Intelligent prediction method for stability of TBM tunneling roof of inclined shaft of coal mine

Abstract

The invention provides a coal mine inclined shaft TBM tunneling roof stability intelligent prediction method comprising the following steps: step 1, collecting TBM tunneling measured data, and generating a TBM operation data set, step 2, dividing the TBM operation data set in the step 1 into a plurality of tunneling periods according to tunneling cycles, and establishing a tunneling period data set, and step 3, based on the tunneling period data set, determining the stability of the TBM tunneling roof. Step 2, calculating and establishing an input parameter data set in the neural network; step 4, utilizing DBNamp; and step 5, comparing with a real working condition, judging whether caving can be really predicted in the early warning range or not, and establishing an inclined shaft TBM tunneling roof caving early warning range criterion. The roof caving risk is analyzed and predicted based on a large number of tunneling parameters collected by the system in the TBM tunneling process. The prediction result can provide high-reliability on-site risk prediction for roof caving excavation by using the TBM inclined shaft.

Description

Technical Field

[0001] The invention relates to the field of risk prediction, and in particular to an intelligent prediction method for the stability of a roof during TBM excavation in a coal mine inclined shaft. Background Art

[0002] Compared with traditional tunnel construction methods, (full-face) tunnel boring machines (TBMs) have the advantages of fast excavation speed and high construction safety. Since TBMs are expensive, correctly evaluating TBM performance can greatly avoid the occurrence of surrounding rock roof collapse during construction, speed up the construction process, increase TBM utilization, and reduce the possibility of machine damage.

[0003] The shortcomings of existing inclined shaft TBM excavation are mainly reflected in the following aspects: the empirical method is used to deal with risks, and repairs are often started only when the roof collapses, which seriously affects the construction speed. Although some achievements have been made in TBM intelligent excavation, most research focuses on horizontal tunnel excavation, and there are relatively few cases for mining inclined shaft construction. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent prediction method for the stability of a roof in a coal mine inclined shaft TBM excavation, so as to analyze and predict the risk of roof collapse.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for intelligently predicting the stability of a coal mine inclined shaft TBM excavation roof comprises the following steps:

[0007] Step 1: Collect TBM excavation measured data and generate TBM operation data set.

[0008] Step 2: Divide the TBM operation data set in step 1 into multiple excavation cycles according to the excavation cycle, and establish an excavation cycle data set.

[0009] Step 3: Based on the excavation cycle data set, calculate and establish the input parameter data set in the neural network.

[0010] Step 4: Use DBN&TSF algorithm to obtain the prediction results of the pilot section and establish the warning range of the caving area.

[0011] Step 5: Compare with the actual working conditions to determine whether the warning range can truly predict the occurrence of roof collapse, and establish the warning range criteria for inclined shaft TBM excavation roof collapse.

[0012] Furthermore, in the above-mentioned coal mine inclined shaft TBM excavation roof stability intelligent prediction method, in step 1, based on the measured data collected by TBM excavation, the measured data are selected and processed once, and the measured data of the trial excavation and debugging state are discarded to generate an optimized TBM operation data set. The optimized TBM operation data set includes: advancement speed v, defined as the distance the TBM excavates per minute; penetration P rev , defined as the thickness of the rock cut into by each TBM cutterhead per rotation; rotation speed w, defined as the number of revolutions per minute of the TBM cutterhead; thrust F, defined as the average thrust of each TBM cutterhead; torque T, defined as the average torque of each TBM cutterhead.

[0013] Furthermore, in the above-mentioned intelligent prediction method for the stability of the TBM excavation roof in the coal mine inclined shaft, in step 2, the TBM operation data set is divided into multiple excavation cycles according to the excavation cycle, and the excavation cycles are numbered from 1 to establish an excavation cycle data set. The screening rule is: within the excavation range of 0 to 1.8 m, an excavation distance greater than 0.4 m is defined as a normal excavation cycle, and an excavation distance less than 0.4 m is defined as an abnormal excavation cycle. An excavation cycle can be divided into a starting loading section, a stable advancement section, and an end unloading section, and the excavation parameters of the stable advancement section are used as the analysis object.

[0014] Furthermore, in the above-mentioned intelligent prediction method for roof stability of TBM excavation in coal mine inclined shaft, in step 3, based on the excavation cycle data set, the penetration rate P is calculated using the measured advancement speed v and cutter head speed w in the caving area and the normal excavation section. rev , the torque T and penetration P rev The parameters TPI, thrust F and penetration P are formed rev The parameter FPI formed is used as the input parameter in the neural network, and the parameter TPI calculation formula is:

[0015]

[0016]

[0017] The unit of TPI parameter is 10 3 kN rpm,

[0018] The calculation formula of parameter FPI is:

[0019]

[0020] The unit of FPI parameter is kN·rpm·mm -1 .

[0021] Further, in the above-mentioned coal mine inclined shaft TBM excavation roof stability intelligent prediction method, in step 4, based on the input parameter data set, the input parameter data set of the pilot section is selected and substituted into the DBN&TSF algorithm for prediction, which specifically includes:

[0022] Step 41: define the number of all excavation cycles in the established excavation cycle data set as m, select the first k cycles as the training group, and the following h cycles as the test group, m=k+h, the data of the training group and the test group take the average value of all TPI and FPI in the stable section of each excavation cycle, define the engineering measured value of the training group as K-train, and the predicted value calculated by the DBN&TSF algorithm as K-test; define the engineering measured value of the test group as H-train, and the predicted value calculated by the DBN&TSF algorithm as H-test. After the DBN&TSF algorithm is trained on the k group, the performance test is performed on the h group.

[0023] Step 42: Starting from the first excavation cycle, substitute the true values ​​of the first five excavation cycles in the m training sets into the DBN&TSF algorithm to find the optimal target value f DBN , and specify the target value as the predicted value of the sixth excavation cycle, and K-test 6 As a new vector input into the algorithm, K-train is discarded 1 , and K-test 7 As the goal, restart a round of calculation;

[0024] Step 43: Neural Network f DBN In this round of training, the upgrade is performed, and then steps 41 to 42 are looped, and the cycle continues until the predicted values ​​of all excavation cycles are obtained.

[0025] Furthermore, in the above-mentioned intelligent prediction method for roof stability of TBM excavation in coal mine inclined shaft, the evaluation index of the caving warning range criterion is specifically:

[0026] Variance contribution rate VAF, Where: var represents variance,

[0027] VAF is an indicator used to evaluate the correlation between the measured value and the predicted value learned in the model. The higher the VAF, the better the model's ability to explain the measured value, and the closer the model's prediction result is to the actual result.

[0028] Mean relative error MRE,

[0029] The mean relative error (MRE) is an indicator that reflects the reliability of measurement. The smaller the MRE, the smaller the relative error between the measured value and the predicted value.

[0030] Coefficient of determination R 1 ,

[0031] Correlation coefficient R 1 It is an indicator to measure the correlation between the true value and the predicted value. 1 The smaller it is, the higher the correlation is and the more accurate the model is.

[0032] Among them, y is the actual measured value of the project, and y′ is the predicted value obtained by the DBN&TSF algorithm.

[0033] The specific criteria for the warning range of roof fall in inclined shaft TBM excavation are to compare the evaluation indicators of the correct warning results of roof fall based on whether there is roof fall on site, and give a range of evaluation indicators, VAF is less than a certain value, MRE is greater than a certain value, R 1 If the value is greater than a certain value, and the evaluation index obtained by the subsequent data set input into the algorithm for prediction is within this range, there is a potential risk and emergency measures need to be taken. The warning range of the roof collapse index is reversely determined according to the collapse degree of the pilot section: VAF<0.80, MRE>0.10 and R 1 <0.85, VAF, MRE, and R 1 When all the above values ​​are met, an early warning will be issued in the falling area.

[0034] Furthermore, in the above-mentioned coal mine inclined shaft TBM excavation roof stability intelligent prediction method, step 5 specifically includes:

[0035] Step 51: Based on the caving area warning range established by the prediction result of the pilot section, the input parameter data sets of multiple caving areas of the non-pilot section are substituted into the DBN&TSF algorithm for prediction to check whether the prediction result is within the established caving area warning range, so as to determine whether the DBN&TSF algorithm can truly predict the caving of the section;

[0036] Step 52: Based on the judgment result of step 51, the warning range of the caving area is expanded or reduced according to the caving degree of the caving area, until all caving areas are included in the determined warning range of the caving area, and the final warning range of the evaluation index of roof caving is determined;

[0037] Step 53: Based on the predetermined warning range of the evaluation index and the measured data of TBM excavation, the possibility of roof caving in the unknown section ahead is predicted, and the warning range criteria for roof caving in inclined shaft TBM excavation are established.

[0038] It can be seen from the analysis that the present invention discloses an intelligent prediction method for roof stability in coal mine inclined shaft TBM excavation, which analyzes and predicts the risk of roof falling based on a large number of excavation parameters (such as thrust, torque, etc.) collected by the system during TBM excavation. The prediction results can provide a highly reliable on-site risk forecast of roof falling when using TBM inclined shaft excavation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. Among them:

[0040] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention.

[0041] Figure 2 This is a normal excavation cycle diagram in a TBM database according to an embodiment of the present invention.

[0042] Figure 3 This is an abnormal excavation cycle diagram in a TBM database according to an embodiment of the present invention.

[0043] Figure 4 It is a schematic diagram of the process of step 4 of the present invention.

[0044] Figure 5 It is a schematic diagram of the process of step 5 of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present invention and does not limit the present invention. In fact, it will be clear to those skilled in the art that modifications and variations may be made in the present invention without departing from the scope or spirit of the present invention. For example, a feature shown or described as a part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, it is desired that the present invention encompasses such modifications and variations within the scope of the appended claims and their equivalents.

[0046] like Figures 1 to 5 As shown, according to an embodiment of the present invention, a method for intelligent prediction of the stability of the roof of a coal mine inclined shaft TBM excavation is provided. A TBM big data platform system based on the TSF (time series prediction) & DBN (deep belief network) algorithm is used to predict the risk of roof collapse in the inclined shaft and to establish the warning range criteria for roof collapse in the inclined shaft TBM excavation. The intelligent prediction method includes the following steps (such as Figure 1 shown):

[0047] Step 1: Collect TBM excavation measured data and generate TBM operation data set.

[0048] Based on the measured data collected during TBM excavation, the measured data are selected and processed, and the measured data of trial excavation and debugging status are discarded. After removing some data that do not meet the requirements, an optimized TBM operation data set is generated.

[0049] The optimized TBM operation data set includes: advancement speed v, defined as the distance the TBM advances per minute; penetration rate P rev , defined as the thickness of the rock cut into by each TBM cutterhead per rotation; rotation speed w, defined as the number of revolutions per minute of the TBM cutterhead; thrust F, defined as the average thrust of each TBM cutterhead; torque T, defined as the average torque of each TBM cutterhead.

[0050] Step 2: divide the TBM operation data set in step 1 into multiple excavation cycles according to the excavation cycle, and establish an excavation cycle data set.

[0051] The TBM operation data set is divided into multiple excavation cycles according to the excavation cycle. The average advancement distance of one excavation cycle is about 1.8m. The excavation cycles are numbered from 1, and the screening rule is: within the excavation range of 0 to 1.8m, an excavation distance greater than 0.4m is defined as a normal excavation cycle (see Figure 2 ), a driving distance of less than 0.4m is defined as an abnormal driving cycle. Too short a driving distance is generally manifested in a significant increase in thrust and torque and a sharp decrease in propulsion speed (see Figure 3 ) or the rock on the TBM face is too hard to allow the cutterhead to rotate normally.

[0052] Figure 2 is a normal excavation cycle diagram, such as Figure 2 As shown in the figure, a tunneling cycle can be divided into the starting loading section, the stable advancement section and the end unloading section. In the starting loading section, the thrust, torque and advancement speed begin to gradually increase. After entering the stable advancement section, the values ​​of each parameter tend to be stable. After 20 minutes of stable advancement, the cutterhead stops, the values ​​of each tunneling parameter decrease, and the tunneling cycle ends.

[0053] Figure 3 is an abnormal excavation cycle diagram, such as Figure 3 As shown, after 10 minutes of excavation, in the framed abnormal area, the thrust and torque increased significantly and the propulsion speed dropped sharply, indicating that the cutter head could not rotate normally.

[0054] The parameters of the stable advancement section of the normal excavation cycle are stable and cover a large part of the data in this cycle. Therefore, the excavation parameters of the stable advancement section are used as the analysis object. According to the real-time data provided by the TBM every minute, the data set formed by the stable advancement section of each normal excavation cycle is used for model training and learning.

[0055] For example, within the effective TBM excavation days, the database records the corresponding excavation measured parameters of the inclined shaft from 33.4m (cycle number 1) to 3768.1m (cycle number 4178). Among them, 1604 excavation cycles with an excavation length of less than 0.4m are abnormal excavation cycles and are not used for subsequent training, learning and analysis.

[0056] A TBM excavation cycle includes the following:

[0057] (1) Start loading phase: The excavation cycle begins, the support shoe cylinder supports the shoe and contacts the rock, and the rear support is lifted;

[0058] (2) Stable propulsion stage: The cutter disc rotates, the rear support is lowered and supported on the ground, the propulsion cylinder is started to propel the cutter disc forward, the cutter disc is advanced, and the main support provides thrust until it advances one stroke;

[0059] (3) End of unloading section: the cutterhead stops, the rear support opens, the support shoe cylinder retracts, the step is changed and the direction and posture of the TBM are adjusted to start a new excavation cycle.

[0060] Step 3: Based on the excavation cycle data set, calculate and establish the input parameter data set in the neural network.

[0061] Based on the excavation cycle data set, the penetration rate P is calculated using the measured advance speed v and cutter head speed w in the caving area and the normal excavation section. rev , the torque T and penetration P rev The parameters TPI, thrust F and penetration P are formed rev The formed parameters FPI are used as input parameters in the neural network.

[0062] The parameter TPI calculation formula is:

[0063]

[0064]

[0065] The unit of TPI parameter is 10 3 ·kN·rpm.

[0066] The calculation formula of parameter FPI is:

[0067]

[0068] The unit of FPI parameter is kN·rpm·mm -1 .

[0069] Step 4: Use the DBN&TSF algorithm to obtain the prediction results of the pilot section and establish the warning range of the caving area.

[0070] Based on the input parameter data set, the input parameter data set of the pilot section is selected and substituted into the DBN&TSF algorithm for prediction.

[0071] The full name of DBN&TSF algorithm is the time series prediction (TSF) algorithm based on deep belief network (DBN). The deep belief network (DBN) is a network model obtained by adding a back propagation (BP) layer to the restricted Boltzmann machine (RBM) model. It performs unsupervised pre-training of parameters and learns layer by layer. Time series prediction (TSF) arranges data at certain time intervals, analyzes these time series, discovers and reveals the laws of the development and change of phenomena, and uses the laws for prediction. The combination of deep belief network (DBN) and time series prediction (TSF) forms the DBN&TSF algorithm.

[0072] The specific steps are as follows (see Figure 4 ):

[0073] Step 41: Define the number of all excavation cycles in the excavation cycle data set established above as m, select the first k cycles as the training group, and the following h cycles as the test group, m=k+h. The data of the training group and the test group take the average value of all TPI and FPI in the stable section of each excavation cycle. The engineering measured value (true value) of the training group is defined as K-train, and the predicted value calculated by the DBN&TSF algorithm is defined as K-test; the engineering measured value (true value) of the test group is defined as H-train, and the predicted value calculated by the DBN&TSF algorithm is defined as H-test. After the DBN&TSF algorithm is trained on k groups, the performance is tested on h groups.

[0074] Step 42: Starting from the first excavation cycle, substitute the true values ​​of the first five excavation cycles in the m training sets into the DBN&TSF algorithm to find the optimal target value f DBN , and the target value is designated as the predicted value of the sixth excavation cycle. 6 As a new vector input into the algorithm, K-train is discarded 1 , and K-test 7 As the goal, restart a round of calculation;

[0075] Step 43: Neural Network f DBN In this round of training, the upgrade is performed, and then steps 41 to 42 are looped, and the cycle continues until the predicted values ​​of all excavation cycles are obtained.

[0076] Figure 4This is the flowchart of the DBN&TSF algorithm. Taking TPI as an example, the average TPI value of the first five tunneling cycles in the m training sets is taken as the true value. These 6 true values ​​are calculated by the RBM layer and BP layer of the algorithm to obtain the optimal target value TPI. 7 As the predicted value of the algorithm, TPI 6 As a new vector input algorithm, while discarding TPI 1 , and TPI 7 As the goal, a new round of calculation is restarted until the TPI prediction values ​​for all excavation cycles are obtained.

[0077] The specific evaluation indicators of the caving warning range criteria are:

[0078] Variance contribution rate VAF, Where: var represents variance.

[0079] Mean relative error MRE, Coefficient of determination R 1 , Where y is the measured value of the project, and y′ is the predicted value obtained by the DBN&TSF algorithm.

[0080] VAF is an indicator used to evaluate the correlation between the measured value and the predicted value learned in the model. The higher the VAF, the better the model's ability to explain the measured value, and the closer the model's prediction result is to the actual result. The mean relative error (MRE) is an indicator that reflects the reliability of the measurement. The smaller the MRE, the smaller the relative error between the measured value and the predicted value. Correlation coefficient R 1 It is an indicator to measure the correlation between the true value and the predicted value. 1 The smaller it is, the higher the correlation is and the more accurate the model is.

[0081] The specific criterion for the early warning range of the stratum cave-in area is to compare the evaluation indicators of the correct early warning cave-in results based on whether the site caves in, and give a range of evaluation indicators, VAF is less than a certain value, MRE is greater than a certain value, R 1 If it is greater than a certain value, and the evaluation index obtained by the subsequent data set collected and put into the algorithm for prediction is within this range, there is a potential risk and emergency measures need to be taken.

[0082] In the example given, the warning range of roof collapse evaluation indexes was determined to be VAF<0.80, MRE>0.10 and R 1 <0.85.

[0083] VAF, MRE, and R 1 When all the above values ​​are met, an early warning will be issued in the falling area.

[0084] Step 5: Compare with the actual working conditions to determine whether the warning range can truly predict the occurrence of roof collapse, and establish the warning range criteria for inclined shaft TBM excavation roof collapse.

[0085] The specific steps for establishing the warning range of roof fall in inclined shaft TBM excavation are as follows (see Figure 5 ):

[0086] Step 51: Based on the caving area warning range established by the prediction result of the pilot section, the input parameter data sets of multiple caving areas of the non-pilot section are substituted into the DBN&TSF algorithm for prediction to check whether the prediction result is within the established caving area warning range, so as to determine whether the DBN&TSF algorithm can truly predict the caving of the section;

[0087] Step 52: Based on the judgment result of step 51, the warning range of the caving area is expanded or reduced according to the caving degree of the caving area, until all caving areas are covered in the determined warning range, and the final warning range of the evaluation index of roof caving is determined;

[0088] Step 53: Based on the predetermined warning range of the evaluation index and the measured data of TBM excavation, the possibility of roof caving in the unknown section ahead is predicted, and the warning range criteria for roof caving in inclined shaft TBM excavation are established.

[0089] In one embodiment of the present invention, based on the learning samples of the pilot section, 12 roof falls in the main inclined shaft were predicted, and 10 were successfully predicted with an accuracy rate of up to 83.3%, indicating that the DBN&TSF algorithm model has a high credibility for predicting roof falls in TBM inclined shafts.

[0090] From the above description, it can be seen that the above embodiments of the present invention achieve the following technical effects:

[0091] In the present invention, by applying the TSF&DBN algorithm to the prediction of roof fall that may occur in inclined shaft excavation, big data is well applied to accurately predict risks, thereby improving the accuracy of judging whether the excavation is normal in the project. The optimized parameters TPI formed by the propulsion speed, penetration, and cutterhead speed obtained after cleaning and sorting the excavation parameters in the measured fall zone and normal section, and the parameters FPI formed by the thrust and penetration are used as input parameters in the neural network, and are transmitted to the TSF&DBN algorithm model for data analysis. The possibility of the fall risk is analyzed in combination with the rock formation conditions ahead, and the results are transmitted to the cab in real time through the platform. The multi-platform combined application of the TBM operation parameter monitoring platform, the algorithm model analysis platform, and the real-time feedback platform combined with the rock formation conditions is realized. The fall risk is upgraded from empirical judgment to intelligent judgment combining big data with actual conditions, which provides an important basis for TBM intelligent construction and avoidance of fall risks. It provides a better standard for improving construction efficiency, improving construction level, and realizing TBM intelligent inclined shaft excavation.

[0092] (1) Taking the TBM excavation of a main inclined shaft as the research background, the real-time excavation parameters of the construction section were used as the basic database. Based on the DBN&TSF algorithm, the relationship between the excavation parameters and the caving degree of the inclined shaft roof was established, and the inverse prediction analysis of the caving area was realized.

[0093] (2) The linear regression results of the TPI and FPI show that there is a certain functional relationship between TBM thrust, torque and penetration, and both can be used as effective parameters in the DBN&TSF algorithm to successfully predict caving.

[0094] (3) The number of samples and the effective study area of ​​the samples determine the prediction accuracy of the DBN&TSF model. The more samples there are and the larger the effective study area is, the higher the prediction accuracy of roof collapse is. The warning range of roof collapse is determined by the collapse degree of the collapse area, which is VAF<0.80, MRE>0.10 and R 1 <0.85.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent prediction method for roof stability of TBM excavation in a coal mine inclined shaft, characterized in that: The following steps are involved: Step 1: Collect TBM excavation measured data and generate TBM operation data set. Step 2: Divide the TBM operation data set in step 1 into multiple excavation cycles according to the excavation cycle, and establish an excavation cycle data set. Step 3: Based on the excavation cycle data set, calculate and establish the input parameter data set in the neural network. Step 4: Use DBN&TSF algorithm to obtain the prediction results of the pilot section and establish the warning range of the caving area. Step 5: Compare with the actual working conditions to determine whether the warning range can truly predict the occurrence of roof collapse, and establish the warning range criteria for inclined shaft TBM excavation roof collapse.

2. The intelligent prediction method for roof stability of TBM excavation in a coal mine inclined shaft according to claim 1 is characterized in that: In step 1, based on the measured data collected during TBM excavation, the measured data are selected and processed, and the measured data of trial excavation and debugging are discarded to generate an optimized TBM operation data set. The optimized TBM operation data set includes: advancement speed v, defined as the distance the TBM advances per minute; penetration rate P rev , defined as the thickness of the rock cut into by each TBM cutterhead per rotation; rotation speed w, defined as the number of revolutions per minute of the TBM cutterhead; thrust F, defined as the average thrust of each TBM cutterhead; torque T, defined as the average torque of each TBM cutterhead.

3. The intelligent prediction method for roof stability of TBM excavation in coal mine inclined shaft according to claim 1 is characterized in that: In step 2, the TBM operation data set is divided into multiple excavation cycles according to the excavation cycle. The excavation cycles are numbered starting from 1 to establish the excavation cycle data set. The screening rule is: within the excavation range of 0 to 1.8 m, an excavation distance greater than 0.4 m is defined as a normal excavation cycle, and an excavation distance less than 0.4 m is defined as an abnormal excavation cycle. A tunneling cycle can be divided into the starting loading section, the stable advancement section and the ending unloading section. The tunneling parameters of the stable advancement section are used as the analysis object.

4. The intelligent prediction method for roof stability of TBM excavation in a coal mine inclined shaft according to claim 1 is characterized in that: In step 3, based on the excavation cycle data set, the penetration rate P is calculated using the measured advancement speed v and cutter head speed w in the caving area and the normal excavation section. rev , the torque T and penetration P rev The parameters TPI, thrust F and penetration P are formed rev The formed parameter FPI is used as the input parameter in the neural network. The parameter TPI calculation formula is: The unit of TPI parameter is 10 3 kN rpm, The calculation formula of parameter FPI is: The unit of FPI parameter is kN·rpm·mm -1 .

5. The intelligent prediction method for roof stability of TBM excavation in coal mine inclined shaft according to claim 4 is characterized in that: In step 4, based on the input parameter data set, the input parameter data set of the pilot section is selected and substituted into the DBN&TSF algorithm for prediction, which specifically includes: Step 41: define the number of all excavation cycles in the established excavation cycle data set as m, select the first k cycles as the training group, and the last h cycles as the test group, m=k+h, and take the average value of all TPI and FPI of the stable section of each excavation cycle for the data of the training group and the test group. The engineering measured value of the training group is defined as K-train, and the predicted value calculated by the DBN&TSF algorithm is defined as K-test; The measured value of the test group is defined as H-train, and the predicted value calculated by DBN&TSF algorithm is defined as H-test. After the DBN&TSF algorithm is trained on k groups, its performance is tested on h groups. Step 42: Starting from the first excavation cycle, substitute the true values ​​of the first five excavation cycles in the m training sets into the DBN&TSF algorithm to find the optimal target value f DBN , and the target value is designated as the predicted value of the sixth excavation cycle, Input K-test6 into the algorithm as a new vector, discard K-train1, and use K-test7 as the target to restart the calculation. Step 43: Neural Network f DBN During this round of training, the process is upgraded and then steps 41 to 42 are looped, and the cycle continues until the predicted values ​​of all excavation cycles are obtained.

6. The intelligent prediction method for roof stability of TBM excavation in a coal mine inclined shaft according to claim 5 is characterized in that: The specific evaluation indicators of the caving warning range criteria are: Variance contribution rate VAF, Where: var represents variance, VAF is an indicator used to evaluate the correlation between the measured value and the predicted value learned in the model. The higher the VAF, the better the model's ability to explain the measured value, and the closer the model's prediction result is to the actual result. Mean relative error MRE, The mean relative error (MRE) is an indicator that reflects the reliability of measurement. The smaller the MRE, the smaller the relative error between the measured value and the predicted value. Coefficient of determination R1, The correlation coefficient R1 is an indicator to measure the correlation between the true value and the predicted value. The smaller the R1, the higher the correlation and the more accurate the model. Among them, y is the actual measured value of the project, and y′ is the predicted value obtained by the DBN&TSF algorithm. The specific criterion for the warning range of roof fall in inclined shaft TBM excavation is to compare the evaluation indicators of the correct warning results of roof fall based on whether there is roof fall on site. Given a range of evaluation indicators, VAF is less than a certain value, MRE is greater than a certain value, and R1 is greater than a certain value. If the evaluation indicators obtained by the subsequent data set input into the algorithm for prediction are within this range, there is a potential risk and emergency measures need to be taken. According to the degree of roof collapse in the pilot section, the warning range of roof collapse indicators is reversely determined to be VAF<0.80, MRE>0.10 and R1<0.

85. When VAF, MRE and R1 all meet the above values, an early warning of the falling area will be issued.

7. The intelligent prediction method for roof stability of TBM excavation in a coal mine inclined shaft according to claim 6 is characterized in that: Step 5 specifically includes: Step 51: Based on the caving area warning range established by the prediction result of the pilot section, the input parameter data sets of multiple caving areas of the non-pilot section are substituted into the DBN&TSF algorithm for prediction to check whether the prediction result is within the established caving area warning range, so as to determine whether the DBN&TSF algorithm can truly predict the caving of the section; Step 52: Based on the judgment result of step 51, the warning range of the caving area is expanded or reduced according to the caving degree of the caving area, until all caving areas are included in the determined warning range of the caving area, and the final warning range of the evaluation index of roof caving is determined; Step 53: Based on the predetermined warning range of the evaluation index and the measured data of TBM excavation, the possibility of roof caving in the unknown section ahead is predicted, and the warning range criteria for roof caving in inclined shaft TBM excavation are established.

Images