Fracturing sand plugging risk intelligent early warning method with sustainable training function

Abstract

The invention discloses a fracturing sand plugging risk intelligent early warning method with a sustainable training function, and the method comprises the following steps: obtaining historical fracturing construction parameter data; establishing a fracturing construction pressure prediction data set and a fracturing construction sand plugging risk identification data set; establishing a data preprocessing mode; designing an AttBiLSTM network structure, and establishing a fracturing construction pressure real-time prediction model; designing a CNN-LSTM network structure, and establishing a fracturing sand plugging risk automatic identification model; the fracturing construction pressure real-time prediction model and the fracturing sand plugging risk automatic identification model are trained, and the weight of the trained construction pressure real-time prediction model and the weight of the trained fracturing sand plugging risk automatic identification model are obtained; and the fracturing construction pressure real-time prediction model and the fracturing sand plugging risk automatic identification model are coupled, a fracturing sand plugging risk early warning model is established, and intelligent early warning of the fracturing sand plugging risk is achieved. According to the method, two neural network models are adopted, the fracturing sand plugging risk early warning method is established, and the sand plugging risk early warning capacity in the fracturing construction process is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent optimized fracturing technology for oil reservoirs, and in particular relates to an intelligent early warning method for fracturing sand plugging risk with sustainable training function. Background Technology

[0002] In unconventional oil and gas reservoir fracturing operations, fracturing technology is widely used to improve reservoir permeability. However, due to the complexity of reservoir conditions and the variability of the fracturing site, the fracturing process faces the risk of sand plugging, which may negatively impact the reservoir and production, and increase additional construction costs. Therefore, accurate identification and early warning of sand plugging risks during fracturing are crucial for unconventional oil and gas fracturing production.

[0003] Currently, traditional methods for sand blockage risk early warning mainly rely on the double logarithmic curve method, net pressure analysis, and human experience. However, these methods are time-consuming, inefficient, and easily affected by human factors, lacking intelligent and automated features. Therefore, to improve the accuracy and efficiency of sand blockage risk early warning, it is necessary to introduce intelligent and automated methods. Summary of the Invention

[0004] The problem this invention aims to solve is to provide an intelligent early warning method for fracturing sand plugging risk with sustainable training capabilities. This method builds a predictive model based on real-time monitoring data and historical data, thereby quickly and accurately identifying and warning of sand plugging risk, reducing sand plugging risk and improving production efficiency, bringing significant progress and competitive advantages to the unconventional oil and gas industry.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an intelligent early warning method for fracturing sand plugging risk with sustainable training function, comprising the following steps,

[0006] S1: Obtain historical fracturing operation parameter data;

[0007] S2: Establish a dataset for predicting fracturing construction pressure and a dataset for identifying sand plugging risks during fracturing construction;

[0008] S3: Establish preprocessing methods for the fracturing construction pressure prediction dataset and the fracturing construction sand plugging risk identification dataset, respectively;

[0009] S4: Design the AttBiLSTM network structure, and establish a real-time prediction model for fracturing operation pressure based on the AttBiLSTM network structure;

[0010] S5: Design a CNN-LSTM network structure, and establish an automatic identification model for fracturing sand plugging risk based on the CNN-LSTM network structure;

[0011] S6: Train the real-time prediction model of fracturing construction pressure and the automatic identification model of fracturing sand plugging risk to obtain the weights of the trained real-time prediction model of construction pressure and the weights of the automatic identification model of fracturing sand plugging risk.

[0012] S7: Couple the real-time prediction model of fracturing construction pressure and the automatic identification model of fracturing sand plugging risk to establish a fracturing sand plugging risk early warning model and realize intelligent early warning of fracturing sand plugging risk.

[0013] Furthermore, in S1, the fracturing construction parameter data is acquired using a fracturing construction data acquisition system. The fracturing construction parameter data includes time, construction pressure, discharge volume, and sand ratio.

[0014] Furthermore, step S2 includes the following steps:

[0015] S21: The fracturing operation parameter data is smoothed using a mean filter to reduce data noise. The mean filter expression is as follows:

[0016]

[0017] In the formula, n represents the filter size; x j This represents the j-th number in parameter x; x ′ n+j-1 It is the new value in parameter x; N represents the number of data points in a fracturing data segment;

[0018] S22: Based on the following formulas, calculate the mean, deviation, and average rate of change of the original parameters, and calculate the total sand ratio at the current moment to obtain all extended parameters.

[0019]

[0020]

[0021] In the formula, i is the index number; t1 is the initial time of the fracturing operation; t i Let be the time step at the i-th moment of the fracturing operation; j be the calculation step size; N be the number of data points in a fracturing data segment; f(t1) be the parameter data at the initial moment; f(t) be the parameter data at the initial moment. i ) for t i Parameter data at any given time; This represents the average value of the parameters at the first i time points; d(t) represents the average rate of change of the parameter over a certain period of time. i ) represents the deviation, i.e., t i The difference between the time parameter data and the average value of the parameters at the previous i time points.

[0022] Furthermore, step S3 includes the following steps:

[0023] S31: Read all parameter data in the dataset, parse and decompose the time column, add the hours, minutes, and seconds as new three columns to the read data, and delete the original time column;

[0024] S32: Based on the fracturing operation time, the sliding window moves in the direction of time growth, and data is sampled using a sampling sliding window. All sampled data are randomly divided into training set and test set according to the proportion.

[0025] Furthermore, S4 includes the following steps:

[0026] S41: The input data of the real-time prediction model for fracturing construction pressure undergoes a bidirectional long short-term memory network (BiLSTM) neural network.

[0027] S42: Use the output of BiLSTM as the input to the Attention module of the attention mechanism;

[0028] S43: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the AttBiLSTM model;

[0029] S44: Real-time prediction of stress output by a fully connected neural network.

[0030] Furthermore, S5 includes the following steps:

[0031] S51: The input data of the automatic identification model for fracturing sand plugging risk is processed by a convolutional neural network (CNN) to achieve feature extraction;

[0032] S52: Use the output of the CNN as the input of the Long Short-Term Memory (LSTM) network;

[0033] S53: Use the output of the LSTM as the input to the Attention module of the attention mechanism;

[0034] S54: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the CNN-LSTM model;

[0035] S55: The result of data recognition after passing through a fully connected neural network.

[0036] Furthermore, step S6 includes the following steps:

[0037] S61: Train the real-time prediction model of fracturing construction pressure until the set maximum number of iterations is reached to obtain the trained real-time prediction model of fracturing construction pressure.

[0038] S62: Train the automatic identification model for fracturing sand plugging until the set maximum number of iterations is reached to obtain the trained automatic identification model for fracturing sand plugging.

[0039] Furthermore, S61 includes the following steps:

[0040] S611: Select training hyperparameters;

[0041] S612: Pre-train the real-time prediction model for fracturing construction pressure;

[0042] S613: Perform formal training on the real-time prediction model for fracturing operation pressure;

[0043] S614: Save the weights of the real-time prediction model for the optimal fracturing construction pressure.

[0044] Furthermore, S62 includes the following steps:

[0045] S621: Select training hyperparameters;

[0046] S622: Pre-train the automatic identification model for fracturing sand plugging;

[0047] S623: Perform formal training on the automatic fracturing sand plugging identification model;

[0048] S624: Save the weights of the optimal automatic identification model for fracturing sand plugging.

[0049] Furthermore, S7 includes the following steps:

[0050] S71: Input the fracturing operation parameter data of the current time step collected in real time into the trained real-time prediction model of fracturing operation pressure to obtain the prediction data of future operation pressure.

[0051] S72: Based on future construction pressure data and the original construction data already obtained, a fracturing sand plugging automatic identification model is coupled to realize intelligent early warning of fracturing sand plugging risk.

[0052] The advantages and positive effects of this invention are:

[0053] 1. This invention employs two neural network models to establish an intelligent early warning method for sand plugging risk in fracturing, significantly improving the early warning capability for sand plugging risk during fracturing operations. This invention utilizes time-series data prediction methods to establish a real-time prediction model for the fracturing process's operating pressure, and combines data classification technology and time data characteristics to establish an automatic identification model for fracturing sand plugging risk. By coupling the real-time operating pressure prediction model and the automatic sand plugging risk identification model, intelligent early warning of fracturing sand plugging risk is achieved.

[0054] 2. This invention utilizes deep learning to provide an efficient and accurate method for intelligent early warning of sand plugging risk in fracturing. Compared to traditional double logarithmic curve methods, net pressure analysis methods, manual identification methods, and other neural network fracturing identification models, this invention eliminates the interference of human factors, significantly reduces the computation time of numerical models, and has the advantages of automatic data identification and efficient early warning of sand plugging risk.

[0055] 3. In establishing the real-time prediction model for construction pressure and the automatic identification model for sand plugging risk, this invention employs a two-step training method and applies transfer learning technology to accelerate the convergence speed of the neural network model. When new fracturing construction data is available, the weights of the existing model can be loaded as the initial model weights, and then a small amount of training can be performed to obtain a new and more reliable neural network model.

[0056] 4. This invention employs a mean filter to preprocess the raw fracturing construction data to reduce noise interference. By calculating the rate of change, mean, deviation, and absolute value of deviation of data at different time steps, the hidden features of data such as construction pressure, discharge rate, and sand ratio are enhanced, and the temporal relationship between different data is strengthened, thereby improving the feature extraction capability of the neural network structure.

[0057] 5. The intelligent early warning method of the present invention is of great significance for the identification and early warning of sand plugging risk in unconventional oil and gas reservoir fracturing operations, and provides an efficient and accurate solution by introducing deep learning and data processing technology. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating an intelligent early warning method for fracturing sand plugging risk based on the dynamic changes in time-series data, as described in this embodiment of the invention.

[0059] Figure 2 This is a flowchart illustrating the construction of a real-time prediction model for fracturing construction pressure in an embodiment of the present invention.

[0060] Figure 3 This is a flowchart illustrating the construction of an automatic identification model for fracturing sand plugging in an embodiment of the present invention.

[0061] Figure 4 This is a graph showing the real-time prediction results of fracturing construction pressure from 15:36:00 to 15:37:38 in a specific example of the present invention.

[0062] Figure 5 The image shows the real-time prediction results of fracturing construction pressure from 15:49:41 to 15:51:07 in a specific example of the present invention.

[0063] Figure 6 The image shows the optimized result of the sand blockage event early warning in a specific example of the present invention at time 15:18:21.

[0064] Figure 7 The image shows the optimized result of the sand blockage event early warning in a specific example of the present invention at time 15:20:32. Detailed Implementation

[0065] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0067] like Figure 1 As shown, a smart early warning method for fracturing sand plugging risk with sustainable training function includes the following steps.

[0068] S1: Obtain historical fracturing operation parameter data, including time, operation pressure, discharge rate, and sand ratio. Specifically, use the fracturing operation data acquisition system to obtain fracturing operation parameter data, including time, operation pressure, discharge rate, and sand ratio for each fracturing operation parameter, and save it in an Excel file.

[0069] S2: Establish a dataset for predicting fracturing pressure and a dataset for identifying sand plugging risks during fracturing. Specifically, S2 includes the following steps:

[0070] S21: Read the construction pressure, discharge rate, and sand ratio data from the Excel file, and smooth these three types of fracturing construction parameter data using a mean filter of size 5 to reduce data noise. The mean filter expression is:

[0071]

[0072] Where n represents the filter size, x j Let x represent the j-th number of parameter x. ′ n+j-1 It is a new value in parameter x, and N represents the number of data points in a fracturing data segment.

[0073] S22: Based on the smoothed construction pressure, discharge volume, and sand ratio parameters, according to formulas (2)-(4), the mean, deviation, and average rate of change of the original parameters are calculated using calculation step sizes of 20 and 30 respectively. The total sand ratio at the current moment is then calculated and divided by 10000 to obtain all extended parameters. Specifically, the original parameters and extended parameters include the following: time (hour:minute:second), construction pressure, average rate of change of construction pressure over 20 steps, average rate of change of construction pressure over 30 steps, mean of construction pressure at the current moment, deviation of construction pressure at the current moment, absolute value of deviation of construction pressure at the current moment, discharge volume, average rate of change of discharge volume over 20 steps, average rate of change of discharge volume over 30 steps, sand ratio, average rate of change of sand ratio over 20 steps, average rate of change of sand ratio over 30 steps, mean of sand ratio at the current moment, deviation of sand ratio at the current moment, absolute value of deviation of sand ratio at the current moment, total sand ratio at the current moment / 10000.

[0074]

[0075] Where i is the index number; t1 is the initial time of the fracturing operation; t i Let be the time step at the i-th moment of the fracturing operation; j be the calculation step size; N be the number of data points in a fracturing data segment; f(t1) be the parameter data at the initial moment; f(t) be the parameter data at the initial moment. i ) for t i Parameter data at any given time; This represents the average value of the parameters at the first i time points; d(t) represents the average rate of change of the parameter over a certain period of time. i ) represents the deviation, i.e., t i The difference between the time parameter data and the average value of the parameters at the previous i time points.

[0076] Specifically, the parameters included in the fracturing construction pressure prediction dataset are shown in Table 1, and the parameters included in the fracturing construction sand plugging identification dataset are shown in Table 2.

[0077] Table 1. Predicted Pressure Data Parameters for Fracturing Operations

[0078]

[0079]

[0080] Table 2 Data Parameters for Sand Plug Identification During Fracturing Operations

[0081]

[0082] S3: Preprocessing methods for establishing the fracturing operation pressure prediction dataset and the fracturing operation sand plugging risk identification dataset. Specifically, S3 includes the following steps:

[0083] S31: Read all parameter data in the dataset, parse and decompose the time column, add the hours, minutes and seconds as three new columns to the read data, and delete the original time column.

[0084] S32: Based on the fracturing operation time, the sliding window moves in the direction of time growth. Data is sampled using a sampling sliding window of size 60. All sampled data are randomly divided into training set and test set in a ratio of 9:1.

[0085] S4: Design the AttBiLSTM network structure and establish a real-time prediction model for fracturing operation pressure based on the AttBiLSTM network structure. Specifically, S4 includes the following steps:

[0086] S41: The input data size of the AttBiLSTM model network for real-time prediction of fracturing construction pressure is (60, 19). The input data first goes through a bidirectional long short-term memory network BiLSTM neural network.

[0087] S42: Use the output of BiLSTM as the input to the Attention module of the attention mechanism.

[0088] S43: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the AttBiLSTM model.

[0089] S44: Finally, the real-time prediction result of the pressure is output through a fully connected neural network. The prediction result size is (40, 1), that is, the future construction pressure data of the next 40 seconds is predicted using the data of 60 seconds.

[0090] S5: Design a CNN-LSTM network structure and establish an automatic identification model for fracturing sand plugging risk based on the CNN-LSTM network structure. Specifically, S5 includes the following steps:

[0091] S51: The input data size of the CNN-LSTM model network for automatic identification of fracturing sand plugs is (60, 12). The input data first goes through a convolutional neural network (CNN) to extract features.

[0092] S52: Use the output of the CNN as the input of the Long Short-Term Memory (LSTM) network.

[0093] S53: Use the output of the LSTM as the input to the Attention module of the attention mechanism.

[0094] S54: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the CNN-LSTM model.

[0095] S55: The output data recognition result is obtained through a fully connected neural network. The output data size is (1, 1), which means that the data of 60 seconds is automatically identified as sand blockage data or normal data.

[0096] S6: Train the real-time prediction model for fracturing operation pressure and the automatic identification model for fracturing sand plugging risk based on the training and test sets, and obtain the weights of the trained real-time prediction model for fracturing operation pressure and the automatic identification model for fracturing sand plugging risk. Specifically, S6 includes the following steps:

[0097] S61: Train the real-time prediction model for fracturing operation pressure until the set maximum number of iterations is reached, resulting in a well-trained real-time prediction model for fracturing operation pressure. Specifically, S61 includes the following steps:

[0098] S611: Select training hyperparameters: Input data dimension is 60×19, number of iterations is 2000, number of data samples captured in 1 training is 2950, ​​optimizer is Adam optimizer, maximum learning rate is 0.0001, minimum learning rate is 0.000001, and learning rate change method is COS function learning rate value.

[0099] S612: Pre-training of real-time pressure prediction model for fracturing operations: Randomly select a portion of fracturing operation data from all fracturing operation pressure prediction datasets to pre-train the real-time pressure prediction model and obtain the optimal real-time pressure prediction model weights from the partial dataset.

[0100] S613: Formal Training of the Real-Time Prediction Model for Fracturing Operation Pressure: Utilizing transfer learning, the optimal network model weights obtained from a portion of the dataset are loaded into the initial AttBiLSTM network model. The real-time prediction model for fracturing operation pressure is trained using the entire fracturing operation dataset, effectively accelerating the training efficiency of the AttBiLSTM network. Furthermore, fracturing operation data from new wells can be continuously added without requiring retraining. Figure 2 As shown.

[0101] S614: Save the weights of the real-time prediction model for optimal fracturing pressure: After reaching the set maximum number of iterations, save the weights of the real-time prediction model for optimal fracturing pressure.

[0102] S62: Train the automatic fracturing sand plugging identification model until the set maximum number of iterations is reached, resulting in a well-trained automatic fracturing sand plugging identification model. Specifically, S62 includes the following steps:

[0103] S621: Select training hyperparameters: Input data dimension is 60×12, number of iterations is 500, number of data samples captured in 1 training is 50, optimizer is Adam optimizer, maximum learning rate is 0.0001, minimum learning rate is 0.000001, and learning rate changes in the COS function learning rate value mode.

[0104] S622: Pre-training of automatic fracturing sand plug identification model: Randomly select a portion of fracturing sand plug identification data from all fracturing construction sand plug identification datasets to pre-train the automatic fracturing sand plug identification model, and obtain the optimal automatic fracturing sand plug identification model weights from the partial dataset.

[0105] S623: Formal Training of the Automatic Sand Blockage Recognition Model for Fracturing: Utilizing transfer learning, the optimal network model weights obtained from a portion of the dataset are loaded into the initial CNN-LSTM network model. The automatic sand blockage recognition model for fracturing is trained using the entire fracturing operation sand blockage recognition dataset, effectively accelerating the training efficiency of the CNN-LSTM network. Furthermore, fracturing operation data from new wells can be continuously added without requiring retraining. Figure 3 As shown.

[0106] S624: Save the weights of the optimal automatic fracturing sand plugging identification model: After reaching the set maximum number of iterations, save the weights of the optimal automatic fracturing sand plugging identification model.

[0107] S7: Couple the real-time prediction model of construction pressure and the automatic identification model of fracturing sand plugging risk to establish a fracturing sand plugging risk early warning model, thereby achieving intelligent early warning of fracturing sand plugging risk. Specifically, S7 includes the following steps:

[0108] S71: Input the fracturing construction parameter data of the current time step collected in real time into the trained real-time construction pressure prediction model to obtain the future construction pressure prediction data.

[0109] S72: Based on future construction pressure data and the obtained original construction data, a fracturing sand plugging automatic identification model is coupled to realize intelligent early warning of fracturing sand plugging risk.

[0110] The present invention will now be described in detail with reference to specific embodiments.

[0111] The above model and method were applied to the early warning of sand plugging risk in offshore fracturing operations. The real-time prediction results of the construction pressure during the fracturing operation of Well No. 1 (which was free of sand plugging) are shown below. Figure 4 and Figure 5As shown, compared with actual construction pressure data, the construction pressure implementation prediction model has a high fitting degree of 95%, indicating that the established AttBiLSTM construction pressure implementation prediction model has reliable practical value. In the later stage of fracturing construction of well No. 2, the pressure rise was significant and the fluctuations were abnormal, exhibiting obvious sand plugging characteristics. A fracturing sand plugging risk early warning model was used to provide sand plugging early warning. Figure 6 and Figure 7 The real-time dynamic early warning results show that the risk of sand blockage can be effectively predicted during the real-time prediction stage of construction pressure, effectively improving the early warning time by more than 30 seconds. Therefore, the established sand blockage risk early warning model can effectively provide early warning of sand blockage.

[0112] In summary, based on Figure 4 , Figure 5 , Figure 6 and Figure 7 The comparison and results show that the model has high accuracy and predictive ability in early warning of sand plugging risks during offshore fracturing operations. This method can provide early warning of sand plugging risks, offering effective risk monitoring and prevention measures for the fracturing process to ensure its safety and smooth progress.

[0113] The advantages and positive effects of this invention are:

[0114] 1. This invention employs two neural network models to establish an intelligent early warning method for sand plugging risk in fracturing, significantly improving the early warning capability for sand plugging risk during fracturing operations. This invention utilizes time-series data prediction methods to establish a real-time prediction model for the fracturing process's operating pressure, and combines data classification technology and time data characteristics to establish an automatic identification model for fracturing sand plugging risk. By coupling the real-time operating pressure prediction model and the automatic sand plugging risk identification model, intelligent early warning of fracturing sand plugging risk is achieved.

[0115] 2. This invention utilizes deep learning to provide an efficient and accurate method for intelligent early warning of sand plugging risk in fracturing. Compared to traditional double logarithmic curve methods, net pressure analysis methods, manual identification methods, and other neural network fracturing identification models, this invention eliminates the interference of human factors, significantly reduces the computation time of numerical models, and has the advantages of automatic data identification and efficient early warning of sand plugging risk.

[0116] 3. In establishing the real-time prediction model for construction pressure and the automatic identification model for sand plugging risk, this invention employs a two-step training method and applies transfer learning technology to accelerate the convergence speed of the neural network model. When new fracturing construction data is available, the weights of the existing model can be loaded as the initial model weights, and then a small amount of training can be performed to obtain a new and more reliable neural network model.

[0117] 4. This invention employs a mean filter to preprocess the raw fracturing construction data to reduce noise interference. By calculating the rate of change, mean, deviation, and absolute value of deviation of data at different time steps, the hidden features of data such as construction pressure, discharge rate, and sand ratio are enhanced, and the temporal relationship between different data is strengthened, thereby improving the feature extraction capability of the neural network structure.

[0118] 5. The intelligent early warning method of the present invention is of great significance for the identification and early warning of sand plugging risk in unconventional oil and gas reservoir fracturing operations, and provides an efficient and accurate solution by introducing deep learning and data processing technology.

[0119] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A smart early warning method for fracturing sand plugging risk with sustainable training function, characterized in that: Includes the following steps, S1: Obtain historical fracturing operation parameter data; S2: Establish a dataset for predicting fracturing construction pressure and a dataset for identifying sand plugging risks during fracturing construction; S3: Establish preprocessing methods for the fracturing construction pressure prediction dataset and the fracturing construction sand plugging risk identification dataset, respectively; S4: Design the AttBiLSTM network structure, and establish a real-time prediction model for fracturing operation pressure based on the AttBiLSTM network structure; S5: Design a CNN-LSTM network structure, and establish an automatic identification model for fracturing sand plugging risk based on the CNN-LSTM network structure; S6: Train the real-time prediction model of fracturing construction pressure and the automatic identification model of fracturing sand plugging risk to obtain the weights of the trained real-time prediction model of construction pressure and the weights of the automatic identification model of fracturing sand plugging risk. S7: Couple the real-time prediction model of fracturing construction pressure and the automatic identification model of fracturing sand plugging risk to establish a fracturing sand plugging risk early warning model and realize intelligent early warning of fracturing sand plugging risk.

2. The intelligent early warning method for fracturing sand plugging risk with sustainable training function according to claim 1, characterized in that: In step S1, the fracturing construction parameter data is acquired using a fracturing construction data acquisition system. The fracturing construction parameter data includes time, construction pressure, discharge volume, and sand ratio.

3. The intelligent early warning method for fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S2 includes the following steps: S21: The fracturing operation parameter data is smoothed using a mean filter to reduce data noise. The mean filter expression is as follows: In the formula, n represents the filter size; x j This represents the j-th number in parameter x; x ′ n+j-1 It is the new value in parameter x; N represents the number of data points in a fracturing data segment; S22: Based on the following formulas, calculate the mean, deviation, and average rate of change of the original parameters, and calculate the total sand ratio at the current moment to obtain all extended parameters. In the formula, i is the index number; t1 is the initial time of the fracturing operation; t i Let be the time step at the i-th moment of the fracturing operation; j be the calculation step size; N be the number of data points in a fracturing data segment; f(t1) be the parameter data at the initial moment; f(t) be the parameter data at the initial moment. i ) for t i Parameter data at any given time; This represents the average value of the parameters at the first i time points; d(t) represents the average rate of change of the parameter over a certain period of time. i ) represents the deviation, i.e., t i The difference between the time parameter data and the average value of the parameters at the previous i time points.

4. The intelligent early warning method for fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S3 includes the following steps: S31: Read all parameter data in the dataset, parse and decompose the time column, add the hours, minutes, and seconds as new three columns to the read data, and delete the original time column; S32: Based on the fracturing operation time, the sliding window moves in the direction of time growth, and data is sampled using a sampling sliding window. All sampled data are randomly divided into training set and test set according to the proportion.

5. A method for intelligent early warning of fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S4 includes the following steps: S41: The input data of the real-time prediction model for fracturing construction pressure undergoes a bidirectional long short-term memory network (BiLSTM) neural network. S42: Use the output of BiLSTM as the input to the Attention module of the attention mechanism; S43: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the AttBiLSTM model; S44: Real-time prediction of stress output by a fully connected neural network.

6. A method for intelligent early warning of fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S5 includes the following steps: S51: The input data of the automatic identification model for fracturing sand plugging risk is processed by a convolutional neural network (CNN) to achieve feature extraction; S52: Use the output of the CNN as the input of the Long Short-Term Memory (LSTM) network; S53: Use the output of the LSTM as the input to the Attention module of the attention mechanism; S54: The output of the Attention module is passed through the hyperbolic tangent Tanh activation function to introduce non-linearity, thereby increasing the expressive power of the CNN-LSTM model; S55: The result of data recognition after passing through a fully connected neural network.

7. A method for intelligent early warning of fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S6 includes the following steps: S61: Train the real-time prediction model of fracturing construction pressure until the set maximum number of iterations is reached to obtain the trained real-time prediction model of fracturing construction pressure. S62: Train the automatic identification model for fracturing sand plugging until the set maximum number of iterations is reached to obtain the trained automatic identification model for fracturing sand plugging.

8. The intelligent early warning method for fracturing sand plugging risk with sustainable training function according to claim 7, characterized in that: S61 includes the following steps: S611: Select training hyperparameters; S612: Pre-train the real-time prediction model for fracturing construction pressure; S613: Perform formal training on the real-time prediction model for fracturing operation pressure; S614: Save the weights of the real-time prediction model for the optimal fracturing construction pressure.

9. The intelligent early warning method for fracturing sand plugging risk with sustainable training function according to claim 7, characterized in that: S62 includes the following steps: S621: Select training hyperparameters; S622: Pre-train the automatic identification model for fracturing sand plugging; S623: Perform formal training on the automatic fracturing sand plugging identification model; S624: Save the weights of the optimal automatic identification model for fracturing sand plugging.

10. A method for intelligent early warning of fracturing sand plugging risk with sustainable training function according to claim 1 or 2, characterized in that: S7 includes the following steps: S71: Input the fracturing operation parameter data of the current time step collected in real time into the trained real-time prediction model of fracturing operation pressure to obtain the prediction data of future operation pressure. S72: Based on future construction pressure data and the original construction data already obtained, a fracturing sand plugging automatic identification model is coupled to realize intelligent early warning of fracturing sand plugging risk.

Images