Method for monitoring ultrasonic plug removal process of deep coal bed gas

Abstract

The invention discloses a deep coal bed gas ultrasonic plug removal process monitoring method, and belongs to the technical field of coal mining, training data is constructed through historical deep coal bed gas ultrasonic plug removal excitation parameters and ultrasonic plug removal performance index labels, and then according to training samples and training labels in the training data, the deep coal bed gas ultrasonic plug removal process is monitored. Training the deep coal bed gas ultrasonic unblocking process monitoring model to obtain a trained deep coal bed gas ultrasonic unblocking process monitoring model, and finally monitoring the ultrasonic unblocking process by adopting the trained deep coal bed gas ultrasonic unblocking process monitoring model to determine a monitoring result corresponding to the deep coal bed gas ultrasonic unblocking process. The deep coal bed gas ultrasonic plug removal process can be effectively monitored, a worker can find the abnormal condition of ultrasonic plug removal in time, meanwhile, the requirement for professional knowledge of the worker can be lowered, and the monitoring efficiency of the deep coal bed gas ultrasonic plug removal process can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of coal mining, and in particular relates to a method for monitoring a deep coalbed methane ultrasonic unblocking process. Background Art

[0002] The deep coal body has high stratum pressure, developed micropore structure and poor permeability. Compared with the middle and shallow coal seams, although the proportion of free gas in deep coalbed methane is relatively high, the main problem of deep coalbed methane production is that the self-blowing period is short and the gas well is in a low-production stage dominated by desorption for a long time. The main influencing factors are as follows: (1) Deep coalbed methane drainage is more affected by stress sensitivity, velocity sensitivity, Jiamin effect and other factors. (2) Deep coalbed methane drainage is seriously blocked by scaling and the adsorbed gas is released slowly. Therefore, in order to achieve efficient drainage of deep coalbed methane and maximize the economic benefits of single wells, it is necessary to carry out secondary reservoir permeability transformation during the drainage process to achieve the goal of unblocking and increasing flow.

[0003] The construction techniques and parameters of traditional reservoir secondary transformation technologies (such as secondary fracturing, CO2 huff and puff, nitrogen disturbance, acidification and plugging removal, etc.) are mostly based on the experience of large-scale fracturing of medium and shallow coal seams or shale gas, and all have technical shortcomings to varying degrees. It is urgent to explore engineering technology methods that are compatible with the geological conditions of deep coal seams. Ultrasonic reservoir transformation technology has obvious advantages such as environmental protection, high efficiency, strong operability, safety and controllability, and has gradually attracted people's attention and recognition in the field of coalbed methane permeability enhancement.

[0004] Deep coalbed methane ultrasonic unblocking technology is an innovative technology that improves the permeability of deep coalbed methane through ultrasonic vibration. This technology changes the pore and fracture structure of the coal seam, releases the water lock effect, and eliminates reservoir sediments by applying a high-power acoustic field, thereby improving the permeability of the coal seam reservoir. Research and practice have shown that when ultrasonic waves propagate in two-phase or even multi-phase media, they can produce vibration effects, thermal effects, and cavitation effects, increase the temperature of the coal seam, promote coalbed methane desorption and seepage, and improve the pore and fracture network channels of the coal body, thereby achieving the purpose of permeability transformation, unblocking, and flow increase of coalbed methane reservoirs. This technology has been applied in actual projects. For example, ultrasonic vibration is used to remove coal powder blocking the screen pipe, which optimizes the coalbed methane mining process. It has the advantages of high efficiency and safety, and has been widely used in recent years. However, the existing ultrasonic unblocking process lacks effective monitoring means, and it is difficult to grasp the unblocking effect and coal seam status in real time. It is easy to over-unblock or incomplete unblocking, which affects mining efficiency and safety. Summary of the invention

[0005] The present invention provides a method for monitoring the ultrasonic unblocking process of deep coalbed methane, which is used to solve the problem that the prior art is difficult to grasp the unblocking effect and coal seam status in real time, and is prone to excessive unblocking or incomplete unblocking, thereby affecting mining efficiency and safety.

[0006] A method for monitoring the ultrasonic plugging removal process of deep coalbed methane, comprising:

[0007] Obtain the historical ultrasonic plugging removal excitation parameters of deep coalbed methane and the ultrasonic plugging removal performance index labels input by the staff through human-computer interaction;

[0008] Construct training data according to the historical ultrasonic plugging removal excitation parameters of deep coalbed methane and the ultrasonic plugging removal performance index labels; wherein, the training data includes training samples and training labels;

[0009] Construct a monitoring model for the ultrasonic plugging removal process of deep coalbed methane, and train the monitoring model for the ultrasonic plugging removal process of deep coalbed methane according to the training samples and training labels in the training data to obtain a trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane;

[0010] Use the trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane to monitor the ultrasonic plugging removal process, and determine the monitoring result corresponding to the ultrasonic plugging removal process of deep coalbed methane.

[0011] In a possible implementation manner, after determining the monitoring result corresponding to the ultrasonic plugging removal process of deep coalbed methane, it further includes: when the monitoring result meets a preset warning condition, a warning is given.

[0012] In a possible implementation manner, constructing training data according to the historical ultrasonic plugging removal excitation parameters of deep coalbed methane and the ultrasonic plugging removal performance index labels includes: constructing training samples according to the historical ultrasonic plugging removal excitation parameters of deep coalbed methane, constructing training labels according to the ultrasonic plugging removal performance index labels corresponding to the historical ultrasonic plugging removal excitation parameters of deep coalbed methane, and using the training samples and the training labels together as training data.

[0013] In a possible implementation manner, constructing a monitoring model for the ultrasonic plugging removal process of deep coalbed methane includes: using a CNN model, a CNN-LSTM model, an RNN model, an LSTM model, a BP neural network model or a CNN-BP neural network model to construct a monitoring model for the ultrasonic plugging removal process of deep coalbed methane.

[0014] In a possible implementation manner, training the monitoring model for the ultrasonic plugging removal process of deep coalbed methane according to the training samples and training labels in the training data to obtain a trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane includes:

[0015] Initialize the model parameters of the monitoring model for the ultrasonic plugging removal process of deep coalbed methane to obtain multiple different model parameter vectors;

[0016] For any model parameter vector, determine the fitness corresponding to each model parameter vector according to the training samples and training labels in the training data;

[0017] According to the fitness corresponding to each model parameter vector, determine the optimal parameter vector among all model parameter vectors;

[0018] Based on the optimal parameter vector, perform adaptive information interaction on the model parameter vector using an adaptive information interaction strategy to determine the model parameter vector after adaptive information interaction;

[0019] Perform joint search on the model parameter vector after adaptive information interaction using an adaptive chaotic joint search strategy to determine the model parameter vector after joint search;

[0020] Perform neighborhood enhancement search on the model parameter vector after joint search using a pivot-around search strategy to determine the model parameter vector after neighborhood enhancement search;

[0021] Perform global random search on the model parameter vector after neighborhood enhancement search using a Levy random walk search strategy to determine the model parameter vector after global random search;

[0022] Repeat the execution of the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the pivot-around search strategy, and the Levy random walk search strategy until the training end condition is satisfied. According to the model parameter vector after global random search in the last training process, obtain the monitoring model for the deep coalbed methane ultrasonic plugging removal process after training.

[0023] In a possible implementation manner, based on the optimal parameter vector, perform adaptive information interaction on the model parameter vector using an adaptive information interaction strategy to determine the model parameter vector after adaptive information interaction, including:

[0024] Based on the current training times, respectively determine an adaptive speed control factor, a first information interaction factor, and a second information interaction factor;

[0025] For any model parameter vector, match multiple other model parameter vectors with the closest Euclidean distance to the model parameter vector to obtain a neighboring group corresponding to the model parameter vector;

[0026] Based on the neighboring group corresponding to the model parameter vector, obtain a neighborhood correction parameter vector;

[0027] According to the first information interaction factor and the optimal parameter vector, obtain a first adaptive information interaction amount;

[0028] Obtain a second adaptive information interaction amount according to the second information interaction factor and the neighborhood correction parameter vector;

[0029] Obtain the training speed in the current training process according to the adaptive speed control factor, the training speed in the previous training process, the first adaptive information interaction amount, and the second adaptive information interaction amount;

[0030] Perform adaptive information interaction on the model parameter vector according to the training speed in the current training process, and determine the model parameter vector after the adaptive information interaction.

[0031] In a possible implementation manner, adopt an adaptive chaotic joint search strategy to perform joint search on the model parameter vector after the adaptive information interaction, and determine the model parameter vector after the joint search, including:

[0032] Obtain a chaotic search factor, and obtain a joint search factor according to the current training times;

[0033] Obtain the fitness corresponding to the model parameter vector after the adaptive information interaction, arrange the model parameter vectors in descending order of fitness, and obtain the sorted model parameter vectors;

[0034] For any sorted model parameter vector, randomly match another model parameter vector to the model parameter vector to obtain the information interaction parameter vector corresponding to the model parameter vector;

[0035] According to the sorted model parameter vectors, allocate a first joint search parameter vector and a second joint search parameter vector to the model parameter vectors;

[0036] Obtain a chaotic search amount according to the chaotic search factor, the first joint search parameter vector, and the second joint search parameter vector;

[0037] Obtain a joint search amount according to the joint search factor and the information interaction parameter vector corresponding to the model parameter vector;

[0038] Perform joint search on the model parameter vector after the adaptive information interaction according to the chaotic search amount, the joint search amount, and the information interaction parameter vector corresponding to the model parameter vector, and determine the model parameter vector after the joint search.

[0039] In a possible implementation manner, adopt a pivot surrounding search strategy to perform neighborhood enhancement search on the model parameter vector after the joint search, and determine the model parameter vector after the neighborhood enhancement search, including:

[0040] Based on the current training times, obtain a neighborhood enhancement search factor;

[0041] For any model parameter vector after joint search, randomly select another model parameter vector as the pivot point for the model parameter vector;

[0042] According to the pivot point corresponding to the model parameter vector after joint search and the neighborhood enhancement search factor, perform neighborhood enhancement search on the model parameter vector after joint search to determine the model parameter vector after neighborhood enhancement search.

[0043] In a possible implementation manner, perform global random search on the model parameter vector after neighborhood enhancement search using the Levy random walk search strategy to determine the model parameter vector after global random search, including:

[0044] Adopt the Levy flight strategy to obtain the Levy random walk factor;

[0045] According to the Levy random walk factor and the optimal parameter vector, combined with the exponential function, perform global random search on the model parameter vector after neighborhood enhancement search to determine the model parameter vector after global random search.

[0046] In a possible implementation manner, repeatedly execute the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the pivot point surrounding search strategy, and the Levy random walk search strategy until the training end condition is met. According to the model parameter vector after global random search in the last training process, obtain the deep coalbed methane ultrasonic plugging removal process monitoring model after training, including:

[0047] Execute the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the pivot point surrounding search strategy, and the Levy random walk search strategy;

[0048] Judge whether the current training times are greater than or equal to the preset maximum training times. If so, re-determine the optimal parameter vector according to the model parameter vector after global random search in the last training process. Otherwise, return to the step of executing the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the pivot point surrounding search strategy, and the Levy random walk search strategy;

[0049] Take the hyperparameters included in the re-determined optimal parameter vector as the final hyperparameters of the deep coalbed methane ultrasonic plugging removal process monitoring model to obtain the deep coalbed methane ultrasonic plugging removal process monitoring model after training.

[0050] A method for monitoring the ultrasonic plugging removal process of deep coalbed methane provided by the present invention constructs training data through the historical ultrasonic plugging removal excitation parameters of deep coalbed methane and ultrasonic plugging removal performance index tags, and then trains the monitoring model for the ultrasonic plugging removal process of deep coalbed methane according to the training samples and training tags in the training data to obtain the trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane. Finally, the trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane is used to monitor the ultrasonic plugging removal process to determine the monitoring result corresponding to the ultrasonic plugging removal process of deep coalbed methane, which can effectively realize the monitoring of the ultrasonic plugging removal process of deep coalbed methane, enable the staff to timely discover abnormal situations in ultrasonic plugging removal, and at the same time can reduce the professional knowledge requirements for the staff and improve the monitoring efficiency of the ultrasonic plugging removal process of deep coalbed methane. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0052] Figure 1 It is a flowchart of a method for monitoring the ultrasonic plugging removal process of deep coalbed methane provided by an embodiment of the present invention.

[0053] Figure 2 It is a flowchart of training a monitoring model for the ultrasonic plugging removal process of deep coalbed methane provided by an embodiment of the present invention.

[0054] Through the above-mentioned accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0056] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0057] As Figure 1 shown, an embodiment of the present invention provides a method for monitoring the ultrasonic plugging removal process of deep coalbed methane, including:

[0058] S101. Obtain the historical deep coalbed methane ultrasonic plugging removal excitation parameters input by the staff through human-computer interaction and the ultrasonic plugging removal performance index labels;

[0059] The historical deep coalbed methane ultrasonic plugging removal excitation parameters may include: parameters related to ultrasonic plugging removal such as excitation power, excitation frequency, excitation time, excitation distance, etc.

[0060] The ultrasonic plugging removal performance index labels can be set as: pore morphology characteristic parameters such as low-permeability coal body pore density, geometric size, specific surface area, etc. or the change rates of these pore morphology characteristic parameters; however, it is worth noting that the ultrasonic plugging removal performance index labels can also be the plugging removal completion rate so that the staff can better understand the plugging removal progress.

[0061] Optionally, in addition to the historical deep coalbed methane ultrasonic plugging removal excitation parameters, it is also possible to: obtain the initial pore morphology characteristic parameters of the deep coalbed methane ultrasonic plugging removal process as identification data and participate in the data analysis process together with the historical deep coalbed methane ultrasonic plugging removal excitation parameters to improve the monitoring accuracy of the deep coalbed methane ultrasonic plugging removal process.

[0062] S102. Construct training data according to the historical deep coalbed methane ultrasonic plugging removal excitation parameters and the ultrasonic plugging removal performance index labels; wherein, the training data includes training samples and training labels;

[0063] In the prior art, it often relies on the staff to use professional equipment for sampling and analysis. Not only is the sampling process complex, but the analysis requires relatively complex calculations, so it is difficult to effectively monitor the deep coalbed methane ultrasonic plugging removal process. During the ultrasonic plugging removal process, both the deep coal ultrasonic plugging removal excitation parameters and the initial pore morphology characteristic parameters have an impact on the plugging removal effect. Therefore, the historical deep coalbed methane ultrasonic plugging removal excitation parameters can be used as training samples, or the historical deep coalbed methane ultrasonic plugging removal excitation parameters and the initial pore morphology characteristic parameters can be used together as training samples, so that deep learning can be carried out on this basis.

[0064] Optionally, during the process of constructing the training data, normalization processing can also be used to reduce the data complexity to increase the data processing speed. Similarly, during the subsequent process of identifying data, the data can also be normalized.

[0065] S103. Construct a monitoring model for the deep coalbed methane ultrasonic plugging removal process, and train the monitoring model for the deep coalbed methane ultrasonic plugging removal process according to the training samples and training labels in the training data to obtain the trained monitoring model for the deep coalbed methane ultrasonic plugging removal process;

[0066] In this embodiment, a deep coalbed methane ultrasonic plugging removal process monitoring model can be constructed using deep learning models (such as convolutional neural networks, long short-term memory networks, etc.). The deep coalbed methane ultrasonic plugging removal process monitoring model can learn the data relationship between training samples and training labels. After training, a trained deep coalbed methane ultrasonic plugging removal process monitoring model can be obtained, and the trained deep coalbed methane ultrasonic plugging removal process monitoring model has the ability to monitor the plugging removal process.

[0067] S104. Use the trained deep coalbed methane ultrasonic plugging removal process monitoring model to monitor the ultrasonic plugging removal process and determine the monitoring result corresponding to the deep coalbed methane ultrasonic plugging removal process.

[0068] The real-time deep coalbed ultrasonic plugging removal excitation parameters can be obtained, and then the real-time deep coalbed ultrasonic plugging removal excitation parameters are used as the input of the trained deep coalbed methane ultrasonic plugging removal process monitoring model to obtain the predicted ultrasonic plugging removal performance index label, so as to monitor the ultrasonic plugging removal process. Using the trained deep coalbed methane ultrasonic plugging removal process monitoring model to monitor the ultrasonic plugging removal process can effectively improve the monitoring efficiency of the plugging removal process and reduce the monitoring difficulty.

[0069] It should be noted that when using the trained deep coalbed methane ultrasonic plugging removal process monitoring model to monitor the ultrasonic plugging removal process, the data structure of the input data should be the same as that of the training samples, so as to realize data identification and analysis.

[0070] A method for monitoring the deep coalbed methane ultrasonic plugging removal process provided by the present invention constructs training data through the historical deep coalbed methane ultrasonic plugging removal excitation parameters and ultrasonic plugging removal performance index labels, and then trains the deep coalbed methane ultrasonic plugging removal process monitoring model according to the training samples and training labels in the training data to obtain a trained deep coalbed methane ultrasonic plugging removal process monitoring model. Finally, the trained deep coalbed methane ultrasonic plugging removal process monitoring model is used to monitor the ultrasonic plugging removal process to determine the monitoring result corresponding to the deep coalbed methane ultrasonic plugging removal process, which can effectively realize the monitoring of the deep coalbed methane ultrasonic plugging removal process, enable the staff to timely discover abnormal situations in ultrasonic plugging removal, and at the same time reduce the professional knowledge requirements for the staff and improve the monitoring efficiency of the deep coalbed methane ultrasonic plugging removal process.

[0071] In a possible implementation manner, after determining the monitoring result corresponding to the deep coalbed methane ultrasonic plugging removal process, it further includes: when the monitoring result meets a preset warning condition, a warning is given.

[0072] For example, the preset warning condition can be the unqualified condition of the ultrasonic plugging removal performance index label. When the identified ultrasonic plugging removal performance index label meets the unqualified condition, a warning can be given.

[0073] In a possible implementation manner, training data is constructed according to the historical deep coalbed methane ultrasonic plugging removal excitation parameters and the ultrasonic plugging removal performance index tags, including: constructing training samples according to the historical deep coalbed methane ultrasonic plugging removal excitation parameters, constructing training tags according to the ultrasonic plugging removal performance index tags corresponding to the historical deep coalbed methane ultrasonic plugging removal excitation parameters, and taking the training samples and the training tags together as training data.

[0074] In a possible implementation manner, a monitoring model for the ultrasonic plugging removal process of deep coalbed methane is constructed, including: using a CNN (Convolutional Neural Network) model, a CNN-LSTM (Long Short-Term Memory) model, an RNN (Recurrent Neural Network) model, an LSTM model, a BP neural network model, or a CNN-BP neural network model to construct a monitoring model for the ultrasonic plugging removal process of deep coalbed methane.

[0075] In the embodiment of the present invention, it is preferably to use a CNN model to construct a monitoring model for the ultrasonic plugging removal process of deep coalbed methane. However, it should be noted that other neural network models can still be used to construct a monitoring model for the ultrasonic plugging removal process of deep coalbed methane.

[0076] As Figure 2 shown, according to the training samples and training tags in the training data, the monitoring model for the ultrasonic plugging removal process of deep coalbed methane is trained to obtain a trained monitoring model for the ultrasonic plugging removal process of deep coalbed methane, including:

[0077] S201. Initialize the model parameters of the monitoring model for the ultrasonic plugging removal process of deep coalbed methane to obtain a plurality of different model parameter vectors;

[0078] For the model parameters (such as connection weights) of a deep learning model, generally, the model parameters have corresponding upper and lower limits. Therefore, the model parameters can be randomly initialized between their upper and lower limits, and the initialized model parameters are combined into a vector, so that a model parameter vector can be obtained. After repeating the initialization multiple times, a plurality of different model parameter vectors can be obtained.

[0079] S202. For any one of the model parameter vectors, determine the fitness corresponding to each model parameter vector according to the training samples and training tags in the training data;

[0080] The model parameter vector can be first applied to the deep coalbed methane ultrasonic unblocking process monitoring model, and then the training samples are used as the input data of the deep coalbed methane ultrasonic unblocking process monitoring model, and the training labels are used as the expected output data of the deep coalbed methane ultrasonic unblocking process monitoring model, so that the loss function value corresponding to the model parameter vector can be obtained. After adding the obtained loss function value to the preset minimum constant and taking the reciprocal, the fitness corresponding to the model parameter vector can be obtained. By setting a minimum constant (such as 0.0001), the situation where the denominator is zero can be effectively avoided.

[0081] S203, according to the fitness corresponding to each model parameter vector, determine the optimal parameter vector among all model parameter vectors; that is, the model parameter vector with the smallest fitness is the optimal parameter vector;

[0082] S204, based on the optimal parameter vector, adopting an adaptive information interaction strategy to perform adaptive information interaction on the model parameter vector, and determining the model parameter vector after the adaptive information interaction;

[0083] S205, using an adaptive chaos joint search strategy to conduct a joint search on the model parameter vector after the adaptive information interaction, and determining the model parameter vector after the joint search;

[0084] S206, using a pivot point around search strategy to perform neighborhood enhanced search on the model parameter vector after the joint search, to determine the model parameter vector after the neighborhood enhanced search;

[0085] S207, using the Levy random walk search strategy to perform a global random search on the model parameter vector after the neighborhood enhanced search, and determining the model parameter vector after the global random search;

[0086] S208. Repeat the adaptive information interaction strategy, the adaptive chaos joint search strategy, the pivot point search strategy and the Levy random walk search strategy until the training end condition is met, and obtain the deep coalbed methane ultrasonic deblocking process monitoring model after training based on the model parameter vector after the global random search in the last training process.

[0087] In the prior art, gradient descent and back propagation are often used to train deep learning models, which leads to the problem that the algorithm falls into local optimality and the training effect is poor, so that the deep coalbed methane ultrasonic unblocking process monitoring model after training cannot effectively perform data recognition, and finally reduces the monitoring effect of the deep coalbed methane ultrasonic unblocking process. Therefore, the embodiment of the present invention provides a new training algorithm to solve the problems existing in the prior art and improve the monitoring accuracy of the deep coalbed methane ultrasonic unblocking process.

[0088] Optionally, after the model parameter vector changes, boundary crossing processing should be performed on it, and at the same time, the optimal model parameter vector should be updated, that is, the optimal model parameter vector should be re-obtained to ensure the training speed of the algorithm.

[0089] In a possible implementation manner, based on the optimal parameter vector, an adaptive information interaction strategy is adopted to perform adaptive information interaction on the model parameter vector to determine the model parameter vector after adaptive information interaction, including:

[0090] S2041. Based on the current training times, respectively determine the adaptive speed control factor, the first information interaction factor, and the second information interaction factor as:

[0091]

[0092] where η represents the adaptive speed control factor, a1 represents the first constant factor, a2 represents the second constant factor, a3 represents the third constant factor, and the first constant factor, the second constant factor, and the third constant factor can be set to 2, 1, and 0.3 respectively, t represents the current training times, T represents the preset maximum training times, c1 represents the first information interaction factor, sin represents the sine function, π represents the pi, c2 represents the second information interaction factor, and cos represents the cosine function;

[0093] S2042. For any model parameter vector, match multiple other model parameter vectors with the closest Euclidean distance to the model parameter vector to obtain the neighboring group corresponding to the model parameter vector;

[0094] S2043. Based on the neighboring group corresponding to the model parameter vector, obtain the neighborhood correction parameter vector as:

[0095]

[0096] where represents the d-th dimensional model parameter of the neighborhood correction parameter vector, d = 1, 2,..., D, D represents the total number of model parameters, L represents the total number of other model parameter vectors in the neighboring group, represents the d-th dimensional model parameter of the optimal parameter vector, f best represents the fitness corresponding to the optimal parameter vector, f l represents the fitness of the l-th other model parameter vector in the neighboring group, represents the d-th dimensional model parameter of the l-th other model parameter vector in the neighboring group;

[0097] S2044. According to the first information interaction factor and the optimal parameter vector, obtain the first adaptive information interaction amount as:

[0098]

[0099] Among them, represents the m-th model parameter vector in the t-th training process, and Δ m1 represents the first adaptive information interaction amount corresponding to the m-th model parameter vector, r1 represents the first random number between (0, 1), m = 1, 2, …, M, and M represents the total number of model parameter vectors;

[0100] S2045. Obtain the second adaptive information interaction amount according to the second information interaction factor and the neighborhood correction parameter vector as:

[0101]

[0102] Among them, Δ m2 represents the second adaptive information interaction amount, and r2 represents the second random number between (0, 1);

[0103] S2046. Obtain the training speed in the current training process according to the adaptive speed control factor, the training speed in the previous training process, the first adaptive information interaction amount, and the second adaptive information interaction amount as:

[0104]

[0105] Among them, represents the training speed corresponding to the m-th model parameter vector in the t-th training process, that is, the training speed in the previous training process; represents the training speed corresponding to the m-th model parameter vector in the (t + 1)-th training process, that is, the training speed in the current training process; exp represents the exponential function with the natural constant e as the base;

[0106] S2047. Perform adaptive information interaction on the model parameter vector according to the training speed in the current training process, and determine the model parameter vector after adaptive information interaction as:

[0107]

[0108] Among them, represents the model parameter vector after the m-th adaptive information interaction.

[0109] In the embodiments of the present invention, an adaptive information interaction strategy is adopted to perform adaptive information interaction on the model parameter vectors, which can effectively combine the position information included in the neighborhood group during the training process, enable all model parameter vectors to perform neighborhood information interaction, avoid the training from falling into local optimum to a certain extent, and at the same time learn the optimal model parameter vectors to ensure that the training of the algorithm is always effective and fast. And perform adaptive control on the training process, which can effectively improve the search accuracy in the later stage of the algorithm and improve the training effect of the algorithm.

[0110] In a possible implementation manner, an adaptive chaotic joint search strategy is adopted to perform joint search on the model parameter vectors after adaptive information interaction to determine the model parameter vectors after joint search, including:

[0111] S2051. Obtain a chaotic search factor, and obtain a joint search factor according to the current training times as:

[0112] λ t+1 = r3λ t ×(1 - λ t )

[0113]

[0114] where λ t represents the chaotic search factor in the t-th training process, λ t+1 represents the chaotic search factor in the (t + 1)-th training process, and the chaotic search factor is randomly generated between (0, 1) at the first training, r3 represents the third random number between (0, 1), β represents the joint search factor, e represents the natural constant, π represents the pi, T represents the preset maximum number of training times, and r4 represents the fourth random number between (0, 1);

[0115] S2052. Obtain the fitness corresponding to the model parameter vectors after adaptive information interaction, and arrange the model parameter vectors in descending order of fitness to obtain the sorted model parameter vectors;

[0116] S2053. For any one of the sorted model parameter vectors, randomly match another model parameter vector to obtain the information interaction parameter vector corresponding to the model parameter vector;

[0117] S2054. According to the sorted model parameter vectors, allocate the first joint search parameter vector and the second joint search parameter vector to the model parameter vectors;

[0118] When the model parameter vectors are ranked first and second, both the first joint search parameter vector and the second joint search parameter vector are set as the information interaction parameter vectors;

[0119] When the sorting of the model parameter vectors is not the first and the second, for the n-th sorted model parameter vector, the (n - 1)-th model parameter vector is taken as the first joint search parameter vector, and the (n - 2)-th model parameter vector is taken as the second joint search parameter vector;

[0120] S2055. Obtain the chaotic search quantity according to the chaotic search factor, the first joint search parameter vector, and the second joint search parameter vector as:

[0121]

[0122] where, represents the n-th sorted model parameter vector in the t-th training process, represents the first joint search parameter vector corresponding to the n-th sorted model parameter vector, represents the second joint search parameter vector corresponding to the n-th sorted model parameter vector;

[0123] S2056. Obtain the joint search quantity according to the joint search factor and the information interaction parameter vector corresponding to the model parameter vector as:

[0124]

[0125] where, Δ n2 represents the joint search quantity corresponding to the n-th sorted model parameter vector, represents the information interaction parameter vector corresponding to the n-th sorted model parameter vector;

[0126] S2057. Perform joint search on the model parameter vector after adaptive information interaction according to the chaotic search quantity, the joint search quantity, and the information interaction parameter vector corresponding to the model parameter vector, and determine the model parameter vector after joint search as:

[0127]

[0128] where, represents the model parameter vector after the n-th joint search, n = 1, 2, …, M, and M represents the total number of model parameter vectors.

[0129] In the embodiments of the present invention, an adaptive chaos joint search strategy is adopted to jointly search the model parameter vector after adaptive information interaction. During the process of searching the solution space of the model parameter vector, other model parameter vectors can be jointly used to search unfamiliar regions. At the same time, while all model parameter vectors are always searching for better positions, joint search is carried out, which can not only improve the traversal of the solution space and enhance the global search ability, but also increase the possibility of finding the optimal solution to a certain extent and balance the local search ability.

[0130] In a possible implementation manner, a pivot around search strategy is adopted to perform neighborhood enhancement search on the model parameter vector after joint search to determine the model parameter vector after neighborhood enhancement search, including:

[0131] S2061. Based on the current training times, obtain the neighborhood enhancement search factor as:

[0132]

[0133] where s represents the neighborhood enhancement search factor, s min represents the minimum value corresponding to the neighborhood enhancement search factor, s max represents the maximum value corresponding to the neighborhood enhancement search factor, μ represents the control parameter of the neighborhood enhancement search factor, and can be set to 5;

[0134] S2062. For any model parameter vector after joint search, randomly select another model parameter vector as the pivot for the model parameter vector;

[0135] S2063. According to the pivot corresponding to the model parameter vector after joint search and the neighborhood enhancement search factor, perform neighborhood enhancement search on the model parameter vector after joint search to determine the model parameter vector after neighborhood enhancement search as:

[0136]

[0137] where, represents the k-th model parameter vector after joint search in the t-th training process, represents the k-th model parameter vector after neighborhood enhancement search, k = 1, 2,..., M, M represents the total number of model parameter vectors, r5 represents the fifth random number between (0, 1), r6 represents the sixth random number between (0, 1), represents the optimal parameter vector, represents the pivot.

[0138] In the embodiment of the present invention, a fulcrum-around search strategy is adopted to perform neighborhood enhancement search on the model parameter vector after joint search, which can strengthen the search for the neighborhood range. At the same time, as the algorithm progresses, the search accuracy will become higher and higher, effectively improving the search ability of the algorithm and increasing a certain global search ability.

[0139] In a possible implementation manner, a Levy random walk search strategy is adopted to perform global random search on the model parameter vector after neighborhood enhancement search to determine the model parameter vector after global random search, including:

[0140] S2071. Obtain a Levy random walk factor Levy using the Levy flight strategy;

[0141] S2072. Perform global random search on the model parameter vector after neighborhood enhancement search according to the Levy random walk factor and the optimal parameter vector, in combination with the exponential function, and determine the model parameter vector after global random search as:

[0142]

[0143]

[0144] where represents the model parameter vector after the j-th neighborhood enhancement search in the t-th training process, represents the model parameter vector after the j-th global random search, j = 1, 2,..., M, where M represents the total number of model parameter vectors, r7 represents the seventh random number between (0, 1), r8 represents the eighth random number between (0, 1), represents the training speed corresponding to the model parameter vector after the j-th neighborhood enhancement search in the (t + 1)-th training process, represents the average training speed in the (t + 1)-th training process, represents the training speed corresponding to the model parameter vector after the j-th neighborhood enhancement search in the t-th training process, c3 represents the information interaction factor, which can be set as a constant between (0.2, 0.5), and e represents the natural constant, represents a random model parameter vector, r9 represents the ninth random number between (-08, 0.8), r 10 represents the tenth random number that is randomly a positive integer, r 11 represents the eleventh random number between (0, 1).

[0145] Optionally, after obtaining the model parameter vector after global random search, it can be determined whether the fitness of the model parameter vector after global random search increases. If so, accept this global random search; otherwise, reject this global random search, which can effectively ensure the training speed of the algorithm.

[0146] In the embodiments of the present invention, through the cooperation of the above several searches, not only the global search ability of the algorithm is greatly improved, but also the local search ability and the convergence fineness are ensured, which greatly increases the training effect of the deep learning model, and finally improves the recognition accuracy of the deep coalbed methane ultrasonic plugging removal process monitoring model for data, and improves the monitoring effect of the plugging removal process.

[0147] In a possible implementation manner, the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the fulcrum around search strategy, and the Levy random walk search strategy are repeatedly executed until the training end condition is met. According to the model parameter vector after the global random search in the last training process, the deep coalbed methane ultrasonic plugging removal process monitoring model after training is obtained, including:

[0148] Execute the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the fulcrum around search strategy, and the Levy random walk search strategy;

[0149] Judge whether the current training times are greater than or equal to the preset maximum training times. If so, re-determine the optimal parameter vector according to the model parameter vector after the global random search in the last training process. Otherwise, return to the step of executing the adaptive information interaction strategy, the adaptive chaotic joint search strategy, the fulcrum around search strategy, and the Levy random walk search strategy;

[0150] Take the hyperparameters included in the re-determined optimal parameter vector as the final hyperparameters of the deep coalbed methane ultrasonic plugging removal process monitoring model to obtain the deep coalbed methane ultrasonic plugging removal process monitoring model after training.

[0151] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein.

[0152] The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include known common general knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A method for monitoring the ultrasonic unblocking process of deep coalbed methane, characterized in that: include: Obtain the historical depth coalbed methane ultrasonic plugging removal excitation parameters and ultrasonic plugging removal performance index labels input by the staff through human-computer interaction; Construct training data according to the historical depth coalbed methane ultrasonic deblocking excitation parameters and ultrasonic deblocking performance index labels; wherein the training data includes training samples and training labels; Constructing a deep coalbed methane ultrasonic unblocking process monitoring model, and training the deep coalbed methane ultrasonic unblocking process monitoring model according to the training samples and training labels in the training data to obtain a trained deep coalbed methane ultrasonic unblocking process monitoring model; The trained deep coalbed methane ultrasonic unblocking process monitoring model is used to monitor the ultrasonic unblocking process and determine the corresponding monitoring results of the deep coalbed methane ultrasonic unblocking process.

2. The deep coalbed methane ultrasonic unblocking process monitoring method according to claim 1 is characterized in that: After determining the monitoring result corresponding to the deep coalbed methane ultrasonic deblocking process, it also includes: when the monitoring result meets the preset warning condition, an early warning is issued.

3. The method for monitoring deep coalbed methane ultrasonic unblocking process according to claim 1 is characterized in that: Training data is constructed according to the historical depth coalbed methane ultrasonic unblocking excitation parameters and ultrasonic unblocking performance index labels, including: constructing training samples according to the historical depth coalbed methane ultrasonic unblocking excitation parameters, constructing training labels according to the ultrasonic unblocking performance index labels corresponding to the historical depth coalbed methane ultrasonic unblocking excitation parameters, and using the training samples and the training labels together as training data.

4. The method for monitoring deep coalbed methane ultrasonic unblocking process according to claim 1, characterized in that: Constructing a deep coalbed methane ultrasonic deblocking process monitoring model, including: using a CNN model, a CNN-LSTM model, a RNN model, a LSTM model, a BP neural network model or a CNN-BP neural network model to construct a deep coalbed methane ultrasonic deblocking process monitoring model.

5. The method for monitoring deep coalbed methane ultrasonic plugging removal process according to claim 1, characterized in that: The deep coalbed methane ultrasonic deblocking process monitoring model is trained according to the training samples and training labels in the training data to obtain the trained deep coalbed methane ultrasonic deblocking process monitoring model, including: Initializing the model parameters of the deep coalbed methane ultrasonic deblocking process monitoring model to obtain a plurality of different model parameter vectors; For any model parameter vector, determine the fitness corresponding to each model parameter vector according to the training samples and training labels in the training data; According to the fitness corresponding to each model parameter vector, the optimal parameter vector is determined among all model parameter vectors; Based on the optimal parameter vector, an adaptive information interaction strategy is used to perform adaptive information interaction on the model parameter vector to determine the model parameter vector after the adaptive information interaction; Adopting an adaptive chaos joint search strategy to jointly search the model parameter vector after the adaptive information interaction, and determining the model parameter vector after the joint search; A pivot point surrounding search strategy is used to perform neighborhood enhanced search on the model parameter vector after the joint search, and the model parameter vector after the neighborhood enhanced search is determined; A Levy random walk search strategy is used to perform a global random search on the model parameter vector after the neighborhood enhanced search, and the model parameter vector after the global random search is determined; The adaptive information interaction strategy, adaptive chaos joint search strategy, pivot point search strategy and Levy random walk search strategy are repeatedly executed until the training end conditions are met. According to the model parameter vector after the global random search in the last training process, the deep coalbed methane ultrasonic deblocking process monitoring model after training is obtained.

6. The method for monitoring the deep coalbed methane ultrasonic plugging removal process according to claim 5 is characterized in that: Based on the optimal parameter vector, an adaptive information interaction strategy is used to perform adaptive information interaction on the model parameter vector, and the model parameter vector after the adaptive information interaction is determined, including: Based on the current number of training times, determining the adaptive speed control factor, the first information interaction factor and the second information interaction factor respectively; For any model parameter vector, multiple other model parameter vectors with the closest Euclidean distance are matched to the model parameter vector to obtain a neighboring group corresponding to the model parameter vector; Obtaining a neighborhood correction parameter vector based on the neighborhood group corresponding to the model parameter vector; Acquire a first adaptive information interaction amount according to the first information interaction factor and the optimal parameter vector; Acquire a second adaptive information interaction amount according to the second information interaction factor and the neighborhood correction parameter vector; Obtaining the training speed in the current training process according to the adaptive speed control factor, the training speed in the previous training process, the first adaptive information interaction amount, and the second adaptive information interaction amount; According to the training speed in the current training process, adaptive information interaction is performed on the model parameter vector to determine the model parameter vector after the adaptive information interaction.

7. The method for monitoring deep coalbed methane ultrasonic plugging removal process according to claim 6 is characterized in that: Adopting an adaptive chaos joint search strategy to jointly search the model parameter vector after the adaptive information interaction, and determining the model parameter vector after the joint search, including: Get the chaotic search factor, and get the joint search factor according to the current training times; Obtain the fitness corresponding to the model parameter vector after the adaptive information interaction, arrange the model parameter vector in descending order of fitness, and obtain the sorted model parameter vector; For any sorted model parameter vector, randomly match the model parameter vector with another model parameter vector to obtain the information interaction parameter vector corresponding to the model parameter vector; According to the sorted model parameter vector, a first joint search parameter vector and a second joint search parameter vector are allocated to the model parameter vector; Obtaining a chaotic search amount according to the chaotic search factor, the first joint search parameter vector and the second joint search parameter vector; Obtaining a joint search amount according to the joint search factor and the information interaction parameter vector corresponding to the model parameter vector; According to the chaotic search amount, the joint search amount and the information interaction parameter vector corresponding to the model parameter vector, a joint search is performed on the model parameter vector after the adaptive information interaction to determine the model parameter vector after the joint search.

8. The method for monitoring deep coalbed methane ultrasonic plugging removal process according to claim 7 is characterized in that: The pivot point search strategy is used to perform neighborhood enhanced search on the model parameter vector after the joint search to determine the model parameter vector after the neighborhood enhanced search, including: Based on the current number of training times, obtain the neighborhood enhancement search factor; For any model parameter vector after the joint search, randomly select another model parameter vector as a fulcrum for the model parameter vector; According to the fulcrum corresponding to the model parameter vector after the joint search and the neighborhood enhanced search factor, a neighborhood enhanced search is performed on the model parameter vector after the joint search to determine the model parameter vector after the neighborhood enhanced search.

9. The method for monitoring deep coalbed methane ultrasonic plugging removal process according to claim 8, characterized in that: The Levy random walk search strategy is used to perform a global random search on the model parameter vector after the neighborhood enhanced search to determine the model parameter vector after the global random search, including: The Levy flight strategy is used to obtain the Levy random walk factor; According to the Levy random walk factor and the optimal parameter vector, combined with an exponential function, a global random search is performed on the model parameter vector after the neighborhood enhanced search to determine the model parameter vector after the global random search.

10. The method for monitoring deep coalbed methane ultrasonic plugging removal process according to claim 5, characterized in that: Repeat the adaptive information interaction strategy, the adaptive chaos joint search strategy, the pivot point search strategy and the Levy random walk search strategy until the training end condition is met. According to the model parameter vector after the global random search in the last training process, the deep coalbed methane ultrasonic deblocking process monitoring model after training is obtained, including: Implement adaptive information interaction strategy, adaptive chaos joint search strategy, pivot point search strategy and Levy random walk search strategy; Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal parameter vector according to the model parameter vector after the global random search in the last training process, otherwise return to the steps of executing the adaptive information interaction strategy, the adaptive chaos joint search strategy, the pivot point surrounding search strategy and the Levy random walk search strategy; The hyperparameters contained in the re-determined optimal parameter vector are used as the final hyperparameters of the deep coalbed methane ultrasonic deblocking process monitoring model to obtain the trained deep coalbed methane ultrasonic deblocking process monitoring model.

Images