Method for automatically optimizing fracturing fluid blowout working system after oil well fracturing
Through the deep learning model, the discharge and spraying of oil well fracturing fluid is dynamically adjusted, which solves the problems of fracturing fluid filtration loss and proppant reflux, and realizes the optimization of the fracturing effect and efficient utilization of resources.
Patent Information
- Application Number
- CN202411300283.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-12-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
After the oil well is fracturing, how to dynamically adjust the discharge and spraying system of fracturing fluid to prevent proppant from flowing back and filtration loss, ensure the reservoir's diversion capacity and safe development, it is difficult for the existing technology to achieve real-time optimization.
Using deep learning model, the real-time data of fracturing fluid loss and return volume can be obtained through sensors, convolutional calculation and pooling, predict bottom-hole pressure and crack conditions, dynamically select the oil nozzle diameter and model to ensure effective suspension of proppant and reduce filter loss.
The automatic optimization of the oil well fracturing fluid discharge and spraying system has been achieved, which has improved operational flexibility and response speed, improved fracturing effect and resource utilization, and reduced costs and risks.
Smart Images
Figure CN119177836A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of reservoir transformation engineering, and more particularly to an automatic optimization method for a fracturing fluid release working system after oil well fracturing. Background Art
[0002] In recent years, most domestic oil fields have entered the middle and late stages of development, and my country's dependence on foreign oil and natural gas has increased day by day. my country's low permeability and unconventional oil and gas resources are an important guarantee for national energy security. Reservoir fracturing transformation is the core technology to achieve efficiency and stable production of old wells in oil fields and efficient development of low permeability and unconventional oil and gas. Post-fracturing fracturing fluid flowback, as a very important link in the hydraulic fracturing process, directly affects the effect of reservoir fracturing transformation: on the one hand, the flowback speed of the fracturing fluid will affect the sedimentation and laying of the proppant in the fracture and even cause the proppant to flow back, ultimately affecting the conductivity of the fracture; on the other hand, during fracturing, the fracturing fluid will continue to filter into the formation, which will cause secondary damage to the oil and gas layer, weaken the fracturing effect and even cause fracturing failure. Therefore, the design of the post-fracturing fracturing fluid release work system is very important. It is necessary to prevent the backflow flow from being too large, resulting in a large amount of proppant backflow in the fracture, and to complete the fracturing fluid flowback as soon as possible to reduce proppant sedimentation and reduce the risk of filtration loss.
[0003] A suitable choke system is crucial for efficient and safe production development; if the choke system is too small, the reservoir production capacity cannot be fully utilized and the development efficiency is reduced; if the choke system is too large, it may cause proppant outflow, sand production, etc., resulting in excessive damage to the reservoir. A higher production capacity may be achieved in a short period of time, but the final output will be greatly reduced.
[0004] Therefore, how to select the nozzle in real time is a problem that those skilled in the art need to solve urgently. Summary of the invention
[0005] In view of this, the present invention provides an automatic optimization method for a fracturing fluid release working system after oil well fracturing, which can select an oil nozzle according to the change of the fracturing fluid over time after the oil well fracturing.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A method for automatically optimizing a fracturing fluid release working system after oil well fracturing, comprising:
[0008] Fracturing the oil well, obtaining input variables related to fracturing fluid loss and flowback volume through sensors, normalizing the input variables, using the processed input variables as input data, and arranging them in time series to construct a variable time series input layer;
[0009] Performing a convolution operation on the input data, and pooling the input data after the convolution operation, performing a full connection operation on the output data after multiple convolution and pooling, completing the training of the fracturing fluid prediction model, inputting the input variables to be predicted, and obtaining the fracturing fluid loss volume and the fracturing fluid return volume;
[0010] Calculating the bottom hole pressure according to the fracturing fluid loss volume and the fracturing fluid flowback volume, and confirming the fracture condition through the bottom hole pressure and the fracture closure pressure;
[0011] The filtration volume, flowback volume and viscosity of the fracturing fluid, as well as the casing pressure and oil pressure are converted into a two-dimensional image as the input of a preset diameter prediction model, the preset diameter prediction model is used to predict the nozzle diameter, and the flowback flow rate of the fracturing fluid is calculated based on the nozzle diameter; the critical reflux flow rate of the proppant is calculated based on the fracture conditions, and the nozzle model is selected based on the critical reflux flow rate and the flowback flow rate.
[0012] Preferably, the input variables include Poisson's ratio, wellhead pressure, fracture height, reservoir thickness, fracturing fluid temperature, fluid level, and wellbore fluid column pressure.
[0013] Preferably, performing convolution operation on the input data specifically includes performing convolution operation on the input data in different directions, specifically:
[0014] S1: Use t convolution kernels to perform vertical convolution on the input variable. The vertical convolution is calculated using the following formula:
[0015] a i,j =f(w m x i,j +z m ),m=1,2,...,t;
[0016] Among them, x i,j is the element in row i and column j of the input data; w m represents the convolution kernel weight, z m is the bias term of the convolution kernel; a i,j is the i-th row and j-th column element of the convolution data; f represents the activation function;
[0017] S2: The input layer data is convolved by t convolution kernels and activated by the relu function. After activation, t neurons are output, each of which contains a data matrix.
[0018] S3: Average pooling is performed on t neurons. The specific formula is as follows:
[0019]
[0020] Among them, q represents the size of the pooling area, D and F are the length and width of a single neuron data matrix. Since the pooling kernel is vertically pooled, only the length of the neuron matrix will decrease, p i / q,j Represents the i / qth row and jth column element of the output neuron matrix of the pooling layer;
[0021] S4: The neurons output by the pooling layer are then horizontally convolved by g convolution kernels and average pooled again, outputting g neurons;
[0022] S5: The neurons output after two convolution poolings are used as the input of the connection layer. The connection layer integrates the feature information represented by the elements of all neuron matrices into the neurons of the connection layer. Assume that the number of neurons in the connection layer is T. After multiple convolution poolings, each neuron output contains k rows and l columns of elements. Each neuron in the connection layer is convolved by g [k, l] convolution kernels on the elements in each neuron matrix. The specific formula is as follows:
[0023]
[0024] Among them, x k,l is the element in row k and column l of the input data; w n is the convolution kernel weight, z n is the bias term of the convolution kernel; the size of the convolution kernel is the same as the size of the neuron matrix, and a value is output after convolution, y u Represents the value in each neuron of the fully connected layer, that is, the value obtained by adding g values after convolution.
[0025] Preferably, the output layer in S5 is a linear weighted summation calculation of the output vector of the connection layer, the number of input neurons is T, and the output is The specific formula is as follows:
[0026]
[0027] in, is the output volume prediction value, is the predicted value of fracturing fluid loss volume, is the predicted value of fracturing fluid flowback volume.
[0028] Preferably, the method further includes updating weight parameters of the fracturing fluid prediction model using back propagation technology, designing independent adaptive learning rates for different parameters by calculating first-order moment estimation and second-order moment estimation of the gradient, updating the weights and bias of the convolutional layer, completing parameter fine-tuning, and minimizing the output error.
[0029] Preferably, the method of updating weight parameters by using back propagation technology specifically includes:
[0030] Determine the back propagation learning parameters: the back propagation algorithm learning rate is α, and the first-order moment estimation attenuation exponent β is determined at the same time 1 , second-order moment estimate decay exponent β 2 , parameter ε 1 , time step t, objective function f(θ), where β 1 and β 2 The value range is [0,1], where the θ parameter is a vector containing the forward training weights and biases;
[0031] Determine the parameters α, β 1 , β 2 After the objective function f(θ) is given, the first-order moment vector m=0, the second-order moment vector v=0 and the time step t=0 are initialized;
[0032] Calculate the first-order moment estimate and second-order moment estimate of the forecast indicator error gradient and make corrections;
[0033] Using the gradient of the previous moment The exponential decay correction value and the square gradient of the previous moment The exponential decay correction value of θ is used to update the θ parameter until the error is less than the set threshold.
[0034] Preferably, the diameter prediction model specifically includes: input layer, convolution layer 1, convolution layer 2, convolution layer 3, three BN layers, three Relu layers, pooling layer 1, pooling layer 2, fully connected layer, and output layer. The diameter prediction model structure is input layer, convolution layer 1, BN layer, Relu layer, pooling layer 1, convolution layer 2, BN layer, Relu layer, pooling layer 2, convolution layer 3, BN layer, Relu layer, fully connected layer, and output layer.
[0035] Preferably, the calculating of the first-order moment estimation and the second-order moment estimation of the prediction index error gradient and the correction specifically include:
[0036] When the parameter θ does not converge, first calculate the gradient at time t, and the calculation formula is as follows:
[0037]
[0038] Then calculate the first moment estimate m of the gradient t , the calculation formula is as follows:
[0039] m t =β 1 ·m t-1 +(1-β 1 )·b t ;
[0040] Calculate the second-order moment estimate of the gradient, the calculation formula is as follows:
[0041] v t =β 2 ·v t-1 +(1-β 2 )·b t 2 ;
[0042] The first-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows:
[0043]
[0044] The second-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows:
[0045]
[0046] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for automatically optimizing the working system of fracturing fluid release after oil well fracturing. The method is based on real-time data and deep learning models, and can dynamically adjust the release of fracturing fluid to improve the flexibility and response speed of operation; adopts back propagation technology, through first-order and second-order moment estimation, the learning rate is adaptively adjusted for different parameters to achieve continuous optimization and performance improvement of the model; through multiple convolution and pooling operations, the model can effectively extract complex feature relationships and improve the accuracy and reliability of prediction; the model can more efficiently learn the complex relationship between fracturing fluid filtration loss and return flow during training, thereby making the prediction more accurate; by combining the traditional oil well management process with intelligent algorithms, intelligent management of fracturing operations is realized, and the efficiency and safety of the entire operation are improved; by accurately predicting the flow rate and selecting the appropriate nozzle model, while achieving the best fracturing effect, the use of resources is optimized and costs are reduced. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] 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 embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0048] Figure 1 A diagram of the method steps provided by the present invention;
[0049] Figure 2 A diagram of the training process of the fracturing fluid prediction model provided by the present invention;
[0050] Figure 3 This is a diagram of the diameter prediction model provided by the present invention. DETAILED DESCRIPTION
[0051] 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.
[0052] The embodiment of the present invention discloses a method for automatically optimizing the fracturing fluid release working system after oil well fracturing. Figure 1 As shown, including:
[0053] Fracturing the oil well, obtaining input variables related to fracturing fluid loss and flowback volume through sensors, normalizing the input variables, using the processed input variables as input data, and arranging them in time series to construct a variable time series input layer;
[0054] Perform convolution operation on the input data, and pool the input data after convolution operation. After multiple convolution and pooling, perform full connection operation on the output data to complete the training of the fracturing fluid prediction model, input the input variables to be predicted, and obtain the fracturing fluid loss volume and fracturing fluid return volume;
[0055] Calculate the bottom hole pressure based on the fracturing fluid loss volume and fracturing fluid flowback volume, and confirm the fracture situation through the bottom hole pressure and fracture closure pressure;
[0056] The filtration volume, flowback volume and viscosity of the fracturing fluid, as well as the casing pressure and oil pressure converted into a two-dimensional image are used as the input of the preset diameter prediction model. The preset diameter prediction model is used to predict the nozzle diameter, and the flowback flow rate of the fracturing fluid is calculated based on the nozzle diameter. The critical reflux flow rate of the proppant is calculated according to the fracture conditions, and the nozzle model is selected based on the critical reflux flow rate and the flowback flow rate.
[0057] In a specific embodiment, when selecting a nozzle model based on the critical return flow rate and the backflow flow rate, the critical return flow rate refers to the minimum flow rate at which the proppant can remain suspended in the fracturing fluid without settling. Only when the backflow flow rate reaches or exceeds this value can the proppant be effectively brought back to the wellhead. The flow rate should be less than or equal to the critical flow rate: In order to prevent the proppant from flowing out, the backflow flow rate should be less than or equal to the critical return flow rate. This means that in theory, the backflow flow rate should not exceed the critical return flow rate to ensure that the proppant is not brought back.
[0058] The flow rate can slightly exceed the critical flow rate: In actual operation, the backflow flow rate can slightly exceed the critical flow rate, but caution is required: slightly exceeding the critical flow rate can provide better flow stability, ensuring that the proppant does not settle during the flow process. Although the flow rate can slightly exceed the critical flow rate, too high a flow rate may cause the proppant to be brought back to the wellhead, thereby increasing the risk of proppant backflow. Therefore, the choice of flow rate needs to find a balance between ensuring proppant suspension and avoiding proppant backflow.
[0059] In a specific embodiment, a sliding window with a width of n from time t to time t+n is selected in the fracturing fluid prediction model, and input variables are input row by row to form a matrix, which is then used as input data.
[0060] X i ={X i (t),X i (t+1),...,X i (t+n)}, i=1,2,...,7;
[0061] Where i is the i-th input variable.
[0062] In a specific embodiment, the input variables include Poisson's ratio, wellhead pressure, fracture height, reservoir thickness, fracturing fluid temperature, fluid level, and wellbore fluid column pressure.
[0063] In a specific embodiment, Figure 2 As shown, the convolution operation on the input data specifically includes performing convolution operations on the input data in different directions, specifically:
[0064] S1: Use t convolution kernels to perform vertical convolution on the input variable. The vertical convolution is calculated using the following formula:
[0065] a i,j =f(w m x i,j +z m ),m=1,2,...,t;
[0066] Among them, x i,j is the element in row i and column j of the input data; w m represents the convolution kernel weight, z m is the bias term of the convolution kernel; a i,j is the i-th row and j-th column element of the convolution data; f represents the activation function; this method selects the relu function as the activation function. Its formula is as follows:
[0067] f(x)=max(0,x);
[0068] S2: The input layer data is convolved by t convolution kernels and activated by the relu function. After activation, t neurons are output, each of which contains a data matrix.
[0069] S3: Average pooling is performed on t neurons. The specific formula is as follows:
[0070]
[0071] Among them, q represents the size of the pooling area, D and F are the length and width of a single neuron data matrix. Since the pooling kernel is vertically pooled, only the length of the neuron matrix will decrease, p i / q,j Represents the i / qth row and jth column element of the output neuron matrix of the pooling layer;
[0072] After average pooling, the length of a single neuron matrix is reduced by half, but the number of output neurons does not change. The neurons output by the pooling layer are then divided into n 2 The convolution kernels are horizontally convolved and average pooled again, and the output is n 2 A neuron.
[0073] S4: The neurons output by the pooling layer are then horizontally convolved by g convolution kernels and average pooled again, outputting g neurons;
[0074] S5: The neurons output after two convolution poolings are used as the input of the connection layer. The connection layer integrates the feature information represented by the elements of all neuron matrices into the neurons of the connection layer. Assume that the number of neurons in the connection layer is T. After multiple convolution poolings, each neuron output contains k rows and l columns of elements. Each neuron in the connection layer is convolved by g [k, l] convolution kernels on the elements in each neuron matrix. The specific formula is as follows:
[0075]
[0076] Among them, x k,l is the kth row and lth column element of the input data; w n is the convolution kernel weight, z n is the bias term of the convolution kernel; the size of the convolution kernel is the same as the size of the neuron matrix, and a value is output after convolution, y u Represents the value in each neuron of the fully connected layer, that is, the value obtained by adding g values after convolution.
[0077] In a specific embodiment, the output layer in S5 is a linear weighted summation calculation of the output vector of the connection layer, the number of input neurons is T, and the output is The specific formula is as follows:
[0078]
[0079] in, is the output volume prediction value, is the predicted value of fracturing fluid loss volume, is the predicted value of fracturing fluid flowback volume.
[0080] In a specific embodiment, the fracture reduction volume should be equal to the sum of the fracturing fluid flowback volume and the fracturing fluid loss volume, as shown below:
[0081]
[0082] Where ΔY(t) is the crack reduction volume at time t, m 3 ; is the cumulative filtration volume of the fracturing fluid at time t, m 3 ; is the cumulative flowback volume at time t, m 3 The fracture volume reduction at time t, the cumulative filtration volume of the fracturing fluid, and the cumulative flowback volume of the fracturing fluid are all functions of the bottom hole pressure at time t during the drainage process. The bottom hole pressure at time t during the drainage process can be iteratively calculated to obtain the accurate bottom hole pressure at time t. At the same time, when the bottom hole pressure is equal to the fracture closure pressure, it can be concluded that the fracture has been closed. The calculation of the critical flow rate of the proppant backflow is a prior art and will not be described in detail here.
[0083] In a specific embodiment, it also includes updating weight parameters of the fracturing fluid prediction model using back propagation technology, designing independent adaptive learning rates for different parameters by calculating the first-order moment estimate and second-order moment estimate of the gradient, updating the weights and bias of the convolutional layer, completing parameter fine-tuning, and minimizing the output error.
[0084] In a specific embodiment, updating the weight parameters using the back propagation technique specifically includes:
[0085] Determine the back propagation learning parameters: the back propagation algorithm learning rate is α, and the first-order moment estimation attenuation exponent β is determined at the same time 1 , second-order moment estimate decay exponent β 2 , parameter ε 1 , time step t, objective function f(θ), where β 1 and β 2 The value range is [0,1], where the θ parameter is a vector containing the forward training weights and biases;
[0086] Determine the parameters α, β 1 , β 2 After the objective function f(θ) is given, the first-order moment vector m=0, the second-order moment vector v=0 and the time step t=0 are initialized;
[0087] Calculate the first-order moment estimate and second-order moment estimate of the forecast indicator error gradient and make corrections;
[0088] Using the gradient of the previous moment The exponential decay correction value and the square gradient of the previous moment The exponential decay correction value of θ is used to update the θ parameter until the error is less than the set threshold.
[0089] In a specific embodiment, Figure 3 As shown, the diameter prediction model specifically includes: input layer, convolution layer 1, convolution layer 2, convolution layer 3, three BN layers, three Relu layers, pooling layer 1, pooling layer 2, fully connected layer, and output layer. The structure of the diameter prediction model is input layer, convolution layer 1, BN layer, Relu layer, pooling layer 1, convolution layer 2, BN layer, Relu layer, pooling layer 2, convolution layer 3, BN layer, Relu layer, fully connected layer, and output layer.
[0090] The function of the BN layer is batch normalization, and the output layer uses the Softmax activation function to directly output the predicted nozzle diameter.
[0091] In a specific embodiment, the parameters of each layer of the diameter prediction model are shown in Table 1:
[0092] Table 1 Parameters of each layer of the diameter prediction model
[0093] Network layer name Convolution kernel size Number of convolution kernels Fill value Convolution kernel sliding distance Network layer output dimensions Convolutional layer 1 5x5 6 2 1 26x26x6 Pooling layer 1 2x2 N N 2 13x13x6 Convolutional Layer 2 5x5 16 0 1 9x9x16 Pooling layer 2 2x2 N N 2 5x5x16 Convolutional layer 3 2x2 120 0 1 2x2x120
[0094] In a specific embodiment, calculating the first-order moment estimate and the second-order moment estimate of the prediction index error gradient and performing correction specifically includes:
[0095] When the parameter θ does not converge, first calculate the gradient at time t, and the calculation formula is as follows:
[0096]
[0097] Then calculate the first moment estimate m of the gradient t , the calculation formula is as follows:
[0098] m t =β 1 ·m t-1 +(1-β 1 )·b t ;
[0099] Calculate the second-order moment estimate of the gradient, the calculation formula is as follows:
[0100] v t =β 2 ·ν t-1 +(1-β 2 )·b t2 ;
[0101] The first-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows:
[0102]
[0103] The second-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows:
[0104]
[0105] In a specific embodiment, the obtained parameters are used to update the model parameter θ, and the calculation formula is as follows:
[0106]
[0107] Update each part iteratively, that is, add 1 to the time step t, update the gradient of the objective function for the parameter θ at this time step, and update the first-order moment estimate m of the deviation t and the second-order raw moment estimate v t , and then calculate the bias-corrected first-order moment estimate and bias-corrected second moment estimates Then use the calculated values to update the model parameters θ until the error is less than the set threshold.
[0108] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0109] 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 method for automatically optimizing the fracturing fluid release system after oil well fracturing, characterized in that: include: Fracturing the oil well, obtaining input variables related to fracturing fluid loss and flowback volume through sensors, normalizing the input variables, using the processed input variables as input data, and arranging them in time series to construct a variable time series input layer; Performing a convolution operation on the input data, and pooling the input data after the convolution operation, performing a full connection operation on the output data after multiple convolution and pooling, completing the training of the fracturing fluid prediction model, inputting the input variables to be predicted, and obtaining the fracturing fluid loss volume and the fracturing fluid return volume; Calculating the bottom hole pressure according to the fracturing fluid loss volume and the fracturing fluid flowback volume, and confirming the fracture condition through the bottom hole pressure and the fracture closure pressure; The filtration volume, flowback volume and viscosity of the fracturing fluid, as well as the casing pressure and oil pressure are converted into a two-dimensional image as the input of a preset diameter prediction model, the preset diameter prediction model is used to predict the nozzle diameter, and the flowback flow rate of the fracturing fluid is calculated based on the nozzle diameter; the critical reflux flow rate of the proppant is calculated based on the fracture conditions, and the nozzle model is selected based on the critical reflux flow rate and the flowback flow rate.
2. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 1, characterized in that: The input variables include Poisson's ratio, wellhead pressure, fracture height, reservoir thickness, fracturing fluid temperature, fluid level, and wellbore fluid column pressure.
3. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 1, characterized in that: The convolution operation on the input data specifically includes performing convolution operations on the input data in different directions, specifically: S1: Use t convolution kernels to perform vertical convolution on the input variable. The vertical convolution is calculated using the following formula: a i,j =f(w m x i,j +z m ),m=1,2,...,t; Among them, x i,j is the element in row i and column j of the input data; w m represents the convolution kernel weight, z m is the bias term of the convolution kernel; a i,j is the i-th row and j-th column element of the convolution data; f represents the activation function; S2: The input layer data is convolved by t convolution kernels and activated by the relu function. After activation, t neurons are output, each of which contains a data matrix. S3: Average pooling is performed on t neurons. The specific formula is as follows: Among them, q represents the size of the pooling area, D and F are the length and width of a single neuron data matrix. Since the pooling kernel is vertically pooled, only the length of the neuron matrix will decrease, p i / q,j Represents the i / qth row and jth column element of the output neuron matrix of the pooling layer; S4: The neurons output by the pooling layer are then horizontally convolved by g convolution kernels and average pooled again, outputting g neurons; S5: The neurons output after two convolution poolings are used as the input of the connection layer. The connection layer integrates the feature information represented by the elements of all neuron matrices into the neurons of the connection layer. Assume that the number of neurons in the connection layer is T. After multiple convolution poolings, each neuron output contains k rows and l columns of elements. Each neuron in the connection layer is convolved by g [k, l] convolution kernels on the elements in each neuron matrix. The specific formula is as follows: Among them, x k,l is the element in row k and column l of the input data; w n is the convolution kernel weight, z n is the bias term of the convolution kernel; the size of the convolution kernel is the same as the size of the neuron matrix, and a value is output after convolution, y u Represents the value in each neuron of the fully connected layer, that is, the value obtained by adding g values after convolution.
4. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 3, characterized in that: The output layer in S5 is a linear weighted summation calculation of the output vector of the connection layer, the number of input neurons is T, and the output is The specific formula is as follows: in, is the output volume prediction value, is the predicted value of fracturing fluid loss volume, is the predicted value of fracturing fluid flowback volume.
5. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 1, characterized in that: The method also includes updating weight parameters of the fracturing fluid prediction model using back propagation technology, designing independent adaptive learning rates for different parameters by calculating first-order moment estimation and second-order moment estimation of gradients, updating weights and biases of convolutional layers, completing parameter fine-tuning, and minimizing output errors.
6. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 5, characterized in that: The use of back propagation technology to update weight parameters specifically includes: Determine the back propagation learning parameters: the back propagation algorithm learning rate is α, and at the same time determine the first-order moment estimation decay index β1, the second-order moment estimation decay index β2, the parameter ε1, the time step t, and the objective function f(θ). The value range of β1 and β2 is [0,1], and the θ parameter is a vector containing the forward training weights and biases. After determining the parameters α, β1, β2 and the given objective function f(θ), initialize the first-order moment vector m=0, the second-order moment vector v=0 and the time step t=0; Calculate the first-order moment estimate and second-order moment estimate of the forecast indicator error gradient and make corrections; Using the gradient of the previous moment The exponential decay correction value and the square gradient of the previous moment The exponential decay correction value of θ is used to update the θ parameter until the error is less than the set threshold.
7. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 1, characterized in that: The diameter prediction model specifically includes: input layer, convolution layer 1, convolution layer 2, convolution layer 3, three BN layers, three Relu layers, pooling layer 1, pooling layer 2, fully connected layer, and output layer. The structure of the diameter prediction model is input layer, convolution layer 1, BN layer, Relu layer, pooling layer 1, convolution layer 2, BN layer, Relu layer, pooling layer 2, convolution layer 3, BN layer, Relu layer, fully connected layer, and output layer.
8. The method for automatically optimizing the fracturing fluid release system after oil well fracturing according to claim 6, characterized in that: The first-order moment estimation and the second-order moment estimation of the error gradient of the prediction index are calculated and corrected, and specifically include: When the parameter θ does not converge, first calculate the gradient at time t, and the calculation formula is as follows: Then calculate the first moment estimate m of the gradient t , the calculation formula is as follows: m t =β1·m t-1 +(1-β1)·b t ; Calculate the second-order moment estimate of the gradient, the calculation formula is as follows: v t =β2·ν t-1 +(1-β2)·b t 2 ; The first-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows: The second-order moment estimate is corrected, and the corrected moment estimate is The calculation formula is as follows: