Multi-task learning coal seam horizontal well fracturing scheme optimization method

Abstract

The invention discloses a multitask learning coal seam horizontal well fracturing scheme optimization method, which comprises the following steps: collecting and counting parameter data of a research area, and preprocessing the parameter data to obtain target data; performing feature extraction on the target data, and randomly dividing a data set into a training set, a verification set and a test set according to a certain proportion by using the extracted feature data; creating a BP neural network model through a newff function, and training the BP neural network model based on the training set to obtain a target BP neural network model; and according to the target BP neural network model, a particle swarm optimization algorithm is adopted to optimize fracturing parameters of the coal bed gas horizontal well, and an optimization result is obtained. According to the method, geology, fracturing, logging and production data are comprehensively considered, the generalization ability and learning performance of the neural network are improved by using multiple methods, prediction precision can be improved, and the method has great significance in improving and optimizing fracturing parameters of the horizontal well.

Description

Technical Field

[0001] The invention belongs to the technical field of coalbed methane mining, and in particular relates to a multi-task learning method for optimizing coalbed horizontal well fracturing schemes. Background Art

[0002] As a key technology to improve the efficiency and safety of coalbed methane mining, the optimization of fracturing parameters for horizontal wells of coalbed methane is of great significance to further release the potential of coal resources and optimize the energy structure in my country. At present, my country's coalbed methane industry faces problems such as complex geological conditions and strong reservoir heterogeneity. However, the fracturing designs under different geological units are often similar. The traditional fracturing construction scheme has been difficult to meet the production of coalbed methane horizontal wells in my country. Machine learning methods have been widely used in coalbed methane well capacity prediction and fracturing parameter optimization. However, the optimization design problem of fracturing parameters for horizontal wells of coalbed methane has complex variable types, high variable dimensions, mutual connections between variables, and many constraints. Single machine learning may fall into local optimality. As the decision variables increase, the algorithm performance also decreases. Therefore, it is particularly important to integrate reservoir geology, fracturing and production factors to build a multi-task artificial intelligence agent model to optimize fracturing parameter design.

[0003] Patents with application numbers 202211572501.8 and 202111581397.4 respectively disclose a method for optimizing horizontal well fracturing parameters for shale gas and shale oil, but both only optimize the fracturing parameters from the perspective of fracturing construction. Application number 202311407780.7 discloses a hierarchical multi-level optimization method for horizontal well fracturing parameters in a shale reservoir well factory, which takes into account geological and fracturing construction parameters, but the geological parameters used are parameters after logging interpretation, ignoring the authenticity of the logging data itself. The differential evolution optimization algorithm used is difficult to perform multi-task learning parameter optimization. The above-mentioned comparative patents all have certain limitations.

[0004] In summary, the current optimization of horizontal well fracturing parameters has obvious shortcomings. First, the geological parameters after logging interpretation are often used for subsequent research, which reduces the authenticity of the logging curve itself and is easily disturbed by uncertain factors such as formula and parameter selection during the interpretation process; second, the current coalbed methane horizontal well production fracturing parameter design has not yet fully integrated reservoir geological conditions, fracturing construction parameters, production data and original logging data; third, there is a lack of a multi-task artificial intelligence agent model that can integrate the above parameters and has a targeted, rapid and accurate average gas production, maximum gas production and long-term gas production capacity of horizontal wells. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a multi-task learning method for optimizing coal seam horizontal well fracturing scheme, comprising:

[0006] Collect and count parameter data of the study area, and pre-process the parameter data to obtain target data;

[0007] Extract features from the target data, and randomly divide the data set into a training set, a validation set, and a test set according to a certain ratio using the extracted feature data;

[0008] Create a BP neural network model through the newff function, train the BP neural network model based on the training set, and obtain a target BP neural network model;

[0009] According to the target BP neural network model, a particle swarm optimization algorithm is used to optimize the fracturing parameters of coalbed methane horizontal wells to obtain optimization results.

[0010] Preferably, the parameter data of the study area include geological parameters, fracturing construction parameters, coal seam section logging parameters, construction parameters and production data;

[0011] The geological parameters include coal seam depth, coal thickness, and interlayer thickness;

[0012] The fracturing construction parameters include the total construction fluid volume of the horizontal well, the total sand volume, the construction displacement of the sand-carrying fluid, the proppant particle size, the amount of different particle sizes, the horizontal section length, and the number of fracturing sections;

[0013] The coal seam section logging parameters include AC, DEN, GR, RD, RS, and SP;

[0014] The production data include average gas production, cumulative gas production within 30 days of peak gas production, and cumulative gas production within 2 years.

[0015] Preferably, the process of preprocessing the parameter data includes:

[0016] Using the box plot method, find out the abnormal points of the parameter data in the study area and eliminate the abnormal data;

[0017] The Pearson correlation coefficients among all parameters of the horizontal wells were calculated, the reservoir parameters were sorted according to the strength of the correlation, and the parameters with the same correlation were filtered out, and only one of them was retained.

[0018] Preferably, the formula for calculating the Pearson correlation coefficient between all parameters of the horizontal well is:

[0019]

[0020] Where r is the Pearson correlation coefficient, x i and i are the sample values ​​of the two variables, and is the sample mean, and n is the sample size.

[0021] Preferably, the process of extracting features from the target data includes:

[0022] Extract features of geological parameters, fracturing operation parameters and production data including mean, median, maximum, minimum and variance;

[0023] For the logging parameters of coal seam sections, the deep convolution automatic encoder-decoder of MATLAB is used to perform nonlinear dimensionality reduction on the logging depth time data series and output characteristic parameters.

[0024] Preferably, the process of randomly dividing the extracted feature data into a training set, a validation set and a test set according to a certain ratio includes:

[0025] The data set was divided into five subsets using five-fold cross validation. One subset was selected as the test set each time without duplication, and the remaining four subsets were used as training sets. The training was repeated five times.

[0026] At the same time, the mean absolute error and root mean square error are used to evaluate the accuracy of the model, wherein the smaller the values ​​of the mean absolute error and root mean square error are, the smaller the gap between the true value and the predicted value is, and the better the accuracy of the model prediction result is;

[0027] The mean absolute error and the root mean square error are respectively:

[0028]

[0029]

[0030] In the formula, n represents the number of samples, represents the predicted value, Represents the true value.

[0031] Preferably, the process of creating a BP neural network model by the newff function includes:

[0032] Based on the newff function, a column of data is regarded as a sample, with m rows and n columns in total, and the topological structure includes an input layer, a hidden layer and an output;

[0033] Wherein, the input layer is geological parameters, fracturing construction parameters and coal seam section logging parameters;

[0034] The output layer is horizontal well production data;

[0035] The hidden layer adopts the trial and error method, uses tansig and purelin functions as the activation functions of the hidden layer and the output layer respectively, and uses the Levenberg-Marquardt method for training. The initial weights and thresholds of each neural unit in the model are randomly assigned, the maximum number of training times is 5000, the expected training error is 0.001, and the learning rate is selected as 0.01;

[0036] Wherein, the formula expression of the newff function is:

[0037] net=newff(data,label,[m,n],{'tansig','purelin'},'trainlm')

[0038] The formula for determining the number of nodes in the hidden layer is:

[0039]

[0040] In the formula, p represents the number of input layer nodes, q represents the number of output layer nodes, and a is a constant between 1 and 10;

[0041] The formula expressions of the tansig and purelin functions are respectively:

[0042]

[0043] purelin(x)=x.

[0044] Preferably, before optimizing the coalbed methane horizontal well fracturing parameters using the particle swarm optimization algorithm, the method further comprises:

[0045] According to the target BP neural network model, the importance of geological parameters, fracturing operation parameters and coal seam section logging parameters is evaluated to obtain the main control factors of the average gas production, maximum gas production and long-term gas production capacity of the horizontal well;

[0046] Multi-task learning is performed by minimizing the loss training model, and the losses of multiple tasks are optimized for prediction at the same time; wherein, the minimization loss training model includes three elements: feature extraction, task sharing and loss function; first, multi-task learning extracts effective features of geology, logging, fracturing and production, and then the multi-task joint features are passed to the cross network module, and the multi-task sharing structure and sharing degree are automatically determined through learning; finally, the loss weight is measured using the same variance uncertainty, the two tasks are partially differentiated, end-to-end learning is achieved, the loss function Lt is calculated, and finally a multi-task loss function is established.

[0047] Preferably, according to the target BP neural network model, the process of optimizing the fracturing parameters of coalbed methane horizontal wells by using a particle swarm optimization algorithm includes:

[0048] Read the parameter data of the study area as learning sample data and initialize the position of each particle and speed

[0049] Calculate the fitness value of each particle and find the individual and global optimal values;

[0050] Update the speed and position of each particle according to the speed and position update formula;

[0051] Determine whether the particle has reached the maximum number of iterations or the global optimal position meets the minimum limit. If so, output the optimal result. Otherwise, continue to initialize the position of each particle. and speed

[0052] According to the output optimal results, the main controlling factors of the average production capacity, long-term production capacity and maximum production capacity of horizontal wells controlled by geology, fracturing construction parameters and logging parameters during the fracturing construction and production of coalbed methane horizontal wells are determined. At the same time, according to the actual fracturing construction plan, geological conditions and logging parameters of coalbed methane horizontal wells today, the multi-task learning coalbed methane horizontal well fracturing parameter design of average production capacity, long-term production capacity and maximum production capacity is optimized.

[0053] Preferably, the speed and position of each particle are updated according to the speed and position update formula, and the calculation formula for the speed and position of each particle is expressed as:

[0054]

[0055]

[0056] in, represents the position of the current particle i at the nth iteration on dimension d, represents the updated position of particle i at the n+1th iteration on dimension d, The velocity of the particle i before and after in the nth iteration on dimension d, represents the updated velocity of particle i at the t+1th iteration on dimension d, ω represents the inertia matrix, which indicates the degree to which the particle continues to move forward in the search space, and c 1 and c 2 are learning factors, representing the influence weights of individuals and groups respectively, represents the optimal solution of particle i in dimension d, represents the global optimal solution of the entire group, r 1 and r 2 They represent the Rand function, which represents a random number between [0,1], and r 1 and r2 The numbers represented are not the same.

[0057] Compared with the prior art, the present invention has the following advantages and technical effects:

[0058] The method of the present invention comprehensively integrates horizontal well logging curves, geology, fracturing and coalbed methane production data, uses convolutional neural networks to extract eigenvalues, is based on BP neural networks, assisted by five-fold cross validation, embeds multi-task shared digital modules, and uses particle swarm optimization algorithms to optimize coalbed methane horizontal well fracturing parameters. The invention emphasizes the authenticity of logging data and adopts multi-type data constraints. The constructed algorithm can find the global optimal solution more easily in the search process compared with traditional geological methods and single artificial intelligence methods, improves the generalization ability and learning performance of neural networks, helps to improve prediction accuracy, and is of great significance to improving and optimizing horizontal well fracturing parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0060] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0061] Figure 2 A flowchart of confirming the training set, validation set, and test set by five-fold cross-validation in an embodiment of the present invention;

[0062] Figure 3 This is a flow chart of identifying the main control factors of coalbed methane horizontal well production control based on BP neural network according to an embodiment of the present invention;

[0063] Figure 4 A schematic diagram of a multi-task shared digital module according to an embodiment of the present invention;

[0064] Figure 5 This is a flow chart of particle swarm optimization of horizontal well fracturing parameters according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0066] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0067] like Figure 1-5 As shown, this embodiment provides a multi-task learning coal seam horizontal well fracturing scheme optimization method, including the following steps:

[0068] The parameter data of the study area are collected and counted, and the parameter data are preprocessed to obtain the target data; the target data are feature extracted, and the extracted feature data are randomly divided into training set, validation set and test set according to a certain ratio; the BP neural network model is created through the newff function, and the BP neural network model is trained based on the training set to obtain the target BP neural network model; according to the target BP neural network model, the particle swarm optimization algorithm is used to optimize the coalbed methane horizontal well fracturing parameters to obtain the optimization results.

[0069] More specifically, the following steps are included:

[0070] Step 1: Collect statistics on the geology, fracturing construction, coal seam logging curves and production data of the study area. The geological parameters include coal seam burial depth, coal thickness, and interlayer thickness; the fracturing construction parameters include the total construction fluid volume of horizontal wells, the total sand volume, the construction displacement of sand-carrying fluid, the proppant particle size, the amount of different particle sizes, the length of the horizontal section, and the number of fracturing sections; the coal seam logging parameters include AC, DEN, GR, RD, RS, and SP; the production data include the average gas production, the cumulative gas production within 30 days of the peak gas production (the highest gas production capacity), and the cumulative gas production within 2 years (long-term gas production capacity).

[0071] Step 2: In the data collection stage, variables such as geological parameters, logging data, construction parameters, and production data are considered. Considering that the actual data may have errors or mistakes during the collection process, the data is preprocessed before conducting relevant analysis work. The box plot method is used to find out the abnormal points and eliminate the abnormal data.

[0072] Step 3: Use formula (1) to calculate the Pearson correlation coefficient between all parameters of the horizontal well, sort the reservoir parameters according to the strength of the correlation, and filter out one of the two parameters with stronger correlation;

[0073]

[0074] Where r is the Pearson correlation coefficient, x i and i are the sample values ​​of the two variables, and is the sample mean, and n is the sample size.

[0075] Step 4: For geological parameters, fracturing construction parameters and production data, since the data are determined, the extracted features include mean, median, maximum, minimum and variance; for logging parameters, since it is a deep time data series, the deep convolution automatic encoder-decoder of Matlab is used to realize the nonlinear dimensionality reduction representation of the logging sequence and output the characteristic parameters.

[0076] Step 5: Randomly divide the data set into a training set, a validation set and a test set according to a certain ratio. The training set is used to train the machine learning model, the validation set verifies the training degree of the model, avoids overfitting and underfitting, and the test set evaluates the final effect of the model. The test set never participates in the training and optimization process of the model. For a small data set, the model may not perform well on a randomly selected test set. In this embodiment, a five-fold cross validation is used to divide the data set. The data set is divided into five subsets. One subset is selected as the test set each time without repetition. The remaining four subsets are used as training sets. The training is repeated five times, so that all samples have the opportunity to participate in model training and testing to solve the above problem. The mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the accuracy of the model. The smaller the MAE and RMSE values ​​are, the smaller the gap between the true value and the predicted value is, and the accuracy of the model prediction result is better. As shown in formula (2)-formula (3), the optimal partitioning model is constructed.

[0077]

[0078] In the formula, n represents the number of samples, represents the predicted value, Represents the true value.

[0079] Step 6: Based on the division of the above data set, use the newff function to create a BP neural network, see formula (4). The newff function will treat a column of data as a sample (m rows and n columns). Its topological structure includes input layer, hidden layer and output layer. The input layer is geological, fracturing construction and logging parameters, and the output layer is horizontal well production data. The hidden layer adopts the trial and error method. The number of nodes in the hidden layer is determined by formula (5). The tansig and purelin functions are shown in formulas (6)-(7) as the activation functions of the hidden layer and output layer respectively. The Levenberg-Marquardt method is used for training. The initial weights and thresholds of each neural unit in the model are randomly assigned. The maximum number of training times is 5000, the expected training error is 0.001, and the learning rate is selected as 0.01.

[0080] net=newff(data,label,[m,n],{'tansig','purelin'},'trainlm') (4)

[0081]

[0082] In the formula, p represents the number of input layer nodes, q represents the number of output layer nodes, and a is a constant between 1 and 10.

[0083]

[0084] purelin(x)=x(7)

[0085] Step 7: Finally, the importance of geological, logging and fracturing characteristic parameters is evaluated based on the BP neural network to obtain the main control factors of the average gas production, maximum gas production and long-term gas production capacity of horizontal wells.

[0086] Step 8: Multi-task learning accurately and effectively predicts by optimizing the losses of multiple tasks at the same time. The model consists of three elements: feature extraction, task sharing and loss function. First, multi-task learning extracts effective features of geology, logging, fracturing and production. Then, the multi-task joint features are passed to the cross network module, and the multi-task sharing structure and sharing degree are automatically determined through learning, see formula (8)-formula (9); finally, the loss weight is measured using the same variance uncertainty, and formula (10)-formula (12) is used as an auxiliary variation formula to partially differentiate the two tasks to achieve end-to-end learning, calculate the loss function Lt, and finally establish a multi-task loss function, see formula (13), and obtain the minimum loss training model:

[0087]

[0088] in and They are the characteristic graphs of average capacity, maximum capacity and long-term capacity at (i, j) respectively; α represents the cross unit, and the smaller its value is, the lower the degree of sharing between tasks is. When α = 0, the layer at (i, j) is a task-specific layer. The value of α can determine the segmentation structure and sharing degree of the network; and They are the feature maps of average capacity, maximum capacity and long-term capacity after cross-unit learning, L t is the total loss function of the multi-task model, L i and w i are the loss function and weight of the i-th task respectively, t is the number of tasks, and σ is the standard deviation of the normal distribution.

[0089] Step 9: The principle of the particle swarm optimization algorithm is to simulate the position and speed changes of particles in the multidimensional search space, and gradually approach the optimal position with the fusion of evolutionary information of individual learning and social learning to achieve the global optimal solution. The whole process involves the fitness change of each particle in the population, and also considers the impact of the global optimal solution. Through iteration, particles adjust their speed and position according to their own state and surrounding environment, and finally achieve the optimal solution to the problem. The calculation formula for the speed and position of each particle is as follows:

[0090]

[0091] in, represents the position of the current particle i at the nth iteration on dimension d, represents the updated position of particle i at the n+1th iteration on dimension d, The velocity of the particle i before and after in the nth iteration on dimension d, represents the updated velocity of particle i at the t+1th iteration on dimension d, ω represents the inertia matrix, which indicates the degree to which the particle continues to move forward in the search space, and c 1 and c 2 are learning factors, representing the influence weights of individuals and groups respectively, represents the optimal solution of particle i in dimension d, represents the global optimal solution of the entire group, r 1 and r 2 They represent the Rand function, which represents a random number between [0,1], and r 1 and r 2 The numbers represented are not the same.

[0092] The specific particle swarm optimization algorithm schematic process is as follows: Figure 5 As shown:

[0093] (1) Select the structure of the BP neural network, including the training set, validation set, test set, etc.;

[0094] (2) Reading learning sample data, including geological parameters such as coal seam burial depth, coal thickness, and interlayer thickness; fracturing operation parameters including total operation fluid volume of horizontal wells, total sand volume, sand-carrying fluid operation displacement, proppant particle size, different particle size dosage, horizontal section length, and number of fracturing sections; coal seam section logging parameters including AC, DEN, GR, RD, RS, and SP; production data including average gas production, cumulative gas production within 30 days of peak gas production (maximum gas production capacity), and cumulative gas production within 2 years (long-term gas production capacity);

[0095] (3) Initialize the position of each particle and speed

[0096] (4) Calculate the fitness value of each particle and find the individual and global optimal values;

[0097] (5) Update the speed and position of each particle according to the speed and position update formula;

[0098] (6) Determine whether the particle has reached the maximum number of iterations or the global optimal position meets the minimum limit;

[0099] If (7) is yes, output the optimal result; otherwise, jump to (3).

[0100] Step 10: Finally, according to the above steps, the main controlling factors of the average production capacity, long-term production capacity and maximum production capacity of horizontal wells controlled by geology, fracturing construction parameters and logging parameters in the fracturing construction and production process of coalbed methane horizontal wells can be determined. At the same time, according to the actual fracturing construction plan, geological conditions and logging parameters of coalbed methane horizontal wells today, the multi-task learning fracturing parameter design of average production capacity, long-term production capacity and maximum production capacity is optimized. Taking a horizontal well as an example, the optimized fracturing construction plan: average gas production is 2034.21 cubic meters, the peak 30-day cumulative gas production is 169,800 cubic meters, and the cumulative production in 2 years is 1.3432 million cubic meters. Compared with the actual fracturing construction plan, the average gas production is increased by 36.5%, the peak 30-day cumulative gas production is increased by 18.9%, and the cumulative gas production in 2 years is increased by 22.7%. See Table 1-Table 2 for details.

[0101] Table 1

[0102]

[0103] Table 2

[0104]

[0105]

[0106] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A multi-task learning method for optimizing coal seam horizontal well fracturing scheme, characterized in that: include: Collect and count parameter data of the study area, and pre-process the parameter data to obtain target data; Extract features from the target data, and randomly divide the data set into a training set, a validation set, and a test set according to a certain ratio using the extracted feature data; Create a BP neural network model through the newff function, train the BP neural network model based on the training set, and obtain a target BP neural network model; According to the target BP neural network model, a particle swarm optimization algorithm is used to optimize the fracturing parameters of coalbed methane horizontal wells to obtain optimization results.

2. The method according to claim 1, characterized in that The parameter data of the study area include geological parameters, fracturing operation parameters, coal seam section logging parameters, operation parameters and production data; The geological parameters include coal seam depth, coal thickness, and interlayer thickness; The fracturing construction parameters include the total construction fluid volume of the horizontal well, the total sand volume, the construction displacement of the sand-carrying fluid, the proppant particle size, the amount of different particle sizes, the horizontal section length, and the number of fracturing sections; The coal seam section logging parameters include AC, DEN, GR, RD, RS, and SP; The production data include average gas production, cumulative gas production within 30 days of peak gas production, and cumulative gas production within 2 years.

3. The method according to claim 1, characterized in that The process of preprocessing the parameter data includes: Using the box plot method, find out the abnormal points of the parameter data in the study area and eliminate the abnormal data; The Pearson correlation coefficients among all parameters of the horizontal wells were calculated, the reservoir parameters were sorted according to the strength of the correlation, and the parameters with the same correlation were filtered out, and only one of them was retained.

4. The method according to claim 3, characterized in that The formula for calculating the Pearson correlation coefficient between all parameters of horizontal wells is: Where r is the Pearson correlation coefficient, x i and i are the sample values ​​of the two variables, and is the sample mean, and n is the sample size.

5. The method according to claim 1, characterized in that The process of extracting features from the target data includes: Extract features of geological parameters, fracturing operation parameters and production data including mean, median, maximum, minimum and variance; For the logging parameters of coal seam sections, the deep convolution automatic encoder-decoder of MATLAB is used to perform nonlinear dimensionality reduction on the logging depth time data series and output characteristic parameters.

6. The method according to claim 1, characterized in that The process of randomly dividing the extracted feature data into training set, validation set and test set according to a certain ratio includes: The data set was divided into five subsets using five-fold cross validation. One subset was selected as the test set each time without duplication, and the remaining four subsets were used as training sets. The training was repeated five times. At the same time, the mean absolute error and root mean square error are used to evaluate the accuracy of the model, wherein the smaller the values ​​of the mean absolute error and root mean square error are, the smaller the gap between the true value and the predicted value is, and the better the accuracy of the model prediction result is; The mean absolute error and the root mean square error are respectively expressed as: In the formula, n represents the number of samples, represents the predicted value, Represents the true value.

7. The method according to claim 1, characterized in that The process of creating a BP neural network model through the newff function includes: Based on the newff function, a column of data is regarded as a sample, with m rows and n columns in total, and the topological structure includes an input layer, a hidden layer and an output layer; Wherein, the input layer is geological parameters, fracturing construction parameters and coal seam section logging parameters; The output layer is horizontal well production data; The hidden layer adopts the trial and error method, uses tansig and purelin functions as the activation functions of the hidden layer and the output layer respectively, and uses the Levenberg-Marquardt method for training. The initial weights and thresholds of each neural unit in the model are randomly assigned, the maximum number of training times is 5000, the expected training error is 0.001, and the learning rate is selected as 0.01; Wherein, the formula expression of the newff function is: net=newff(data,label,[m,n],{'tansig','purelin'},'trainlm') The formula for determining the number of nodes in the hidden layer is: In the formula, p represents the number of input layer nodes, q represents the number of output layer nodes, and a is a constant between 1 and 10; The formula expressions of the tansig and purelin functions are respectively: purelin(x)=x.

8. The method according to claim 1, characterized in that Before optimizing the coalbed methane horizontal well fracturing parameters using the particle swarm optimization algorithm, the method further includes: According to the target BP neural network model, the importance of geological parameters, fracturing operation parameters and coal seam section logging parameters is evaluated to obtain the main control factors of the average gas production, maximum gas production and long-term gas production capacity of the horizontal well; Multi-task learning is performed by minimizing the loss training model, and the losses of multiple tasks are optimized for prediction. The minimization loss training model includes three elements: feature extraction, task sharing and loss function. First, multi-task learning extracts effective features of geology, logging, fracturing and production. Then, the multi-task joint features are passed to the cross network module, and the multi-task sharing structure and sharing degree are automatically determined through learning. Finally, the loss weight is measured using the same variance uncertainty, the two tasks are partially differentiated, end-to-end learning is achieved, the loss function Lt is calculated, and finally a multi-task loss function is established.

9. The method according to claim 1, characterized in that: According to the target BP neural network model, the process of optimizing the fracturing parameters of coalbed methane horizontal wells using the particle swarm optimization algorithm includes: Read the parameter data of the study area as learning sample data and initialize the position of each particle and speed Calculate the fitness value of each particle and find the individual and global optimal values; Update the speed and position of each particle according to the speed and position update formula; Determine whether the particle has reached the maximum number of iterations or the global optimal position meets the minimum limit. If so, output the optimal result. Otherwise, continue to initialize the position of each particle. and speed According to the output optimal results, the main controlling factors of the average production capacity, long-term production capacity and maximum production capacity of horizontal wells controlled by geology, fracturing construction parameters and logging parameters during the fracturing construction and production of coalbed methane horizontal wells are determined. At the same time, according to the actual fracturing construction plan, geological conditions and logging parameters of coalbed methane horizontal wells today, the multi-task learning coalbed methane horizontal well fracturing parameter design of average production capacity, long-term production capacity and maximum production capacity is optimized.

10. The method according to claim 9, characterized in that Update the speed and position of each particle according to the speed and position update formula. The calculation formula for the speed and position of each particle is: in, represents the position of the current particle i at the nth iteration on dimension d, represents the updated position of particle i at the n+1th iteration on dimension d, The velocity of the particle i before and after in the nth iteration on dimension d, represents the speed of the updated particle i at the t+1th iteration on dimension d, ω represents the inertia matrix, which indicates the degree to which the particle continues to move forward in the search space, c1 and c2 are learning factors, representing the influence weights of individuals and groups respectively, represents the optimal solution of particle i in dimension d, represents the global optimal solution of the entire group, r1 and r2 represent the Rand function respectively, the Rand function represents a random number between [0,1], and the numbers represented by r1 and r2 are different.

Images