Rock burst intensity grade prediction model training method and grade prediction method

Abstract

The invention relates to the technical field of electric digital data processing, discloses a rockburst intensity grade prediction model training method and a grade prediction method, and aims to solve the problem that an existing rockburst intensity prediction scheme is poor in accuracy, consistency and efficiency. Dividing the rockburst intensity into a plurality of rockburst intensity grades; obtaining a plurality of sample data sets corresponding to the feature parameters, and performing normalization processing on the sample data sets; an SOM network is trained according to the sample data sets and the rockburst intensity level, after the SOM network converges, a clustering result is determined according to output of the SOM network, and the clustering result is used for representing the mapping relation between each sample data set and the actual rockburst intensity level; and training a BP network according to a clustering result, and after the BP network converges, obtaining a rockburst intensity grade prediction model. The accuracy, consistency and efficiency of rock burst intensity prediction are improved, and the method is particularly suitable for rock burst prediction of underground engineering.

Description

Technical Field

[0001] The present application relates to the technical field of electrical digital data processing, and in particular to a rockburst intensity grade prediction model training method and a grade prediction method. Background Art

[0002] Rockburst intensity refers to the strength and scale of rock damage when a rockburst occurs, and is used to describe the severity of a rockburst. Rockburst refers to the phenomenon of sudden destruction in the open rock mass in deep underground mining or areas with high tectonic stress. The cause of the occurrence is that the strain energy accumulated in the open rock mass is suddenly and violently released, causing the rock mass to undergo brittle fracture like an explosion. The impact ground pressure causes a large amount of rock to collapse, and produces a huge noise and air wave impact, which can not only destroy the mine, but also the shock wave can endanger the ground buildings. Therefore, the prevention and prediction research of rockburst is of great significance for preventing rockbursts and protecting the safety of workers.

[0003] At present, the main method of predicting rockburst intensity is for geological experts to predict the rockburst intensity level based on actual engineering data. This method is greatly influenced by personal experience and subjectivity, and the accuracy and consistency of rockburst prediction are poor.

[0004] Application publication number CN115130375A discloses a method for predicting rockburst intensity, the main scheme of which is to establish a rockburst intensity prediction sample data set based on the determined rockburst intensity prediction evaluation index and rockburst intensity classification, and then establish a rockburst intensity prediction model based on the predicted sample data set, thereby realizing the grade prediction of rockburst intensity. This method has at least the following problems: First, when establishing the sample data set, it is necessary to manually mark the sample data, that is, the actual rockburst intensity grade corresponding to the actual evaluation index data is determined by geological experts, and the accuracy and consistency of the sample data marking are poor, which leads to the problem of poor accuracy and consistency in the prediction of rockburst intensity; second, when constructing the prediction model, the repeatability of the influence of each evaluation index on the rockburst grade is ignored, resulting in the need to use a sample data set containing a large amount of data for training to obtain relatively accurate prediction results, the efficiency of prediction model training is low, and the prediction effect is poor. Summary of the invention

[0005] The present application aims to solve the problems of poor accuracy, consistency and efficiency in existing rockburst intensity prediction schemes, and proposes a rockburst intensity grade prediction model training method and a grade prediction method.

[0006] The technical solution adopted by this application to solve the above technical problems is: In a first aspect, the present application provides a method for training a grade prediction model for rockburst intensity, the method comprising: Determine the characteristic parameters that affect the rockburst intensity and classify the rockburst intensity into multiple rockburst intensity levels; Acquire multiple sample data groups corresponding to the characteristic parameters, and perform normalization processing on the sample data groups; Training a SOM network according to the sample data groups and the rockburst intensity level, and after the SOM network converges, determining a clustering result according to the output of the SOM network, wherein the clustering result is used to represent a mapping relationship between each sample data group and an actual rockburst intensity level; The BP network is trained according to the clustering results, and after the BP network converges, a grade prediction model of rock burst intensity is obtained.

[0007] Furthermore, the sample data group is normalized, specifically including: The range standardization method is used to transform the sample data in the sample data group into the interval [0, 1]. The transformation formula is as follows: ; in, Indicates The first Sample data, Represents the normalized The first Sample data, Indicates the maximum value of the corresponding sample data, Indicates the minimum value of the corresponding sample data.

[0008] Further, training the SOM network according to the sample data set and the rockburst intensity level specifically includes: S201, setting the number of neurons in the SOM network input layer according to the dimension of the sample data group, and setting the number of neurons in the SOM network competition layer according to the dimension of the rockburst intensity level; S202, randomly select a sample data group and input it into the SOM network input layer, calculate the Euclidean distance between the current sample data group and each neuron in the SOM network competition layer, and take the neuron corresponding to the minimum Euclidean distance as the optimal neuron; S203, determining an optimal neuron neighborhood range including an optimal neuron and its neighboring neurons according to a preset domain radius, and updating connection weights of neurons within the optimal neuron neighborhood range; S204, determine whether the SOM network has reached the convergence condition, if so, the training ends, otherwise, enter S202.

[0009] Furthermore, the calculation formula of the Euclidean distance is as follows: ; in, Indicates The sample data set and the SOM network competition layer The Euclidean distance between neurons is Indicates A sample data set, Indicates the first neurons, represents the norm, express At this moment, the first The input of the neuron The first Sample data, express At this moment, the first The neurons in the competition layer The connection weights between neurons, Indicates the number of sample data in the sample data group.

[0010] Furthermore, the update formula of the connection weight is as follows: ; in, express At this moment, the first The neurons in the competition layer The connection weights between neurons, express The learning rate of the SOM network at this moment.

[0011] Further, training a BP network according to the clustering result specifically includes: S301, inputting the sample data group corresponding to the clustering result into the BP network input layer, the BP network input layer transmits it to the BP network hidden layer, and transmits it to the output layer after being processed by the activation function of the BP network hidden layer, and the BP network output layer calculates the predicted rockburst intensity level corresponding to the current sample data group according to the output result of the BP network hidden layer; S302, calculate the error between the predicted rockburst intensity level and the corresponding actual rockburst intensity level, and determine whether the error is within a preset range. If so, the training ends; otherwise, the network weights and thresholds of the BP network are updated according to the error, and then enter S301.

[0012] Furthermore, the processing formula of the hidden layer of the BP network is as follows: ; in, Represents the hidden layer of BP network The output of a neuron is represents the activation function, Indicates the first The first The first network weight of the neuron corresponding to the sample data, Represents the hidden layer of BP network The first threshold of neurons, Indicates the number of sample data in the sample data group, Represents the number of neurons in the hidden layer of the BP network; The calculation formula for predicting the rockburst intensity level is as follows: ; in, Indicates The predicted rockburst intensity level corresponding to the sample data set is Indicates the output layer of the BP network and the hidden layer of the BP network The neuron corresponding to The second network weights of neurons, Represents the output layer of the BP network The second threshold of neurons, Represents the number of neurons in the output layer of the BP network.

[0013] Furthermore, the update formula of the first network weight is as follows: ; ; in, represents the updated first network weight, represents the first network weight before update, represents the learning rate of BP network, Indicates The first Sample data, Indicates The error corresponding to the sample data set is Indicates The actual rockburst intensity level corresponding to the sample data set; The update formula of the second network weight is as follows: ; in, represents the updated weight of the second network, represents the weight of the second network before updating; The updating formula of the first threshold is as follows: ; in, represents the first threshold after update, represents the first threshold before updating; The updating formula of the second threshold is as follows: ; in, represents the updated second threshold, Indicates the second threshold before updating.

[0014] Furthermore, the characteristic parameters include rock uniaxial compressive strength, rock uniaxial tensile strength, elastic energy index and maximum shear stress of the cave wall.

[0015] In a second aspect, the present application provides a method for predicting the grade of rockburst intensity, the method comprising: Acquire a feature parameter group to be predicted, and perform normalization processing on the feature parameter group; The normalized characteristic parameter group is input into the grade prediction model of rockburst intensity to obtain the rockburst intensity grade corresponding to the characteristic parameter group. The grade prediction model of rockburst intensity is trained by the grade prediction model training method of rockburst intensity as described in the first aspect.

[0016] The beneficial effects of the present application are as follows: the present application trains a rockburst intensity grade prediction model based on the SOM network and the BP network, and uses the trained grade prediction model to predict the rockburst intensity grade, and uses the unsupervised SOM network to automatically form the clustering results of the sample data, without the need to manually label the sample data, thereby avoiding the influence of personal experience and subjectivity, improving the accuracy and consistency of the sample data, and thus improving the accuracy and consistency of the rockburst intensity grade prediction. By utilizing the nonlinear mapping and strong feature recognition capabilities of the BP network, the prediction model can better reflect the correlation between the rockburst characteristic parameters, and does not rely on a large amount of sample data during the model training process, thereby improving the accuracy and training efficiency of the prediction model, thereby improving the accuracy and efficiency of the rockburst intensity grade prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a rockburst intensity grade prediction model training method provided in an embodiment of the present application; Figure 2 A flow chart of a SOM network training method provided in an embodiment of the present application; Figure 3 A flowchart of a BP network training method provided in an embodiment of the present application; Figure 4 A schematic flow chart of a method for predicting the grade of rockburst intensity provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0019] In some of the processes described in the specification of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or in parallel, and the sequence number of the operation is only used to distinguish between different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to different types.

[0020] The technical solution of the embodiment of the present application is applicable to application scenarios where rockburst level prediction is required, such as rockburst tendency prediction of underground projects such as water diversion tunnels, water delivery tunnels, underground power plants, highway tunnels, railway tunnels, and mines.

[0021] The current rockburst intensity prediction scheme is to obtain actual engineering data and its corresponding rockburst intensity level, and then establish a sample data set, use the sample data set to train a prediction model, and then use the prediction model to predict the rockburst intensity.

[0022] The inventors found through research that in the existing rockburst intensity prediction scheme, when establishing the sample data set, it is necessary to manually label the sample data, that is, to determine the actual rockburst intensity level corresponding to the actual evaluation index data. Affected by personal experience and subjectivity, the accuracy and consistency of the sample data labeling are poor. At the same time, the neural network corresponding to the existing prediction model tends to ignore the repeatability of the impact of each evaluation index on the rockburst level, which also leads to poor accuracy of the prediction results.

[0023] Based on this, the technical solution of the present application is proposed. In an embodiment of the present application, characteristic parameters that affect the rockburst intensity are determined, and the rockburst intensity is divided into multiple rockburst intensity levels; multiple sample data groups corresponding to the characteristic parameters are obtained, and the sample data groups are normalized; a SOM network is trained according to the sample data groups and the rockburst intensity levels, and after the SOM network converges, a clustering result is determined according to the output of the SOM network, and the clustering result is used to represent the mapping relationship between each sample data group and the actual rockburst intensity level; a BP network is trained according to the clustering result, and after the BP network converges, a grade prediction model for the rockburst intensity is obtained.

[0024] It can be understood that the SOM network (Self-Organizing Maps) is an unsupervised artificial neural network. The SOM network uses a competitive learning strategy and relies on the competition between neurons to gradually optimize the network. The BP network (Backpropagation Neural Network) is a multi-layer feedforward neural network based on the error back propagation algorithm. It has the ability to classify arbitrarily complex patterns and excellent multi-dimensional function mapping capabilities.

[0025] Specifically, the embodiment of the present application uses an unsupervised SOM network to automatically form clustering results of sample data, without the need to manually label sample data, avoiding the influence of personal experience and subjectivity, improving the accuracy and consistency of sample data, and thus improving the accuracy and consistency of rockburst intensity level prediction. By utilizing the nonlinear mapping and strong feature recognition capabilities of the BP network, the prediction model can better reflect the correlation between rockburst characteristic parameters, and does not rely on a large amount of sample data during the model training process, thereby improving the accuracy and training efficiency of the prediction model, thereby improving the accuracy and efficiency of rockburst intensity level prediction.

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0027] Figure 1 A flow chart of a rockburst intensity prediction model training method provided in an embodiment of the present application is shown. Figure 1 , the method comprises the following steps: S101. Determine characteristic parameters that affect rockburst intensity, and divide the rockburst intensity into multiple rockburst intensity levels.

[0028] In the embodiment of the present application, the characteristic parameters include the uniaxial compressive strength of rock, the uniaxial tensile strength of rock, the elastic energy index and the maximum shear stress of the cave wall. The rockburst intensity level is divided into level 1, level 2, level 3 and level 4, where level 1 means no rockburst, level 2 means slight rockburst, level 3 means moderate rockburst and level 4 means severe rockburst.

[0029] S102: Acquire multiple sample data groups corresponding to the characteristic parameters, and perform normalization processing on the sample data groups.

[0030] In practical applications, after obtaining the characteristic parameters in the actual project, multiple corresponding sample data groups are generated. The sample data set is , Indicates The first Sample data, Indicates the number of sample data in the sample data group.

[0031] In the embodiment of the present application, the sample data group is normalized, specifically including: The range standardization method is used to transform the sample data in the sample data group into the interval [0, 1]. The transformation formula is as follows: ; in, Indicates The first Sample data, Represents the normalized The first Sample data, Indicates the maximum value of the corresponding sample data, Indicates the minimum value of the corresponding sample data.

[0032] In the embodiment of the present application, The sample data set is ,Right now =4, Respectively represent The rock uniaxial compressive strength, rock uniaxial tensile strength, elastic energy index and maximum shear stress of the cave wall of the sample data set are collected. The present application embodiment collects 10 sets of sample data sets corresponding to the real characteristic parameters, please refer to Table 1.

[0033] Table 1 Data table of sample data sets corresponding to 10 sets of real feature parameters

[0034] S103, training a SOM network according to the sample data set and the rockburst intensity level, and after the SOM network converges, determining a clustering result according to an output of the SOM network.

[0035] See also Figure 2 In the embodiment of the present application, the SOM network is trained according to the sample data set and the rockburst intensity level, specifically including the following steps: S201. Setting the number of neurons in the SOM network input layer according to the dimension of the sample data group, and setting the number of neurons in the SOM network competition layer according to the dimension of the rockburst intensity level.

[0036] It can be understood that the input layer and competition layer of the SOM network are the core components of its network structure. The input layer is responsible for receiving input data, while the competition layer processes and classifies the input data through competition between neurons. The number of neurons in the input layer is determined by the dimension of the input sample data, and each feature parameter corresponds to a neuron; the number of neurons in the competition layer is determined by the dimension of the classification result.

[0037] Based on this, in an embodiment of the present application, the dimension of the sample data group is 4, that is, each sample data group includes sample data corresponding to 4 characteristic parameters, namely, sample data corresponding to the uniaxial compressive strength of rock, the uniaxial tensile strength of rock, the elastic energy index and the maximum shear stress of the cave wall, so the SOM network input layer includes 4 neurons; the dimension of the rock burst intensity level is 4, that is, it includes 4 rock burst intensity levels, so the SOM network competition layer also includes 4 neurons.

[0038] S202, randomly select a sample data group and input it into the SOM network input layer, calculate the Euclidean distance between the current sample data group and each neuron in the SOM network competition layer, and take the neuron corresponding to the minimum Euclidean distance as the optimal neuron.

[0039] It can be understood that the size of the Euclidean distance represents the mapping relationship between the input data and the rockburst intensity level. The smaller the Euclidean distance, the higher the matching degree between the input data and the corresponding rockburst intensity level, and vice versa. The role of the Euclidean distance in the competition layer is to perform cluster analysis on the input data to ensure that the neuron with the smallest Euclidean distance is selected as the winning neuron.

[0040] In the embodiment of the present application, the calculation formula of the Euclidean distance is as follows: ; in, Indicates The sample data set and the SOM network competition layer The Euclidean distance between neurons is Indicates A sample data set, Indicates the first neurons, represents the norm, express At this moment, the first The input of the neuron The first Sample data, express At this moment, the first The neurons in the competition layer The connection weights between neurons are Indicates the number of sample data in the sample data group.

[0041] S203, determining an optimal neuron neighborhood range including an optimal neuron and its neighboring neurons according to a preset domain radius, and updating connection weights of neurons within the optimal neuron neighborhood range.

[0042] After determining the optimal neuron neighborhood range, the connection weights can be updated according to the learning rate of the SOM network, and the optimal neuron neighborhood range is gradually reduced with each update iteration.

[0043] In the embodiment of the present application, the update formula of the connection weight is as follows: ; in, express At this moment, the first The neurons in the competition layer The connection weights between neurons are express The learning rate of the SOM network at this moment.

[0044] S204, determine whether the SOM network has reached the convergence condition, if so, the training ends, otherwise, enter S202.

[0045] After the SOM network reaches the convergence condition, the preliminary clustering of the mapping relationship between the sample data group and the rockburst intensity level can be achieved, and the clustering results used to represent the mapping relationship between each sample data group and the actual rockburst intensity level can be obtained.

[0046] S104, training a BP network according to the clustering results, and obtaining a grade prediction model of rockburst intensity after the BP network converges.

[0047] See also Figure 3In the embodiment of the present application, training the BP network according to the clustering result specifically includes the following steps: S301. Input the sample data group corresponding to the clustering result into the BP network input layer, which passes it to the BP network hidden layer, and then passes it to the output layer after being processed by the activation function of the BP network hidden layer. The BP network output layer calculates the predicted rockburst intensity level corresponding to the current sample data group according to the output result of the BP network hidden layer.

[0048] It can be understood that the hidden layer of the BP network is a neural network layer located between the input layer and the output layer, which is used to extract features and perform nonlinear transformations. The role of the hidden layer is to abstract and transform the input data for better prediction.

[0049] In the embodiment of the present application, the processing formula of the hidden layer of the BP network is as follows: ; in, Represents the hidden layer of BP network The output of a neuron is represents the activation function, , Indicates the first The first The first network weight of the neuron corresponding to the sample data, Represents the hidden layer of BP network The first threshold of neurons, Indicates the number of sample data in the sample data group, Represents the number of neurons in the hidden layer of the BP network; In the BP network, the output layer calculates the rockburst intensity level according to the output results of each neuron in the hidden layer, and uses it as the predicted rockburst intensity level. The calculation formula is as follows: ; in, Indicates The predicted rockburst intensity level corresponding to the sample data set is Indicates the output layer of the BP network and the hidden layer of the BP network The neuron corresponding to The second network weights of neurons, Represents the output layer of the BP network The second threshold of neurons, Represents the number of neurons in the output layer of the BP network.

[0050] S302, calculate the error between the predicted rockburst intensity level and the corresponding actual rockburst intensity level, and determine whether the error is within a preset range. If so, the training ends; otherwise, the network weights and thresholds of the BP network are updated according to the error, and then enter S301.

[0051] In order to ensure the accuracy of the prediction model, the network weights and thresholds in the BP network need to be updated according to the error between the predicted rockburst intensity level and the corresponding actual rockburst intensity level.

[0052] Specifically, the error between the predicted rockburst intensity level and the corresponding actual rockburst intensity level is first calculated, and the calculation formula is as follows: ; in, Indicates The error between the predicted rockburst intensity level and the actual rockburst intensity level corresponding to the sample data group is Indicates The actual rockburst intensity level corresponding to the sample data set.

[0053] In practical applications, when the error is within the preset range, it means that the error is small, and the BP network converges; when the error is not within the preset range, it means that the error is large. At this time, it is necessary to update the network weights and thresholds according to the error and continue to iterate until the expected effect is achieved.

[0054] In the embodiment of the present application, the network weight includes the first network weight of the neuron in the BP network input layer and the second network weight of the neuron in the BP network output layer, and the threshold includes the first threshold of the neuron in the BP network hidden layer and the second threshold of the neuron in the BP network output layer, wherein the threshold refers to the minimum stimulation level required for the neuron to be activated, and the threshold determines when the neuron is activated. When the weighted sum of the input signal minus the threshold is greater than 0, the neuron is activated; otherwise, the neuron remains in an inactive state.

[0055] In the embodiment of the present application, the updating formula of the first network weight is as follows: ; in, represents the updated first network weight, represents the first network weight before update, represents the learning rate of BP network, Indicates The first Sample data.

[0056] The update formula of the second network weight is as follows: ; in, represents the updated weight of the second network, Represents the weight of the second network before updating.

[0057] The updating formula of the first threshold is as follows: ; in, represents the first threshold after update, Indicates the first threshold before updating.

[0058] The updating formula of the second threshold is as follows: ; in, represents the updated second threshold, Indicates the second threshold before updating.

[0059] After the BP network training is completed, a grade prediction model of rock burst intensity is obtained, and the grade prediction model can be used to predict the grade of rock burst intensity. Figure 4 A schematic diagram showing a flow chart of a rockburst intensity prediction method provided in an embodiment of the present application is shown. Figure 4 , the method comprises the following steps: S401: Obtain a feature parameter group to be predicted, and perform normalization processing on the feature parameter group.

[0060] S402: Input the normalized characteristic parameter group into a rockburst intensity level prediction model to obtain the rockburst intensity level corresponding to the characteristic parameter group.

[0061] In practical applications, the characteristic parameter group of the project to be predicted is first obtained, and the characteristic parameter group is normalized using the same method as the sample data group. The normalized characteristic parameter group is then input into the rockburst intensity grade prediction model trained by the rockburst intensity grade prediction model training method described in the embodiment of the present application, so that the rockburst intensity grade of the project to be predicted can be obtained, thereby realizing the rockburst intensity grade prediction.

[0062] In summary, the rockburst intensity level prediction model training method and level prediction method provided in the embodiment of the present application use an unsupervised SOM network to automatically form a clustering result of sample data, without the need to manually label sample data, thereby avoiding the influence of personal experience and subjectivity, improving the accuracy and consistency of sample data, and thus improving the accuracy and consistency of rockburst intensity level prediction. By utilizing the nonlinear mapping of the BP network and its strong feature recognition capability, the prediction model can better reflect the correlation between rockburst characteristic parameters, does not rely on a large amount of sample data during the model training process, improves the accuracy of the prediction model and the training efficiency, and thus improves the accuracy and efficiency of rockburst intensity level prediction.

Claims

1. A rockburst intensity level prediction model training method, characterized in that: The method comprises: Determine the characteristic parameters that affect the rockburst intensity and classify the rockburst intensity into multiple rockburst intensity levels; Acquire multiple sample data groups corresponding to the characteristic parameters, and perform normalization processing on the sample data groups; Training a SOM network according to the sample data groups and the rockburst intensity level, and after the SOM network converges, determining a clustering result according to the output of the SOM network, wherein the clustering result is used to represent a mapping relationship between each sample data group and an actual rockburst intensity level; The BP network is trained according to the clustering results, and after the BP network converges, a grade prediction model of rock burst intensity is obtained.

2. The rockburst intensity grade prediction model training method according to claim 1, characterized in that: The sample data group is normalized, specifically comprising: The range standardization method is used to transform the sample data in the sample data group into the interval [0, 1]. The transformation formula is as follows: ； in, Indicates The first Sample data, Represents the normalized The first Sample data, Indicates the maximum value of the corresponding sample data, Indicates the minimum value of the corresponding sample data.

3. The rockburst intensity grade prediction model training method according to claim 1, characterized in that: Training the SOM network according to the sample data set and the rockburst intensity level specifically includes: S201, setting the number of neurons in the SOM network input layer according to the dimension of the sample data group, and setting the number of neurons in the SOM network competition layer according to the dimension of the rockburst intensity level; S202, randomly select a sample data group and input it into the SOM network input layer, calculate the Euclidean distance between the current sample data group and each neuron in the SOM network competition layer, and take the neuron corresponding to the minimum Euclidean distance as the optimal neuron; S203, determining an optimal neuron neighborhood range including an optimal neuron and its neighboring neurons according to a preset domain radius, and updating connection weights of neurons within the optimal neuron neighborhood range; S204, determine whether the SOM network has reached the convergence condition, if so, the training ends, otherwise, enter S202.

4. The rockburst intensity grade prediction model training method according to claim 3 is characterized in that: The calculation formula of the Euclidean distance is as follows: ； in, Indicates The sample data set and the SOM network competition layer The Euclidean distance between neurons is Indicates A sample data set, Indicates the first neurons, represents the norm, express At this moment, the first The input of the neuron The first Sample data, express At this moment, the first The neurons in the competition layer The connection weights between neurons are Indicates the number of sample data in the sample data group.

5. The rockburst intensity grade prediction model training method according to claim 4, characterized in that: The update formula of the connection weight is as follows: ； in, express At this moment, the first The neurons in the competition layer The connection weights between neurons are express The learning rate of the SOM network at this moment.

6. The rockburst intensity grade prediction model training method according to claim 1, characterized in that: Training the BP network according to the clustering results specifically includes: S301, inputting the sample data group corresponding to the clustering result into the BP network input layer, the BP network input layer transmits it to the BP network hidden layer, and transmits it to the output layer after being processed by the activation function of the BP network hidden layer, and the BP network output layer calculates the predicted rockburst intensity level corresponding to the current sample data group according to the output result of the BP network hidden layer; S302, calculate the error between the predicted rockburst intensity level and the corresponding actual rockburst intensity level, and determine whether the error is within a preset range. If so, the training ends; otherwise, the network weights and thresholds of the BP network are updated according to the error, and then enter S301.

7. The rockburst intensity grade prediction model training method according to claim 6, characterized in that: The processing formula of the hidden layer of the BP network is as follows: ； in, Represents the hidden layer of BP network The output of each neuron is represents the activation function, Indicates the first The first The first network weight of the neuron corresponding to the sample data, Represents the hidden layer of BP network The first threshold of neurons, Indicates the number of sample data in the sample data group, Represents the number of neurons in the hidden layer of the BP network; The calculation formula for predicting the rockburst intensity level is as follows: ； in, Indicates The predicted rockburst intensity level corresponding to the sample data set is Indicates the output layer of the BP network and the hidden layer of the BP network The neuron corresponding to The second network weights of neurons, Represents the output layer of the BP network The second threshold of neurons, Represents the number of neurons in the output layer of the BP network.

8. The rockburst intensity grade prediction model training method according to claim 7, characterized in that: The update formula of the first network weight is as follows: ； ； in, represents the updated first network weight, represents the first network weight before update, represents the learning rate of BP network, Indicates The first Sample data, Indicates The error corresponding to the sample data set is Indicates The actual rockburst intensity level corresponding to the sample data set; The update formula of the second network weight is as follows: ； in, represents the updated weight of the second network, represents the weight of the second network before updating; The updating formula of the first threshold is as follows: ； in, represents the first threshold after update, represents the first threshold before updating; The updating formula of the second threshold is as follows: ； in, represents the updated second threshold, Indicates the second threshold before updating.

9. The rockburst intensity grade prediction model training method according to claim 1, characterized in that: The characteristic parameters include the uniaxial compressive strength of rock, the uniaxial tensile strength of rock, the elastic energy index and the maximum shear stress of the cave wall.

10. A method for predicting the grade of rockburst intensity, characterized in that: The method comprises: Acquire a feature parameter group to be predicted, and perform normalization processing on the feature parameter group; The normalized characteristic parameter group is input into the grade prediction model of rockburst intensity to obtain the rockburst intensity grade corresponding to the characteristic parameter group, and the grade prediction model of rockburst intensity is trained by the grade prediction model training method of rockburst intensity as described in any one of claims 1 to 9.

Images