Method and device for predicting water scouring and breaking of soft coal seam, equipment and storage medium
By training a fracturing prediction model that combines source domain, auxiliary domain, and target domain data, the scale effect and working condition differences in the field application of the fracturing prediction model for water scour in soft coal seams were resolved, achieving higher prediction accuracy and reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TAIYUAN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-05-13
- Publication Date
- 2026-07-21
AI Technical Summary
In existing technologies, the prediction model for water scouring and fracturing of soft coal seams based on laboratory data suffers from insufficient accuracy due to scale effects and differences in working conditions when applied to field drilling operations, making it difficult to meet actual engineering needs.
By acquiring source, auxiliary, and target domain data, an initial rupture prediction model is trained. The model is then gradually optimized through a combination of pre-training and adversarial training to adapt it to the actual scale. Pseudo-labels and correction factors are used for further optimization to improve prediction accuracy.
It improves the accuracy and reliability of predicting water scouring and fracturing in soft coal seams, making the model closer to the actual field conditions, adapting to the fracturing patterns of different scale scenarios, and enhancing the accuracy and reliability of the prediction results.
Smart Images
Figure CN122197733B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of coal mine engineering safety and risk prediction technology, specifically relating to a method, device, equipment, and storage medium for predicting water erosion and fracturing in soft coal seams. Background Technology
[0002] In coal mine gas control, geological exploration, and drainage drilling, water injection drilling or hydraulically assisted drilling is often used for soft and fractured coal seams. This involves continuously injecting water into the borehole to remove slag, cool the drill bit, and stabilize the drilling process. However, soft and fractured coal seams have a loose structure, low cementation, and weak shear and erosion resistance. Under the continuous scouring action of high-velocity, high-dynamic-pressure water flow within the borehole, the coal seam is highly susceptible to localized spalling, structural damage, and even extended fractures. Therefore, predicting the fracture risk of soft and fractured coal seams under borehole water injection scouring conditions is of significant engineering importance for optimizing drilling parameters and ensuring construction safety.
[0003] Current research on predicting water erosion-induced fracturing in soft coal seams typically involves constructing predictive models and incorporating coal mechanics parameters, water flow parameters, and construction parameters for risk assessment. However, the data relied upon for training existing models mainly comes from small-scale specimen tests under laboratory conditions or limited field experience data. In laboratory studies, limitations imposed by specimen size, boundary conditions, and the form of water flow make it difficult to accurately reflect the flow characteristics of the injected water flow within the borehole and its coupling mechanism with the coal structure during drilling operations.
[0004] Due to the scale effect and differences in working conditions mentioned above, the prediction models established based on laboratory data often have insufficient adaptability when applied to field drilling construction. The accuracy and reliability of the prediction results are difficult to meet the actual needs of the project, thus limiting their effective application in drilling parameter optimization and construction risk prevention and control. Summary of the Invention
[0005] The purpose of this application is to provide a method, apparatus, equipment, and storage medium for predicting water scour and fracturing in soft coal seams, in order to solve the problem of low accuracy in fracturing prediction caused by scale effects and differences in operating conditions.
[0006] A first aspect of this application provides a method for predicting water scour and fracturing in soft coal seams, comprising:
[0007] Obtain the target coal sample parameters and target hydrodynamic parameters in the target fractured and soft coal seam;
[0008] Input the target coal sample parameters and target hydrodynamic parameters into the target fracturing prediction model to obtain the fracturing prediction results of the target soft coal seam;
[0009] The training process of the target rupture prediction model includes:
[0010] Acquire source domain data, auxiliary domain data, and target domain data. The source domain data includes multiple sets of laboratory data, each set including experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracturing labels. The auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set including auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracturing labels. The target domain data includes multiple sets of target soft coal seam data, each set including historical coal sample parameters and historical hydrodynamic parameters. The similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data must exceed a preset threshold.
[0011] An initial rupture prediction model is trained based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model that is adapted to the actual scale.
[0012] Input the target domain data into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data;
[0013] The intermediate rupture prediction model is trained based on auxiliary domain data, target domain data, and pseudo-labels to obtain the target rupture prediction model.
[0014] A second aspect of this application provides a device for predicting water scour and fracturing in soft coal seams, comprising:
[0015] The data acquisition module is used to acquire target coal sample parameters and target hydrodynamic parameters in the target broken and soft coal seam;
[0016] The fracture prediction module is used to input the target coal sample parameters and target hydrodynamic parameters into the target fracture prediction model to obtain the fracture prediction results of the target fractured soft coal seam.
[0017] The training process of the target rupture prediction model includes:
[0018] Acquire source domain data, auxiliary domain data, and target domain data. The source domain data includes multiple sets of laboratory data, each set including experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracturing labels. The auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set including auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracturing labels. The target domain data includes multiple sets of target soft coal seam data, each set including historical coal sample parameters and historical hydrodynamic parameters. The similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data must exceed a preset threshold.
[0019] An initial rupture prediction model is trained based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model that is adapted to the actual scale.
[0020] Input the target domain data into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data;
[0021] The intermediate rupture prediction model is trained based on auxiliary domain data, target domain data, and pseudo-labels to obtain the target rupture prediction model.
[0022] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting the fracturing of soft coal seams by water scouring.
[0023] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the fracturing of soft coal seams by water scouring.
[0024] The beneficial effects of the method, apparatus, equipment, and storage medium for predicting water scour and fracturing in soft coal seams provided in this application are as follows:
[0025] This application embodiment acquires source domain data (multiple sets of laboratory data), auxiliary domain data (multiple sets of auxiliary soft coal seam data with a similarity greater than a preset threshold to the target soft coal seam), and target domain data (multiple sets of historical data of the target soft coal seam). Based on the source domain data and auxiliary domain data, an initial fracturing prediction model is trained to obtain an intermediate fracturing prediction model adapted to the actual scale. This considers the correlation between data at different scales and the impact of working condition differences. The auxiliary domain data compensates for the scale difference between laboratory data and actual field data, solving the problems of scale effects and working condition differences. This makes the trained model closer to the actual field situation, thereby improving the accuracy of predicting water scour fracturing in soft coal seams. Secondly, this application embodiment also trains the intermediate fracturing prediction model based on auxiliary domain data, target domain data, and pseudo-labels to obtain a target fracturing prediction model. Utilizing information from the target domain data, the model is further optimized through pseudo-labels, enabling the model to better adapt to the characteristics of the target soft coal seam, further enhancing the accuracy and reliability of the prediction results. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 This is a flowchart illustrating a method for predicting the fracturing of soft coal seams by water scouring according to an embodiment of this application.
[0028] Figure 2 A flowchart illustrating the training process of a target fracture prediction model provided in an embodiment of this application;
[0029] Figure 3 This is a structural block diagram of a water scour and fracture prediction device for soft coal seams provided in an embodiment of this application;
[0030] Figure 4 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0031] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0032] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.
[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.
[0034] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting water scour and fracturing in soft coal seams according to an embodiment of this application. The method can be executed by an electronic device and may include: S101-S102.
[0035] S101: Obtain the target coal sample parameters and target hydrodynamic parameters in the target broken and soft coal seam.
[0036] In this embodiment, the target fractured soft coal seam refers to a specific coal seam requiring fracture prediction, i.e., a coal seam for which the risk of water scour fracture needs to be assessed in detail to avoid disasters. The target coal sample refers to a coal body sample collected from the aforementioned target fractured soft coal seam that represents the overall characteristics of the coal seam. Target coal sample parameters refer to parameters characterizing the specific properties of the target coal body sample, such as particle size distribution, porosity, natural moisture content, and permeability coefficient. Target hydrodynamic parameters refer to parameters related to water scour in the actual field conditions of the target fractured soft coal seam, such as water velocity, flow rate, and head difference.
[0037] S102: Input the target coal sample parameters and target hydrodynamic parameters into the target fracturing prediction model to obtain the fracturing prediction results of the target soft coal seam.
[0038] In this embodiment, the target rupture prediction model refers to the prediction model after training, which is used to eliminate the scale effect caused by the small-scale sample training data in the laboratory and the actual large-scale (greater than the first preset scale) application scenario on site.
[0039] Specifically, refer to Figure 2 The training process of the target rupture prediction model includes: S201-S204.
[0040] S201: Obtain source domain data, auxiliary domain data, and target domain data.
[0041] In this embodiment, the source domain data includes multiple sets of laboratory data, each set of laboratory data including: experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracture tags; the auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set of auxiliary soft coal seam data including: auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracture tags; the target domain data includes multiple sets of target soft coal seam data, each set of target soft coal seam data including: historical coal sample parameters and historical hydrodynamic parameters of the target soft coal seam; the similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data is greater than a preset threshold.
[0042] In this embodiment, a domain-adaptive method is selected to solve the aforementioned scale problem. Specifically, the source domain data refers to controllable data (small scale) smaller than the second preset scale obtained in a laboratory environment. Its role can be understood as providing the basic correlation between coal body properties and hydrodynamic effects and whether or not rupture occurs.
[0043] In this embodiment, the auxiliary domain data is actual field data, but not the actual field data of the target fractured soft coal seam. Instead, it is actual data of other fractured soft coal seams that have a high similarity to the target fractured soft coal seam data. The auxiliary domain data can be obtained based on publicly available datasets. The preset threshold can be determined empirically to ensure that the core characteristics (such as coal and rock type and pore structure) of the auxiliary domain coal seam are sufficiently close to those of the target domain coal seam, avoiding the model learning invalid patterns. The role of the auxiliary domain data is to bridge the differences between the small-scale laboratory and the large-scale field, compensate for the shortcomings of the source domain data in being detached from field conditions, and provide fracture patterns under field conditions.
[0044] In this embodiment, the target domain data refers to the historical data collected from the target broken soft coal seam on-site. Since the target broken soft coal seam is a protected object, the tag data cannot be obtained through actual fracturing. Therefore, each group only contains the target coal sample parameters and the target hydrodynamic parameters, without fracturing tags.
[0045] S202: Train an initial rupture prediction model based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model that is adapted to the actual scale.
[0046] In this embodiment, the initial rupture prediction model refers to the basic model that has not been adapted to field data. The initial rupture prediction model includes: a feature extractor, a task predictor, and a domain discriminator.
[0047] In one embodiment, the process of training an initial rupture prediction model may include:
[0048] The feature extractor and task predictor are pre-trained based on source domain data to obtain the initial rupture prediction model after pre-training.
[0049] The initial rupture prediction model after pre-training is trained adversarially based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model adapted to the actual scale; the loss function in the pre-training process is different from the loss function in the adversarial training process.
[0050] In this embodiment, the feature extractor can extract high-dimensional features reflecting the correlation between coal body properties and hydrodynamic effects from the input coal sample parameters and hydrodynamic parameters. The task predictor is the prediction output module of the initial fracture prediction model, and its function is to transform the high-dimensional features output by the feature extractor into specific fracture prediction results (such as fracture probability). The domain discriminator determines whether the high-dimensional features output by the feature extractor come from the source domain or the auxiliary domain. Pre-training refers to supervised training of the feature extractor and the task predictor using only source domain data (small-scale laboratory data) to enable the initial fracture prediction model to learn the basic correlation between coal sample parameters, hydrodynamic parameters, and fracture probability under laboratory conditions.
[0051] In this embodiment, adversarial training refers to training that simultaneously inputs source domain data and auxiliary domain data, allowing the feature extractor and domain discriminator to form an adversarial relationship. This enables the trained intermediate rupture prediction model to overcome the limitations of small-scale laboratory settings, learn to identify rupture patterns in large-scale field environments, and complete the transformation from a laboratory model to a field-adapted model.
[0052] In this embodiment, the core objective of adversarial training is to address the scaling effect and enable the feature extractor to learn domain-invariant features, which are general features that can reflect laboratory patterns and adapt to field conditions.
[0053] Training can be divided into two parallel computation processes. The first is the task prediction branch, which inputs features from the source and auxiliary domain data into the task predictor and calculates the error between the predicted result and the true rupture label to ensure the model's rupture prediction accuracy. The second is the domain adversarial branch, which inputs the features output from the feature extractor into the domain discriminator and calculates the domain discriminator's discrimination error against the feature source. This allows the feature extractor and the domain discriminator to compete against each other, thereby obtaining an intermediate rupture prediction model adapted to the actual scale.
[0054] In this embodiment, the goal of pre-training is to make the model more accurate in predicting on laboratory data. Therefore, a relatively simple task loss function can be used, focusing on the error between the prediction result and the true label. On the other hand, the goal of adversarial training is to make the model both accurate in predicting and adaptable across domains. Therefore, a joint loss function of task loss + domain adversarial loss can be used to ensure the accuracy of the fracture prediction while driving the feature extractor to learn domain-invariant features, thus achieving a leap from laboratory scale to field scale.
[0055] S203: Input the target domain data into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data.
[0056] In this embodiment, pseudo-labels refer to the prediction results of the intermediate fracture prediction model for the target domain data, specifically the fracture probability output by the model. These pseudo-labels are filtered by confidence level, with only high-confidence predictions retained as pseudo-labels. Essentially, they replace the true labels with the prediction results, transforming the originally unlabeled target domain data into labeled data that the model can continue training on.
[0057] In this embodiment, the intermediate rupture prediction model has been adapted to the field scale. Using this model to label the target domain data can ensure the rationality of the labels to the greatest extent. The confidence screening is to eliminate low-quality pseudo-labels and avoid interfering with subsequent training.
[0058] S204: Train the intermediate rupture prediction model based on auxiliary domain data, target domain data, and pseudo-labels to obtain the target rupture prediction model.
[0059] In this embodiment, through multiple rounds of iterative training until preset iteration conditions are met, such as reaching a preset number of iterations or the difference in the loss function after N consecutive iterations being less than a preset loss threshold, a target fracturing prediction model that can be used to predict the probability of fracturing of the target soft coal seam is finally obtained.
[0060] As can be seen from the above, the embodiments of this application train an initial fracture prediction model based on source domain data and auxiliary domain data. This allows the model to overcome the limitations of small-scale laboratory environments during adversarial training, learn to identify fracture patterns in large-scale field environments, and obtain an intermediate fracture prediction model adapted to actual scales. This effectively solves the problems of scale effects and differences in working conditions, enabling the model to better adapt to field conditions and providing a foundation for accurate prediction. In the initial fracture prediction model training process, a combination of pre-training and adversarial training is adopted. Pre-training uses only source domain data to supervise the feature extractor and task predictor, enabling the model to learn the basic correlation between coal sample parameters, hydrodynamic parameters, and fracture probability under laboratory conditions. Secondly, in adversarial training, both source domain data and auxiliary domain data are input simultaneously, creating an adversarial relationship between the feature extractor and the domain discriminator. A joint loss function of task loss and domain adversarial loss is used, ensuring fracture prediction accuracy while driving the feature extractor to learn domain-invariant features, achieving learning from laboratory scale to field scale. This phased training method gradually optimizes model performance, ensuring that the trained model possesses both accurate prediction capabilities and adaptability to different scale scenarios.
[0061] In one embodiment of this application, an adversarial training process is performed on the pre-trained initial fracture prediction model based on source domain data and auxiliary domain data to obtain an intermediate fracture prediction model adapted to the actual scale, including:
[0062] The source domain data and auxiliary domain data are input into the pre-trained feature extractor to obtain high-dimensional feature vectors;
[0063] The high-dimensional feature vector is input into the pre-trained task predictor, and the corresponding loss information is calculated based on the cross-entropy loss function.
[0064] The high-dimensional feature vector is input into the domain discriminator, and the corresponding loss information is calculated based on the binary cross-entropy loss function.
[0065] The loss information corresponding to the joint loss function is determined based on the loss information corresponding to the cross-entropy loss function and the loss information corresponding to the binary cross-entropy loss function.
[0066] The parameters of the feature extractor, task predictor, and domain discriminator are alternately updated based on the backpropagation algorithm so that the loss information corresponding to the joint loss function meets the preset conditions, thus obtaining the intermediate breakage prediction model.
[0067] In this embodiment, the cross-entropy loss function can be used to measure the error between the predicted break label output by the task predictor and the actual break label of the data. The smaller the error, the higher the break prediction accuracy. The binary cross-entropy loss function is the basis for calculating the domain adversarial loss. It is used to measure the error between the domain classification result output by the domain discriminator and the actual domain label (source domain / auxiliary domain) of the data. The smaller the error, the stronger the domain discriminator's ability to distinguish cross-scale data.
[0068] In this embodiment, the joint loss function can be a weighted combination of the cross-entropy loss function and the binary cross-entropy loss function. The weights can be set empirically, and the resulting loss information can be understood as the loss value. The optimization objectives of the feature extractor and the task predictor are to minimize the joint loss; the optimization objective of the domain discriminator is to maximize the domain adversarial loss. These three components alternately adjust their parameters during backpropagation, forming an adversarial relationship until a preset condition is reached, thus obtaining the intermediate breakdown prediction model. The preset condition is generally set as the joint loss value decreasing to a preset loss threshold, or the difference in loss values across multiple consecutive iterations being less than a preset difference threshold, indicating that the model training has converged.
[0069] As can be seen from the above, the feature extractor and task predictor in this embodiment aim to minimize the joint loss and continuously adjust their parameters during training, enabling the model to learn general features that reflect both laboratory patterns and adapt to field conditions. The domain discriminator aims to maximize the domain adversarial loss, forming an adversarial relationship with the feature extractor. This prompts the feature extractor to continuously optimize and extract more general domain-invariant features, thereby overcoming the limitations of small-scale laboratory settings and learning to identify fracture patterns in large-scale field environments. This allows the intermediate fracture prediction model to better adapt to real-world scenarios.
[0070] In one embodiment of this application, target domain data is input into an intermediate rupture prediction model to obtain pseudo-labels corresponding to the target domain data, including:
[0071] Input the target domain data into the intermediate fracture prediction model and obtain the fracture probability and confidence level corresponding to each set of target fractured soft coal seam data output by the task predictor.
[0072] Filter the target broken and soft coal seam data in each group of target broken and soft coal seam data with a confidence level greater than the preset confidence threshold to obtain standard target broken and soft coal seam data;
[0073] The fracturing probability corresponding to the standard target soft coal seam data is used as the pseudo-label corresponding to the target domain data.
[0074] In this embodiment, an intermediate fracturing prediction model adapted to the field scale is used to make a preliminary risk assessment on the unlabeled target domain data. At the same time, the confidence level corresponding to the prediction result is obtained. After the confidence level is filtered, the target domain data with reliable prediction results, that is, the standard target broken soft coal seam data, is retained, and the fracturing probability corresponding to the standard target broken soft coal seam data is used as the pseudo label corresponding to the target domain data.
[0075] In one embodiment of this application, the intermediate fracturing prediction model further includes a correction module; the correction module is connected to the output of the task predictor, and the parameters of the correction module include a correction factor; the correction factor is calculated based on the target coal sample parameters and the auxiliary coal sample parameters, and is used to correct the prediction results of the intermediate fracturing prediction model to adapt to the target soft coal seam.
[0076] The intermediate rupture prediction model is trained based on auxiliary domain data, target domain data, and pseudo-labels corresponding to the target domain data, resulting in the target rupture prediction model, including:
[0077] The auxiliary domain data, target domain data, and the pseudo-labels corresponding to the target domain data are mixed to obtain a hybrid training dataset;
[0078] Using the label loss of the mixed training dataset as the optimization objective, the intermediate rupture prediction model is trained in multiple rounds based on the mixed training dataset and the correction factor to obtain the target rupture prediction model;
[0079] After each iteration, the mixed training dataset is updated, and the correction factor is also updated.
[0080] In this embodiment, the correction module is a result correction module for the intermediate fracture prediction model. It is directly connected to the output of the task predictor. That is, after the task predictor outputs the initial fracture probability, the correction module will perform a secondary correction on the result before outputting the final prediction value. The parameters of the correction module include a correction factor, which is calculated based on the core differences between the target coal sample parameters and the auxiliary coal sample parameters (such as the relative difference rate of porosity, permeability coefficient, and particle size fractal dimension). The greater the difference, the greater the correction magnitude of the correction factor. Essentially, it quantifies the degree of deviation between the auxiliary domain rules and the target domain characteristics.
[0081] In this embodiment, the intermediate fracture prediction model only completed the scale adaptation between the small scale in the laboratory and the large scale in the field. However, there are still unique characteristics between the target fractured soft coal seam and the auxiliary fractured soft coal seam (such as the target fractured soft coal seam having a higher natural moisture content). The role of the correction factor is to make the model prediction results fit the actual working conditions of the target fractured soft coal seam through parameter correction, rather than simply applying the rules of the auxiliary fractured soft coal seam.
[0082] In this embodiment, label loss refers to the error between the prediction result (the result after correction factor) and the labels in the mixed dataset (true labels in the auxiliary domain and pseudo labels in the target domain). It can be calculated using the cross-entropy loss function. The smaller the loss, the higher the consistency between the prediction result and the label.
[0083] In one embodiment, updating the hybrid training dataset includes:
[0084] The target domain data prediction results output after the current iteration of training are filtered according to a preset confidence threshold to obtain updated high-confidence pseudo-labels.
[0085] The updated high-confidence pseudo-labels and their corresponding target domain data are then remixed with the auxiliary domain data to obtain the updated mixed training dataset.
[0086] In this embodiment, updating the hybrid training dataset means using the model trained in this iteration to re-predict the target domain data, eliminating low-quality pseudo-labels that may have existed in the previous iteration, selecting new high-confidence pseudo-labels, replacing the old pseudo-labels in the original dataset, and then mixing them with the auxiliary domain data to form a new training set. This allows the dataset to be updated synchronously with the model optimization, thus conforming to the true characteristics of the target fractured soft coal seam.
[0087] In one embodiment, updating the correction factor includes:
[0088] The relative difference rate is calculated based on the target domain data and auxiliary domain data in the updated mixed training dataset.
[0089] The updated correction factor is obtained based on the relative difference rate.
[0090] In this embodiment, the relative difference rate is an indicator that quantifies the degree of difference in parameters between the target fractured soft coal seam and the auxiliary fractured soft coal seam. Specifically, it can be the absolute difference between the target domain parameter value and the mean value of the auxiliary domain parameter, divided by the mean value of the auxiliary domain parameter. The larger the value, the more obvious the parameter difference between the two types of coal seams.
[0091] In this embodiment, since there are multiple parameters in both the target domain data and the auxiliary domain data, and each parameter corresponds to a relative difference rate, the average of the relative difference rates corresponding to each parameter in the target domain data and the auxiliary domain data can be taken to obtain the target relative difference rate. The correction factor is then updated based on the target relative difference rate. Specifically, the correction factor can be equal to 1 minus the target difference rate; that is, the closer the correction factor is to 1, the smaller the correction magnitude to the initial prediction result of the intermediate model.
[0092] As can be seen from the above, this embodiment utilizes an intermediate fracturing prediction model adapted to the field scale to process unlabeled target domain data, obtaining the fracturing probability and confidence level corresponding to each group of target soft coal seam data. By filtering data with a confidence level greater than a preset confidence threshold, standard target soft coal seam data is obtained, and its fracturing probability is used as a pseudo-label. This not only performs preliminary risk prediction on the target domain data but also retains reliable prediction data through confidence level filtering, effectively eliminating data that may have large errors, thus improving the accuracy and reliability of the pseudo-label and providing a high-quality data foundation for model training. The correction module in the intermediate fracturing prediction model is connected to the output of the task predictor, and its parameter correction factor is calculated based on the target coal sample parameters and auxiliary coal sample parameters. Due to the characteristic differences between the target soft coal seam and the auxiliary soft coal seam, such as the target soft coal seam having a higher natural moisture content, the correction factor can perform secondary correction on the prediction results of the intermediate fracturing prediction model, making the model prediction results fit the actual working conditions of the target soft coal seam, avoiding the simple application of the rules of the auxiliary soft coal seam, thereby improving the accuracy and reliability of the prediction results.
[0093] Corresponding to the water scouring and fracturing prediction method for soft coal seams in the above embodiments, Figure 3 This is a structural block diagram of a water scour and fracturing prediction device for soft coal seams provided in one embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 3 The water scouring and fracturing prediction device 30 for soft coal seams includes: a data acquisition module 31, a fracturing prediction module 32, and a prediction model training module 33.
[0094] Data acquisition module 31 is used to acquire target coal sample parameters and target hydrodynamic parameters in the target broken soft coal seam;
[0095] The fracture prediction module 32 is used to input the target coal sample parameters and target hydrodynamic parameters into the target fracture prediction model to obtain the fracture prediction results of the target fractured soft coal seam;
[0096] The prediction model training module 33 is used to acquire source domain data, auxiliary domain data, and target domain data during the training process of the target fracture prediction model. The source domain data includes multiple sets of laboratory data, each set including experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracture labels. The auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set including auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracture labels. The target domain data includes multiple sets of target soft coal seam data, each set including historical coal sample parameters and historical hydrodynamic parameters of the target soft coal seam. The similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data is greater than a preset threshold.
[0097] An initial rupture prediction model is trained based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model that is adapted to the actual scale.
[0098] Input the target domain data into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data;
[0099] The intermediate rupture prediction model is trained based on auxiliary domain data, target domain data, and pseudo-labels to obtain the target rupture prediction model.
[0100] In one embodiment of this application, the initial rupture prediction model includes: a feature extractor and a task predictor;
[0101] The prediction model training module 33 is specifically used to pre-train the feature extractor and the task predictor based on the source domain data to obtain the initial rupture prediction model after pre-training.
[0102] The initial rupture prediction model after pre-training is trained adversarially based on source domain data and auxiliary domain data to obtain an intermediate rupture prediction model adapted to the actual scale; the loss function in the pre-training process is different from the loss function in the adversarial training process.
[0103] In one embodiment of this application, the initial rupture prediction model further includes: a domain discriminator;
[0104] The prediction model training module 33 is specifically used to input source domain data and auxiliary domain data into the pre-trained feature extractor to obtain high-dimensional feature vectors.
[0105] The high-dimensional feature vector is input into the pre-trained task predictor, and the corresponding loss information is calculated based on the cross-entropy loss function.
[0106] The high-dimensional feature vector is input into the domain discriminator, and the corresponding loss information is calculated based on the binary cross-entropy loss function.
[0107] The loss information corresponding to the joint loss function is determined based on the loss information corresponding to the cross-entropy loss function and the loss information corresponding to the binary cross-entropy loss function.
[0108] The parameters of the feature extractor, task predictor, and domain discriminator are alternately updated based on the backpropagation algorithm so that the loss information corresponding to the joint loss function meets the preset conditions, thus obtaining the intermediate breakage prediction model.
[0109] In one embodiment of this application, the intermediate rupture prediction model includes a task predictor;
[0110] The prediction model training module 33 is specifically used to input the target domain data into the intermediate fracture prediction model and obtain the fracture probability and confidence level corresponding to each set of target fractured soft coal seam data output by the task predictor.
[0111] Filter the target broken and soft coal seam data in each group of target broken and soft coal seam data with a confidence level greater than the preset confidence threshold to obtain standard target broken and soft coal seam data;
[0112] The fracturing probability corresponding to the standard target soft coal seam data is used as the pseudo-label corresponding to the target domain data.
[0113] In one embodiment of this application, the intermediate fracturing prediction model further includes a correction module; the correction module is connected to the output of the task predictor, and the parameters of the correction module include a correction factor; the correction factor is calculated based on the target coal sample parameters and the auxiliary coal sample parameters, and is used to correct the prediction results of the intermediate fracturing prediction model to adapt to the target soft coal seam.
[0114] The prediction model training module 33 is specifically used to mix the auxiliary domain data, the target domain data, and the pseudo-labels corresponding to the target domain data to obtain a mixed training dataset.
[0115] Using the label loss of the mixed training dataset as the optimization objective, the intermediate rupture prediction model is trained in multiple rounds based on the mixed training dataset and the correction factor to obtain the target rupture prediction model;
[0116] After each iteration, the mixed training dataset is updated, and the correction factor is also updated.
[0117] In one embodiment of this application, the prediction model training module 33 is further configured to filter the target domain data prediction results output after the current iteration of training according to a preset confidence threshold to obtain updated high-confidence pseudo-labels.
[0118] The updated high-confidence pseudo-labels and their corresponding target domain data are then remixed with the auxiliary domain data to obtain the updated mixed training dataset.
[0119] In one embodiment of this application, the prediction model training module 33 is further used to calculate the relative difference rate based on the target domain data and auxiliary domain data in the updated mixed training dataset;
[0120] The updated correction factor is obtained based on the relative difference rate.
[0121] See Figure 4 , Figure 4 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 4The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of each module / unit in the above-described device embodiments, for example... Figure 3 The functions of the data acquisition module 31, the rupture prediction module 32, and the prediction model training module 33 are shown.
[0122] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.
[0123] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.
[0124] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.
[0125] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the water scouring and fracturing prediction method for soft coal seams provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.
[0126] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.
[0127] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.
[0128] Those skilled in the art will recognize that the modules / units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.
[0129] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0130] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules, units, or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or modules / units, or it may be an electrical, mechanical, or other form of connection.
[0131] The modules / units described as separate components may or may not be physically separate. Similarly, the components shown as modules / units may or may not be physical modules / units; they may be located in one place or distributed across multiple network modules / units. Some or all of the modules / units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.
[0132] Furthermore, the functional modules / units in the various embodiments of this application can be integrated into one processing module / unit, or each module / unit can exist physically separately, or two or more modules / units can be integrated into one module / unit. The integrated modules / units described above can be implemented in hardware or in the form of software functional modules / units.
[0133] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for predicting water scour and fracturing in soft coal seams, characterized in that, include: Obtain the target coal sample parameters and target hydrodynamic parameters in the target fractured and soft coal seam; The target coal sample parameters and the target hydrodynamic parameters are input into the target fracturing prediction model to obtain the fracturing prediction results of the target soft coal seam; The training process of the target rupture prediction model includes: Acquire source domain data, auxiliary domain data, and target domain data; wherein, the source domain data includes multiple sets of laboratory data, each set of laboratory data including: experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracturing labels; the auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set of auxiliary soft coal seam data including: auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracturing labels; the target domain data includes multiple sets of target soft coal seam data, each set of target soft coal seam data including: historical coal sample parameters and historical hydrodynamic parameters of the target soft coal seam; the similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data is greater than a preset threshold; An initial fracturing prediction model is trained based on the source domain data and the auxiliary domain data to obtain an intermediate fracturing prediction model adapted to the actual scale. The target domain data is input into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data; The intermediate rupture prediction model is trained based on the auxiliary domain data, the target domain data, and the pseudo-labels to obtain the target rupture prediction model; The initial rupture prediction model includes: a feature extractor, a task predictor, and a domain discriminator; The process of training an initial fracturing prediction model based on the source domain data and the auxiliary domain data to obtain an intermediate fracturing prediction model adapted to the actual scale includes: The feature extractor and the task predictor are pre-trained based on the source domain data to obtain the initial rupture prediction model after pre-training. The source domain data and the auxiliary domain data are input into the pre-trained feature extractor to obtain a high-dimensional feature vector. The high-dimensional feature vector is input into the pre-trained task predictor, and the corresponding loss information is calculated based on the cross-entropy loss function. The high-dimensional feature vector is input into the domain discriminator, and the corresponding loss information is calculated based on the binary cross-entropy loss function. Based on the loss information corresponding to the cross-entropy loss function and the loss information corresponding to the binary cross-entropy loss function, determine the loss information corresponding to the joint loss function; The parameters of the feature extractor, the task predictor, and the domain discriminator are alternately updated based on the backpropagation algorithm so that the loss information corresponding to the joint loss function meets the preset conditions, thereby obtaining the intermediate breakage prediction model; wherein, the loss function in the pre-training process is different from the loss function in the adversarial training process; The intermediate rupture prediction model contains a task predictor; The step of inputting the target domain data into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data includes: The target domain data is input into the intermediate fracturing prediction model to obtain the fracturing probability and confidence level corresponding to each group of target soft coal seam data output by the task predictor. Filter the target broken and soft coal seam data in each group of target broken and soft coal seam data with a confidence level greater than the preset confidence threshold to obtain standard target broken and soft coal seam data; The fracturing probability corresponding to the standard target soft coal seam data is used as the pseudo-label corresponding to the target domain data.
2. The method for predicting water scour and fracturing in soft coal seams as described in claim 1, characterized in that, The intermediate fracturing prediction model also includes a correction module; the correction module is connected to the output of the task predictor, and the parameters of the correction module include a correction factor; the correction factor is calculated based on the target coal sample parameters and the auxiliary coal sample parameters, and is used to correct the prediction results of the intermediate fracturing prediction model to adapt to the target soft coal seam; The process of training the intermediate rupture prediction model based on the auxiliary domain data, the target domain data, and the pseudo-labels corresponding to the target domain data to obtain the target rupture prediction model includes: The auxiliary domain data, the target domain data, and the pseudo-labels corresponding to the target domain data are mixed to obtain a hybrid training dataset. Using the label loss of the mixed training dataset as the optimization objective, the intermediate rupture prediction model is trained iteratively for multiple rounds based on the mixed training dataset and the correction factor to obtain the target rupture prediction model; After each iteration, the mixed training dataset is updated, and the correction factor is updated.
3. The method for predicting water scour and fracturing in soft coal seams as described in claim 2, characterized in that, The updating of the hybrid training dataset includes: The target domain data prediction results output after the current iteration of training are filtered according to the preset confidence threshold to obtain updated high-confidence pseudo-labels; The updated high-confidence pseudo-labels and their corresponding target domain data are then remixed with the auxiliary domain data to obtain the updated mixed training dataset.
4. The method for predicting water scour and fracturing in soft coal seams as described in claim 3, characterized in that, Updating the correction factor includes: Based on the target domain data and auxiliary domain data in the updated hybrid training dataset, the relative difference rate is calculated. The updated correction factor is obtained based on the relative difference rate.
5. A device for predicting water scour and fracturing in soft coal seams, characterized in that, For implementing the method as described in any one of claims 1-4, the apparatus comprises: The data acquisition module is used to acquire target coal sample parameters and target hydrodynamic parameters in the target broken and soft coal seam; The fracture prediction module is used to input the target coal sample parameters and the target hydrodynamic parameters into the target fracture prediction model to obtain the fracture prediction results of the target fractured soft coal seam; The prediction model training module, during the training of the target fracturing prediction model, is used to acquire source domain data, auxiliary domain data, and target domain data. The source domain data includes multiple sets of laboratory data, each set including experimental coal sample parameters, experimental hydrodynamic parameters, and corresponding fracturing labels. The auxiliary domain data includes multiple sets of auxiliary soft coal seam data, each set including auxiliary coal sample parameters, auxiliary hydrodynamic parameters, and corresponding fracturing labels. The target domain data includes multiple sets of target soft coal seam data, each set including historical coal sample parameters and historical hydrodynamic parameters of the target soft coal seam. The similarity between each set of auxiliary soft coal seam data and each set of target soft coal seam data is greater than a preset threshold. An initial fracturing prediction model is trained based on the source domain data and the auxiliary domain data to obtain an intermediate fracturing prediction model adapted to the actual scale. The target domain data is input into the intermediate rupture prediction model to obtain the pseudo-labels corresponding to the target domain data; The intermediate rupture prediction model is trained based on the auxiliary domain data, the target domain data, and the pseudo-labels to obtain the target rupture prediction model.
6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 4.
7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4.