An intelligent prediction method for the height of two zones of overburden in working face based on drilling parameters

Abstract

The present invention relates to the technical field of prediction of the height of two zones of coal mining working faces, and in particular to an intelligent prediction method for the height of two zones of overburden in a working face based on drilling parameters. According to different drilling conditions, measuring stations are selected at different positions to perform drilling operations; real-time monitoring data are acquired, and data preprocessing is performed on the real-time monitoring data and the geomechanical parameters of the rock mass in the two zones, and a training data set and a test data set are generated; a learning model is optimized using an optimization algorithm of an Adam optimizer with a preliminary adaptive learning rate, the learning model is trained using a training data set, and a three-dimensional space algorithm is built into the learning model; the accuracy of the learning model is tested using test set data; and a hyperparameter optimization algorithm and an L2 regularization technology are used to optimize the learning model, and the height of the two zones is predicted using a learning model that meets accuracy requirements.

Description

Technical Field

[0001] The invention relates to the technical field of overburden height prediction of two zones of coal mining working face, and in particular to an intelligent prediction method for overburden height of two zones of working face based on drilling parameters. Background Art

[0002] Coal mining activities in my country mainly rely on underground mining methods. In the process of coal resource mining, the rock strata above the roof of the goaf will form a specific "three-zone" structure after collapse, which are the collapse zone, fracture zone and curved sinking zone from bottom to top. Among them, the height of the collapse zone and the fracture zone (collectively referred to as the water-conducting fracture zone, referred to as the "two zones") is a crucial reference for formulating the retention plan of safe coal and rock pillars in the outcrop area and designing safe mining strategies under water bodies. Therefore, in coal mining operations, it is particularly important to determine the height of the "two zones" of the roof of the working face.

[0003] At present, there are three main methods used to determine the height of the "two belts" at home and abroad: empirical formula calculation, computer simulation and field measurement. However, these methods all have certain limitations. The prediction accuracy of the empirical formula calculation method is often greatly reduced in mines with complex geological conditions; the computer simulation method is difficult for ordinary mine technicians to use effectively because of the cumbersome model construction and parameter acquisition process; as for the field measurement method, although it includes a variety of methods such as the double-end water plugging device tilted side leakage method, optical fiber method and electromagnetic method, the double-end water plugging device tilted side leakage method is widely favored for its high precision and strong operability, but it is also often challenged by problems such as hole collapse, inability to install equipment or test leakage under the conditions of soft rock roof. Other measurement methods are also susceptible to such factors.

[0004] In view of this, it is crucial to develop an intelligent prediction method that can overcome the influence of adverse factors such as hole collapse for the accurate prediction of the height of the "two zones" of the overburden of the working face. As a cutting-edge technical means, the measurement while drilling technology can capture and monitor a series of key parameters such as drilling speed, rotation speed, torque and drilling pressure in real time during the drilling operation, and then accurately analyze and predict the lithology and stress state based on these parameters. When the working face completes the mining task, the rock strata above the goaf will experience different degrees of deformation and damage, and these deformations and damages have their own characteristics in different zones. Specifically, the distribution of rock cracks, the degree of rock looseness, and the rock stress state in each zone will show significant differences. When the drill rig passes through these zones with different characteristics, its drilling speed, rotation speed, torque, and the pressure on the drill bit will change significantly. It is based on this unique principle that the measurement while drilling technology can play an important role in the prediction of the height of the "two zones" of the roof of the goaf.

[0005] Compared with the traditional "two-zone" height prediction method, the intelligent prediction method based on drilling parameters only needs to use the drilling parameters obtained during the drilling process to realize the intelligent prediction of the "two-zone" height of the overburden of the working face. This method not only significantly reduces the workload of subsequent on-site measurements, but also effectively avoids the measurement failure caused by factors such as hole collapse. More importantly, this feature makes this method more applicable under soft rock roof conditions. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcoming that the existing two-zone height prediction method is difficult to predict in soft rock conditions, and then propose an intelligent prediction method for the two-zone height of the overburden rock of the working face based on drilling parameters, comprising the following steps:

[0007] Step 1: Select measuring stations at different locations for drilling operations according to different drilling conditions;

[0008] Step 2: Acquire real-time monitoring data, perform data preprocessing on the real-time monitoring data and the rock mass geomechanical parameters in the two zones, and generate a training data set and a test data set;

[0009] Step 3: Use the Adam optimizer to optimize the learning model, use the training data set to train the learning model, and build a three-dimensional space algorithm into the learning model;

[0010] Step 4: Use the test set data to test the accuracy of the learning model, and use the hyperparameter optimization algorithm and L2 regularization technology to optimize the model;

[0011] Step 5: Use a learning model that meets the accuracy requirements to predict the height of the two zones of overburden at the working face.

[0012] In a preferred embodiment, in step 1, drilling operations are performed on the overlying strata of the goaf in adjacent tunnels, and the drilling parameters are recorded, and the steps are as follows:

[0013] S11. Preliminary determination of drilling depth, directivity arrangement and station spacing;

[0014] S12. Drilling is performed on the overlying rock strata of the goaf in the roof of the adjacent tunnel of the goaf, and the borehole penetrates the water-conducting fracture zone of the overlying rock strata of the goaf and enters the upper curved sinking zone for 5 to 10 m;

[0015] S13, recording all drilling parameters of the drilling rig and deriving the drilling parameters obtained during the drilling.

[0016] In a preferred embodiment, in step 2, the real-time monitoring data is cleaned to obtain a screened data set, and a stepwise feature selection algorithm is used to implement feature screening, starting from the initial feature, and continuously adding new features until the stop condition is reached. The specific steps are as follows:

[0017] (1) Initialize the algorithm and set the initial feature subset to an empty set;

[0018] (2) Setting the stopping condition: adding new features several times in a row but the accuracy does not improve any more;

[0019] (3) Use the cross-validation algorithm to evaluate each unselected feature separately and select the best performing feature to add to the current feature subset;

[0020] (4) After adding new features to the current feature subset, re-evaluate the performance of the feature combination. If the performance improves, retain the new features; otherwise, exclude the new features.

[0021] (5) Repeat steps (3) and (4) until the preset stop condition is reached.

[0022] In a preferred embodiment, the learning model training process in step 3 is as follows:

[0023] (1) Data input and initialization:

[0024] ①Time series input: ,in is the input vector at time step t, d is the feature dimension, and T is the number of input vectors;

[0025] ② Drilling parameter input: The drilling parameters in the training data set are used as initial data to input the model;

[0026] ③ Initial hidden state: Initialize the input time series and drilling parameters;

[0027] (2) Gated recurrent unit calculation:

[0028] ① Combine the calculated and newly input drilling parameters by resetting the gate:

[0029] ;

[0030] in: is the output parameter of the reset gate; is the Sigmoid activation function; is the weight matrix of the reset gate; is the hidden state of the previous time step t-1; is the input parameter of the current time step t; is the bias term for the reset gate;

[0031] ② Filter the calculated while-drilling parameters through the update gate and save the filtered parameters to the current state:

[0032] ;

[0033] in: is the output of the update gate; is the weight matrix of the update gate; is the bias term of the update gate;

[0034] ③ Combine the current input parameters and reset the parameters of the gate to make it a candidate hidden state:

[0035] ;

[0036] in: is a candidate hidden state; Represents element-wise multiplication; is the weight matrix of the candidate hidden states; is the bias term of the candidate hidden state;

[0037] ④Weighted combination of update gate and candidate hidden state parameters to make it the final hidden state:

[0038] ;

[0039] in: is the hidden state at the current time step;

[0040] (3) Model output:

[0041] The output of the GRU model is mapped to the target variable through a fully connected layer:

[0042] ;

[0043] in: is the model at time step The predicted value of is the weight matrix of the output layer; is the bias term of the output layer;

[0044] (4) Calculate the loss function:

[0045] Minimize the error between the predicted value and the true value, and use the mean square error MSE as the loss function:

[0046] ;

[0047] in: are the true height values ​​of the two zones; is the predicted value of the model; is the sample size;

[0048] (5) Optimization model:

[0049] The Adam optimizer is used to update the model parameters through the gradient descent method to minimize the loss function. The parameter update formula is:

[0050] ;

[0051] in: are model parameters; is the learning rate; is the gradient of the loss function with respect to the parameters.

[0052] In a preferred embodiment, a three-dimensional space algorithm is built in to directly calculate the borehole length obtained by the drilling parameters as the two-zone height. The specific calculation steps are as follows:

[0053] (1) Adjust the drilling inclination , drilling azimuth Feed the trained model as input parameters;

[0054] (2) Calculate the horizontal distance between the two crack locations and the monitoring point D :

[0055] ;

[0056] Where: is the horizontal distance between the two crack locations and the monitoring point / m;

[0057] (3) Calculate the vertical distance d between the two crack positions and the monitoring point along the tunnel direction:

[0058] ;

[0059] Where: is the vertical distance between the two crack positions and the monitoring point along the tunnel direction; is the drilling azimuth;

[0060] (4) Calculate the strike distance a between the two crack positions and the monitoring point along the tunnel direction: ;

[0061] (5) Calculate the vertical height of the two crack locations from the monitoring point :

[0062] ;

[0063] Where: is the vertical height of the two cracks from the monitoring point; L is the length of the crack area from the monitoring point; is the drilling inclination angle.

[0064] In a preferred embodiment, in step 4, when the model is tested using a test data set, the square absolute error is used as an evaluation index to evaluate the result, and its specific analytical formula is as follows:

[0065] ;

[0066] If the model does not meet the test accuracy, the model parameters are optimized by adjusting the hyperparameter method of the number of gated recurrent unit layers and the number of hidden units and the Dropout regularization method, which specifically includes the following steps:

[0067] ① Increase the number of gated recurrent unit layers. When optimizing the model, start from layer 1 and gradually increase the number of layers. For L-layer GRU, The hidden state of is calculated as:

[0068] ;

[0069] Among them: input layer ; is the final hidden state;

[0070] ② Adjust the number of hidden units to balance the model’s expressiveness and computational complexity;

[0071] ③Use L2 regularization technology to limit the size of model parameters and add L2 regularization terms to the loss function:

[0072] ;

[0073] in, is the new loss function, is the original loss function, is the regularization coefficient, is the i-th model parameter.

[0074] In a preferred embodiment, data cleaning is used in conjunction with an optimization algorithm for built-in numerical calculations to preprocess the data.

[0075] In a preferred embodiment, the feed speed and rotation speed of the drilling parameters are used as initial input parameters of the drilling rig, and the displacement, pressure, torque and power are used as output parameters of the drilling rig to predict the height of the two belts.

[0076] Compared with the prior art, the present invention has the following advantages:

[0077] (1) The prediction method provided by the present invention can solve the problem of difficulty in predicting the height of the two zones in a soft rock environment, and realize rapid and accurate prediction of the height of the two zones in a soft rock environment. It can be used to guide the design of the protection of coal and rock pillars and safe mining under water bodies, and can be better applied to various working faces such as the retention of coal pillars and the retention of goafs, providing strong support for underground engineering construction.

[0078] (2) Feature selection is performed through a stepwise feature selection algorithm to solve the problem that the model size is large and the efficiency is reduced when the data is directly used as input parameters due to the large number of operation data and complex types. The model is optimized by using an adaptive learning rate optimization algorithm to improve the efficiency and stability of the optimization algorithm, making the model more generalizable.

[0079] (3) The present invention is based on on-site drilling operations, establishes a learning prediction model and a built-in three-dimensional space algorithm, and utilizes learning technology to greatly improve the prediction accuracy, further improve the accuracy of the two-zone height prediction method, reduce the prediction time, and reduce the difficulty of operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A schematic diagram of the two zones of overburden and drilling arrangement for the coal pillar;

[0081] Figure 2 This is a schematic diagram of the two overburden zones and drilling arrangement of the gob-side entry retention working face;

[0082] Figure 3 This is a schematic diagram of the two-zone height calculation of the model's built-in three-dimensional space algorithm;

[0083] Figure 4 It is a flow chart of the intelligent prediction method of the working face two-zone height based on drilling parameters of the present invention;

[0084] Figure 5 It is a schematic diagram of a specific implementation process of the method for intelligently predicting the height of two zones of a working face based on drilling parameters of the present invention.

[0085] Explanation of numbers in the figure: 1-MWD borehole; 2-tunnels adjacent to the goaf; 3-coal pillars; 4-caving zone; 5-fracture zone; 6-other rock masses; 7-walls of goaf-retained lanes; 8-lane roof plane; 9-lane direction. DETAILED DESCRIPTION

[0086] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, 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. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0087] In the drawings of the specific embodiments of the present invention, in order to better and more clearly describe the working principles of the various components in the system, the connection relationship of the various parts in the device is shown, which only clearly distinguishes the relative position relationship between the various components, and cannot constitute a limitation on the signal transmission direction, connection sequence and structural size, size and shape of each part within the component or structure.

[0088] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0089] The present invention is further described below by way of embodiments in conjunction with the accompanying drawings, but is not limited thereto and is based on the embodiments of the present invention.

[0090] Embodiment 1

[0091] This embodiment provides a method for intelligently predicting the height of two zones of a working face based on drilling parameters, and the specific steps are as follows:

[0092] Step 1: Select measuring stations at different locations for drilling operations according to different drilling conditions.

[0093] The on-site drilling operation drills into the overburden of the goaf and records the drilling parameters. The steps are as follows:

[0094] S11. Preliminarily determine the drilling depth, directional arrangement and measuring station spacing.

[0095] The required drilling depth, directional arrangement and station spacing were preliminarily determined using empirical formulas for predicting the height of the two zones, previous measured experience, the displacement and subsidence characteristics of the two zones and the mutual influence between different measuring stations.

[0096] Before selecting the empirical formula for the height of the two zones, the lithology of the roof overburden should be analyzed through tests and other means to obtain the compressive strength of the roof rock, and classify it into one of hard, medium-hard and weak, and select the corresponding empirical formula according to its different strengths. The specific selection method is based on the maximum height calculation formula of the coal seam mining collapse zone and fracture zone, see Table 1. Secondly, based on previous measured experience, the height of the two zones of soft overburden is 10-15 times the mining height, the medium-hard overburden is 10-20 times the mining height, and the hard overburden is 15-25 times the mining height. Therefore, in order to ensure that the borehole can smoothly penetrate the two zones, the drilling depth should take the larger value of the empirical formula calculation and the field measured experience.

[0097] It can be known from the mine pressure theory that the influence area of ​​the tunnel is generally 6 times the diameter. Therefore, in order to prevent mutual influence between different boreholes, the borehole spacing should be greater than or equal to 6 times the borehole diameter. The spacing between measuring stations should be arranged according to the progress of the project. For mines with good geological conditions, the spacing between measuring stations can be set to 10 meters, starting from 20 meters behind the working face.

[0098] In order to improve the accuracy of two-zone height monitoring and balance the anisotropy of the rock itself, drilling can be carried out at multiple points. In order to prevent mutual influence between boreholes, according to the plastic zone range theory, the borehole spacing should be greater than or equal to 6 times the borehole diameter. In order to reduce the error caused by the differences in the rock itself, the boreholes should be as close as possible. Therefore, the borehole spacing is taken as 6 times the borehole diameter, that is, 6d.

[0099] Table 1 Calculation formula for the maximum height of the collapse zone and fracture zone in coal seam mining:

[0100]

[0101] In the formula For cumulative mining height, is the maximum height of the collapse zone, is the maximum height of the fracture zone.

[0102] S12. Drill into the overlying rock strata of the goaf in the adjacent roadway roof. The borehole penetrates the water-conducting fracture zone of the overlying rock strata of the goaf and enters the upper curved sinking zone for 5 to 10 m. Figure 1 , Figure 2 shown.

[0103] Due to the disturbance caused by mining, the overburden has two zones, namely the collapse zone and the fracture zone. The overburden is drilled using a measurement while drilling rig. Since the working face has been mined out, the overburden on the working face will sink until the collapse zone collapses and a stable structure is re-formed. The rock properties of the collapse zone and the fracture zone will change significantly, so the drilling parameters will fluctuate when the drill passes through these areas.

[0104] The measurement while drilling rig in S12 is a special drilling rig which generally includes four parts: drilling system, loading system, pressure chamber device and monitoring control system. It is a drilling rig equipped with measurement while drilling function. In addition to the functions of ordinary drilling rigs, it also has the function of monitoring drilling data such as the pressure on the drill bit, drilling speed and drilling depth of the drilling rig.

[0105] S13, recording all drilling parameters of the drilling rig and deriving the drilling parameters obtained during the drilling.

[0106] Drill the overlying rock strata in the goaf and record the initial drilling parameters during the drilling process. Since the working face has been mined, the overlying rock strata on the working face will sink until the collapse zone collapses and a stable structure is re-formed. The rocks in the collapse zone will be obviously broken. When the drill rig passes through these broken areas, the drilling parameters of the drill rig will fluctuate violently. The drilling distance at the start and end of the fluctuation can be recorded to get the length of the borehole in the collapse zone. There is no obvious crushing of the rock in the fracture zone, so the drilling parameters of the drill rig do not change significantly. Only by comparing with the drilling parameters obtained before mining can the length of the borehole in the fracture zone be obtained, and then the height of the two zones can be obtained by calculation.

[0107] Step 2: Obtain real-time monitoring data, perform data preprocessing on the real-time monitoring data and the rock geomechanical parameters in the two zones, and generate training data sets and test data sets.

[0108] The real-time monitoring data are the drilling parameters that are displayed, recorded and exported in real time during the drilling operation. The drilling parameters include: displacement, feed pressure, rotational pressure, torque, power, feed speed and rotational speed; displacement refers to the drilling depth of the drill pipe, feed pressure refers to the hydraulic oil pressure of the feed cylinder, rotational pressure refers to the motor hydraulic oil pressure, both rotational pressure and feed pressure are measured by sensors, power refers to the power supplied, feed speed refers to the speed at which the feed mechanism feeds the drill bit during drilling, and rotational speed refers to the rotational speed of the spindle.

[0109] The preprocessing process is to clean the real-time monitoring data to obtain the screened data set. Since the data detected by the operation is numerous and complex, directly using it as an input parameter can easily make the model bulky and reduce efficiency. Therefore, a stepwise feature selection algorithm is used to implement feature screening to improve the model learning efficiency. The monitoring data is used as the input parameter of the model after feature selection, and the rock geomechanical parameters are used as the output parameter of the model after screening. The screened input data set and output data set are normalized and processed by sliding windows to generate sample data, and the sample data are used to generate training data set and test data set in a ratio of 7:3.

[0110] The principle of the stepwise feature selection algorithm is to gradually add features, starting from the initial feature and continuously adding new features until the stopping condition is reached. The specific steps are as follows:

[0111] (1) Initialize the algorithm and set the initial feature subset to an empty set;

[0112] (2) Setting the stopping condition as adding new features several times in a row without improving the accuracy;

[0113] (3) Use the cross-validation algorithm to evaluate each unselected feature separately and select the best performing feature to add to the current feature subset;

[0114] (4) After adding new features to the current feature subset, re-evaluate the performance of the feature combination. If the performance improves, retain the new features; otherwise, exclude the new features.

[0115] (5) Repeat steps (3) and (4) until the preset stop condition is reached.

[0116] Step 3: Use the Adam optimizer preliminary optimization algorithm to optimize the learning model, use the training data set to train the learning model, and build a three-dimensional space algorithm into the learning model.

[0117] The parameters obtained by the measurement while drilling for the two-zone height prediction are huge in number and change in real time. GRU is an improved recurrent neural network specially designed for processing time series data. It solves the gradient vanishing and gradient exploding problems that are prone to occur in traditional recurrent neural networks in long sequence training by introducing a gating mechanism. At the same time, it can reduce model parameters, improve training efficiency, and solve various problems of while drilling parameters. Therefore, GRU recurrent neural network is selected as the learning model for two-zone height prediction.

[0118] The learning model training process is as follows:

[0119] (1) Data input and initialization:

[0120] ①Time series input: ,in is the input vector at time step t, d is the feature dimension, and T is the number of input vectors.

[0121] ② Drilling parameter input: The drilling parameters in the training data set are input into the model as initial data.

[0122] ③ Initial hidden state: Initialize the input time series and drilling parameters.

[0123] (2) Gated recurrent unit GRU calculation:

[0124] ① Combine the calculated and newly input drilling parameters by resetting the gate:

[0125] ;

[0126] in:

[0127] is the output parameter of the reset gate.

[0128] is the Sigmoid activation function.

[0129] is the weight matrix of the reset gate.

[0130] is the hidden state at the previous time step t-1.

[0131] is the input parameter for the current time step t.

[0132] is the bias term for the reset gate.

[0133] ② Filter the calculated while-drilling parameters through the update gate and save the filtered parameters to the current state:

[0134] ;

[0135] in:

[0136] is the output of the update gate.

[0137] is the weight matrix of the update gate.

[0138] is the bias term of the update gate.

[0139] ③ Combine the current input parameters and reset the parameters of the gate to make it a candidate hidden state:

[0140] ;

[0141] in:

[0142] is a candidate hidden state.

[0143] Represents element-wise multiplication.

[0144] is the weight matrix of the candidate hidden states.

[0145] is the bias term for the candidate hidden state.

[0146] ④Weighted combination of update gate and candidate hidden state parameters to make it the final hidden state:

[0147] ;

[0148] in:

[0149] is the hidden state at the current time step.

[0150] (3) Model output:

[0151] The output of the GRU model is mapped to the target variable (two bands of height) through a fully connected layer (Dense Layer):

[0152] ;

[0153] in:

[0154] is the model at time step The predicted value of .

[0155] is the weight matrix of the output layer.

[0156] is the bias term of the output layer.

[0157] (4) Calculate the loss function:

[0158] The goal of model training is to minimize the error between the predicted value and the true value. Here, the mean square error (MSE) is used as the loss function:

[0159] ;

[0160] in:

[0161] It is the real height value of the two belts.

[0162] is the predicted value of the model.

[0163] is the sample size.

[0164] (5) Optimization model:

[0165] The model parameters are updated by gradient descent to minimize the loss function. The Adam optimizer is used and the parameter update formula is:

[0166] ;

[0167] in:

[0168] are the model parameters (including weights and biases).

[0169] is the learning rate.

[0170] is the gradient of the loss function with respect to the parameters.

[0171] The ReLU activation function is selected for the initialization model, the learning rate is set to 0.001, and the number of training rounds is selected to be 50. The training data set is input into the learning model for training, and the effectiveness of the training is judged according to the preset results. The choice of attenuation factor ρ and learning rate has a great influence on the effect of the algorithm, and it is necessary to adjust and optimize through multiple rounds of training.

[0172] Built-in 3D space algorithm Figure 3 The three-dimensional spatial information of the two-zone damage positions includes: the vertical height of the position where the drilling parameters fluctuate violently from the monitoring point , the horizontal distance between the monitoring points of the two crack locations , the vertical distance between the two crack positions and the monitoring point along the tunnel direction And the strike distance between the two crack positions and the monitoring points along the tunnel direction In step 3, a three-dimensional space algorithm is built in, which can directly calculate the borehole length obtained by the drilling parameters into the height of the two zones. The specific calculation steps are as follows:

[0173] (1) Adjust the drilling inclination , drilling azimuth Pass the trained model as input parameters.

[0174] (2) Calculate the horizontal distance between the two crack locations and the monitoring point D :

[0175] ;

[0176] Where:

[0177] is the horizontal distance between the two crack locations and the monitoring point / m.

[0178] (3) Calculate the vertical distance d between the two crack positions and the monitoring point along the tunnel direction:

[0179] ;

[0180] Where:

[0181] is the vertical distance between the two crack positions and the monitoring point along the tunnel direction / m;

[0182] is the drilling azimuth / °;

[0183] (4) Calculate the strike distance a between the two crack positions and the monitoring point along the tunnel direction: ;

[0184] (5) Calculate the vertical height of the two crack locations from the monitoring point :

[0185] ;

[0186] Where:

[0187] is the vertical height of the two cracks from the monitoring point / m;

[0188] L is the length of the crack area from the monitoring point / m;

[0189] is the drilling inclination angle / °.

[0190] Step 4: Use the test set data to test the accuracy of the learning model.

[0191] This step mainly includes two types of tests: other field measured data comparison tests and test data set tests.

[0192] The main purpose of other field measured data comparison tests is to verify the accuracy of this method for two-zone height prediction. The main steps are as follows: use the conventional two-zone height prediction method to predict the location where this method is used for two-zone height prediction, compare the prediction results obtained by the two-zone height prediction method with those by the conventional prediction method, and finally prove the feasibility of the two-zone height prediction method.

[0193] The main purpose of the test data set test is to adjust the prediction accuracy. The main steps are as follows: preset the prediction accuracy, set the error between the predicted value and the true value to be less than or equal to 10%, input the test data set into the trained learning model, and judge the generated prediction results. If it meets the prediction accuracy requirements, the model training is completed and the prediction results are output; if not, return to step three, adjust the model parameters (attenuation factor, learning rate, training rounds, amount of training data, etc.), and train the model again until the prediction accuracy requirements are met.

[0194] When using the test data set to test the model, the square absolute error is used as the evaluation indicator to evaluate the results. The specific analytical formula is as follows:

[0195] ;

[0196] If the model does not meet the test accuracy, the model parameters are optimized by adjusting the hyperparameter method of the number of GRU layers and hidden units and the Dropout regularization method. The specific steps include:

[0197] ① Increase the number of GRU layers to enhance the model's expressiveness and capture more complex features. When optimizing the model, start with 1 layer and gradually increase the number of layers (such as 2-3 layers) to avoid overfitting due to too many layers.

[0198] For L-layer GRU, the The hidden state of is calculated as:

[0199] ;

[0200] in: (Input layer).

[0201] is the final hidden state.

[0202] ② Adjust the number of hidden units to balance the model’s expressiveness and computational complexity. Try different numbers of hidden units (such as 32, 64, 128, 256) and choose the appropriate value based on the data size and task complexity.

[0203] ③ Use L2 regularization technology to limit the size of model parameters and prevent overfitting. That is, add L2 regularization terms to the loss function:

[0204] ;

[0205] in Usually set to 0.0001-0.01, is the new loss function, is the original loss function, is the regularization coefficient, is the i-th model parameter.

[0206] Step 5: Use the learning model that meets the accuracy requirements to predict the height of the two zones of overburden at the working face.

[0207] like Figure 1 , Figure 2 As shown in the figure, drilling is carried out on the tunnel roof, and the drilling parameters are recorded. The drilling parameters generated by drilling are used as input data after data preprocessing and input into the trained learning model to predict the height of the two zones of the overlying strata on the working face. Figure 1 This is a schematic diagram of the height prediction of the two zones of overburden in the goaf under the condition of leaving coal pillars. Figure 2 This is a schematic diagram of the predicted height of the two zones of overburden in the goaf under the condition of retaining a road along the goaf.

[0208] The specific steps are as follows:

[0209] (1) Drilling into the overlying strata of the goaf in the tunnel and recording the drilling parameters;

[0210] (2) The drilling parameters generated by the operation are used as input data after data preprocessing and input into the trained learning model to obtain the prediction results. The heights of the two zones are determined based on the predicted geomechanical parameters of the surrounding rock.

[0211] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0212] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An intelligent prediction method for the height of two zones of overburden in a working face based on drilling parameters, characterized in that: The following steps are involved: Step 1: Select measuring stations at different locations for drilling operations according to different drilling conditions; Step 2: Acquire real-time monitoring data, perform data preprocessing on the real-time monitoring data and the rock mass geomechanical parameters in the two zones, and generate a training data set and a test data set; Step 3: Use the Adam optimizer to optimize the learning model, use the training data set to train the learning model, and build a three-dimensional space algorithm into the learning model; Built-in 3D space algorithm directly calculates the borehole length obtained by drilling parameters into two-zone height. The specific calculation steps are as follows: (1) Adjust the drilling inclination , drilling azimuth Feed the trained model as input parameters; (2) Calculate the horizontal distance between the two crack locations and the monitoring point D : ； Where: is the horizontal distance between the two crack locations and the monitoring point; (3) Calculate the vertical distance d between the two crack positions and the monitoring point along the tunnel direction: ； Where: is the vertical distance between the two crack positions and the monitoring point along the tunnel direction; (4) Calculate the strike distance a between the two crack positions and the monitoring point along the tunnel direction: ； (5) Calculate the vertical height of the two crack locations from the monitoring point : ； Where: is the vertical height of the two cracks from the monitoring point; L is the length of the crack area from the monitoring point; Step 4: Use the test data set to test the accuracy of the learning model, and use the hyperparameter optimization algorithm and L2 regularization technology to optimize the model; Step 5: Use the learning model that meets the accuracy requirements to predict the height of the two zones of the overburden at the working face; The feed speed and rotation speed in the drilling parameters are used as initial input parameters of the drilling rig, and the displacement, pressure, torque and power are used as output parameters of the drilling rig to predict the height of the two belts.

2. The method for intelligently predicting the height of two zones of overburden of a working face based on drilling parameters according to claim 1 is characterized in that: In step 1, drilling operations are carried out on the overlying strata of the goaf in adjacent tunnels, and the drilling parameters are recorded. The steps are as follows: S11. Preliminary determination of drilling depth, directivity arrangement and station spacing; S12. Drilling is performed on the overlying rock strata of the goaf in the roof of the adjacent tunnel of the goaf, and the borehole penetrates the water-conducting fracture zone of the overlying rock strata of the goaf and enters the upper curved sinking zone for 5 to 10 m; S13, recording all drilling parameters of the drilling rig and exporting the drilling parameters obtained during the drilling.

3. The method for intelligently predicting the height of two zones of overburden of a working face based on drilling parameters according to claim 1 is characterized in that: In step 2, the real-time monitoring data is cleaned to obtain the filtered data set, and the feature selection is implemented using a stepwise feature selection algorithm. Starting from the initial feature, new features are continuously added until the stop condition is reached. The specific steps are as follows: (1) Initialize the algorithm and set the initial feature subset to an empty set; (2) Setting the stopping condition: adding new features several times in a row but the accuracy does not improve any more; (3) Use the cross-validation algorithm to evaluate each unselected feature separately and select the best performing feature to add to the current feature subset; (4) After adding new features to the current feature subset, re-evaluate the performance of the feature combination. If the performance improves, retain the new features; otherwise, exclude the new features. (5) Repeat steps (3) and (4) until the preset stop condition is reached.

4. The method for intelligently predicting the height of two zones of overburden of a working face based on drilling parameters according to claim 1 is characterized in that: The learning model training process in step 3 is as follows: (1) Input time series and drilling parameters, and initialize the input time series and drilling parameters; (2) Gated recurrent unit calculation: ① Combine the calculated and newly inputted drilling parameters by resetting the gate: ； in: is the output parameter of the reset gate; is the Sigmoid activation function; is the weight matrix of the reset gate; is the hidden state of the previous time step t-1; is the input parameter of the current time step t; is the bias term for the reset gate; ② Filter the calculated while-drilling parameters through the update gate and save the filtered parameters to the current state: ； in: is the output of the update gate; is the weight matrix of the update gate; is the bias term of the update gate; ③ Combine the current input parameters and the parameters of the reset gate to obtain the candidate hidden state: ； in: is a candidate hidden state; Represents element-wise multiplication; is the weight matrix of the candidate hidden states; is the bias term of the candidate hidden state; ④Weighted combination of update gate and candidate hidden state parameters to obtain the final hidden state : ； (3) Mapped to the target variable through the fully connected layer: ； in: is the model at time step The predicted value of is the weight matrix of the output layer; is the bias term of the output layer; (4) Use mean square error (MSE) as the loss function to minimize the error between the predicted value and the true value: ； in: are the true two-band height values; is the sample size; (5) Using the Adam optimizer, update the model parameters by gradient descent method: ； in: are model parameters; is the learning rate; is the gradient of the loss function with respect to the parameters.

5. The method for intelligently predicting the height of two zones of overburden of a working face based on drilling parameters according to claim 1 is characterized in that: Use data cleaning with built-in numerical calculation optimization algorithm to preprocess data.

Images