opening degree, connecting the production layer and the wellbore to establish a production channel. The intelligent switching valves at other non-target fracture positions are in a closed state, and the opening and closing status of each intelligent switching valve is fed back to the wellhead system through the uplink wavecode pulse. Step 2.2.2: The flow monitoring device in the intelligent switch valve at the target fracture segment is used to monitor the cumulative production of the target fracture segment in real time and compare it with the preset cumulative production. If the preset cumulative production is not reached, the intelligent switch valve at the target fracture segment maintains the current opening and continues production. Otherwise, the wavecode generator of the intelligent switch valve modulates the production completion signal into an upward wavecode pulse and feeds it back to the wellhead system. After the wellhead system verifies that the upward wavecode pulse is correct, the wellhead system sends an end pulse command to the downhole system. Step 2.2.3: After receiving the end pulse command, the intelligent switch valve at the target fracture section of the downhole system immediately drives the motor of the intelligent switch valve to rotate in the opposite direction to close the valve core of the intelligent switch valve and cut off the production channel. At this time, the intelligent switch valve at the target fracture section cuts off the power supply to the external device and enters the sleep state.

5. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 1, is characterized in that... Step 3 includes the following sub-steps: Step 3.1: Obtain the raw measurement data collected by the downhole system and construct a flow rate sequence. Pressure sequence and temperature sequence The historical dataset, the traffic sequence is The pressure sequence is The temperature sequence is ,in, For the time of collection, , , Corresponding to Flow data, pressure data, and temperature data are collected in real time; Step 3.2: After smoothing and normalizing the measurement data within each sequence of the historical dataset, the data is then processed according to the pressure sequence. Obtain the pressure change characteristics of each pressure data, including moving average, rate of change and volatility. Set pressure stage labels for each pressure data in the pressure series according to the pressure change characteristics to mark whether the pressure data is in the normal pressure stage, pressure rise warning stage, high pressure danger stage or overpressure emergency stage. Step 3.3: Set up a sliding window, with the direction of increasing sampling time as the sliding window movement direction, and sample data in the flow, pressure and temperature sequences respectively. Randomly allocate the collected sample data to the training set and test set according to a preset ratio to establish a pressure database.

6. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 1, is characterized in that... Step 4 includes the following sub-steps: Step 4.1: Establish a pressure early warning model based on the LSTM-Transformer hybrid network structure, and set the pressure early warning triggering mechanism of the pressure early warning model; Step 4.2: Set the training batch, optimizer, learning rate, and loss function for the stress warning model; Step 4.3: Randomly select multiple sample data from the training set according to the training batch and input them into the pressure early warning model. Train the pressure early warning model to predict the pressure at future times based on the flow rate, temperature and pressure of the sample data. Calculate the loss function value of the pressure early warning model during each training process. Step 4.4: Compare the calculated loss function value with the preset loss function value. If the calculated loss function value is greater than the preset loss function value, adjust the hyperparameters of the stress warning model and return to step 4.3 to continue training the stress warning model. Otherwise, stop training the stress warning model and proceed to step 4.

5. Step 4.5: Randomly select multiple sample data from the test set and input them into the trained stress early warning model to predict stress. Obtain the loss function value of the stress early warning model. If the loss function value of the stress early warning model is greater than the preset loss function value, return to step 4.3; otherwise, proceed to step 4.

6. Step 4.6: Stop the verification of the pressure early warning model and obtain the verified pressure early warning model.

7. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 6, is characterized in that... In step 4, the pressure early warning model includes an input layer, an LSTM encoding layer, a Transformer encoding layer, and an output layer. The LSTM encoding layer is built on an LSTM neural network and includes an input gate, a forget gate, and an output gate. It is used to obtain candidate cell state values ​​and update cell states. The calculation formula is as follows: ； ； ； ； ； In the formula, for The output of the input gate is always being input; It is the sigmoid function; This is the weight matrix of the input gate; This is the bias term for the input gate; for The hidden state of the LSTM at time step; for The hidden state of the LSTM at time step; for Input feature data at any given time; for The output of the forget gate; Here is the weight matrix for the forget gate; For the bias term of the forget gate; for Candidate values ​​for cell state at time step; This is the weight matrix for candidate cell states; The bias term for candidate cell state values; It is the hyperbolic tangent function; for Cellular state at any given moment for Cellular state at any given moment; This indicates element-wise multiplication; for The output of the output gate is always being output; This is the weight matrix of the output gate; This is the bias term for the output gate; The Transformer coding layer is used for self-attention mechanism calculation, multi-head attention calculation, and feedforward network calculation. The calculation formula is as follows: ； ； ； In the formula, For attention output; , , All of them are matrices obtained by mapping input feature data; It is the transpose matrix; matrix elements The dimension; This is a function that normalizes the similarity score into attention weights; For multi-headed attention output; For the first The output of an independent self-attention head; A function that combines the outputs of multiple attention heads; The weight matrix for multi-head attention output; This is the output of the feedforward network; This serves as the input to the feedforward network; It is the ReLU activation function; , These are all weight matrices of the feedforward network; , These are all bias terms of the feedforward network; The output layer is used to output pressure prediction results, including single-step pressure prediction results and multi-step pressure prediction results, calculated using the following formula: ； ； In the formula, for The pressure value predicted by the pressure early warning model at all times is used to represent the single-step pressure prediction result; This is the weight matrix of the output layer; This is the function used to perform the flattening operation; The output features of the Transformer encoding layer; For the bias term of the output layer; From Time's up The pressure value is constantly predicted using the pressure early warning model; This represents the multi-step pressure prediction results of the pressure early warning model; From Time's up The sequence of input feature data at each time step; The length of the input feature data sequence.

8. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well, as described in claim 6, is characterized in that... The pressure warning triggering mechanism sets multi-level dynamic warning thresholds based on the pressure sequence in the historical dataset, including a normal state threshold, a first-level warning threshold, and a second-level warning threshold. The calculation formula is as follows: ； ； ； in, ； ； ； ； In the formula, This is the threshold for the normal state. The threshold for Level 1 early warning; The threshold is set at level two. This is the baseline threshold for normal operation. This is the baseline threshold for Level 1 early warning; The threshold is the baseline threshold for Level II early warning; This represents the historical average pressure. The standard deviation of historical stress; For dynamic adjustment factors; To adjust the weighting coefficients; for Rate of change of pressure data over time; This represents the highest historical pressure value. This represents the historical minimum pressure. Multi-level early warnings are issued based on the pressure prediction results from the pressure early warning model and the pressure early warning triggering mechanism. The triggering conditions for the multi-level early warnings are set as follows: ； In the formula, for The pressure value is predicted by the pressure early warning model at all times.

9. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well according to claim 6, characterized in that, The loss function is set as follows: ； in, ； ； In the formula, The loss function; To reduce prediction accuracy loss; This is the regularization loss; For serial numbers; The total number of sample data points involved in the calculation; For the first The true stress value of each sample data point; For the first Each sample data point uses pressure values ​​predicted by a pressure early warning model; This is the mean square error term; This is the mean absolute error term; The weighting coefficient for the mean absolute error term; This is the set of hyperparameters for the pressure early warning model; For L1 regularization; For L2 regularization; These are the weighting coefficients for the L2 regularization term; The weight coefficients are the L1 regularization terms.

10. The intelligent control method for fine injection and production between fractured sections in a multi-stage fracturing horizontal well according to claim 1, characterized in that, In step 5, when the pressure prediction value exceeds the secondary warning threshold during the pressure trend simulation of the pressure early warning model, it is determined that an overpressure situation has occurred. At this time, the multi-stage fracturing horizontal wellhead system triggers the pressure alarm device to issue an alarm, automatically executes the shutdown of injection and production, and shuts down the blowout preventer.

Images