Monitoring and Prediction Model for Anisotropic Time-Dependent Deformation of Roadway Surrounding Rock under Mining Disturbance

Abstract

The present invention discloses a monitoring and prediction model for anisotropic time-dependent deformation of roadway surrounding rock under mining disturbance, comprising the following steps: Step 1, in-situ specimen preparation; Step 2, construction of specimen installation cavity; Step 3, specimen installation and monitoring system layout; Step 4, construction of in-situ test environment and in-situ stress recovery; Step 5, data acquisition; Step 6, construction of time-dependent deformation model; Step 7, construction of time-dependent deformation prediction model. By installing anisotropic specimens in the deep underground roadway site and combining with an automated monitoring system, the present invention can collect the stress-strain characteristics of the roadway surrounding rock under the change of mining stress in real time and establish a time-dependent deformation prediction model, which can provide a scientific basis for the safe mining of underground mines.

Description

Technical Field

[0001] The present invention relates to the technical fields of deep underground rock mass mechanics, deep geological engineering, and intelligent monitoring and prediction technology. In particular, it is directed to the time-dependent deformation monitoring and prediction of deep anisotropic roadway surrounding rocks under mining disturbance conditions, and a dynamic early warning model is constructed using multi-sensor data fusion and deep reinforcement learning. Background Technique

[0002] The exploitation of deep coal resources faces complex geological conditions and strong mining disturbance effects. In particular, the surrounding rocks of deep roadways have significant anisotropic characteristics, and their mechanical behaviors show obvious differences in different directions. During the mining process, the dynamic change of mining stress will cause time-dependent deformation of the roadway surrounding rocks, that is, deformation that changes with time. This kind of deformation may trigger disasters such as roadway instability and rock bursts, seriously threatening the safe production of coal mines.

[0003] At present, most of the research on the time-dependent deformation of roadway surrounding rocks relies on indoor tests or numerical simulation methods. However, indoor tests are often difficult to truly reproduce the dynamic stress path under mining disturbance, and the specimen size effect may lead to distortion of anisotropic characteristics. In terms of numerical simulation, due to the use of idealized constitutive models, it is also difficult to comprehensively reflect the complex behaviors of multi-field coupling such as damage, seepage, and stress of roadway surrounding rocks under real mining conditions. In addition, traditional monitoring technologies such as single-point displacement gauges and bolt stress gauges can only obtain local and single-direction data, and it is difficult to achieve spatial continuity characterization of the anisotropic time-dependent deformation of roadway surrounding rocks.

[0004] Therefore, there is an urgent need to develop a field monitoring and prediction method that can synchronously obtain the time-dependent deformation evolution data of anisotropic roadway surrounding rocks under real mining disturbance conditions, and comprehensively consider the microscopic fracture characteristics and macroscopic mechanical responses, so as to provide a scientific basis for deeply revealing the damage mechanism of roadway surrounding rocks, realizing real-time dynamic monitoring, and accurately predicting future deformation trends, and providing strong technical support for the safe mining of coal mines. Summary of the Invention

[0005] The present invention aims to provide a monitoring and prediction model for the anisotropic time-dependent deformation of roadway surrounding rocks under mining disturbance. By installing anisotropic specimens in the coal mine underground site and combining with an automated monitoring system, the stress-strain characteristics of the roadway surrounding rocks under the change of mining stress are collected in real time, and a time-dependent deformation prediction model is established to provide a scientific basis for the safe mining of coal mines.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A monitoring and prediction model for the anisotropic time-dependent deformation of roadway surrounding rocks under mining disturbance, and the model construction includes the following steps:

[0008] Step 1. In-situ specimen preparation: Select the surrounding rock of the roadway in the advanced abutment pressure area of the coal mining face that has not been disturbed by mining. Obtain coal samples by coring, and cut them along three directions: the horizontal bedding of the coal sample obtained by coring at 0°, the vertical bedding at 90°, and the inclined bedding at 45° to obtain three types of specimens. The three types of specimens have the same size, and the specimens are cuboids.

[0009] Step 2. Construction of the specimen installation cavity: Construct an installation cavity in the surrounding rock area of the goaf roadway in front of the coal mining face that has not been disturbed by mining. The size of the installation cavity is larger than that of the specimen.

[0010] Step 3. Specimen installation and monitoring system layout: Place the specimen in the installation cavity, and arrange a hydraulic loading device at the open end of the installation cavity. The specimen is supported by the hydraulic loading device between it and the five side walls of the installation cavity. The hydraulic loading devices are all supported on the end face of the specimen, finally forming a triaxial support.

[0011] The loading end of the hydraulic loading device is provided with a stress sensing device. The stress sensing device is internally inlaid with a fiber Bragg grating stress sensor. A strain gauge is arranged on the telescopic arm of the hydraulic loading device. A temperature sensor and a humidity sensor are arranged in the installation cavity. The stress sensor, the strain gauge, the temperature sensor, and the humidity sensor are all connected to a data acquisition device, and the data acquisition device is connected to the ground control center.

[0012] The hydraulic loading device is controlled by the ground control center.

[0013] Step 4. In-situ test environment construction and in-situ stress recovery: According to the on-site exploration data, determine the target value of the in-situ ground stress of the specimen, which is represented as three directions of x, y, and z in space. Apply a pre-tightening force to the specimen through the hydraulic loading device to simulate the in-situ stress state.

[0014] Step 5. Data acquisition: Under mining disturbance, collect the data of the stress sensor, the strain gauge, the temperature sensor, and the humidity sensor in real time to correct the thermal stress effect and other environmental factors during the accelerated test process.

[0015] Step 6. Construct an aging deformation model: Based on the data obtained in Step 5, construct an aging deformation model.

[0016] Step 7. Construct an aging deformation prediction model: Based on the data obtained in Step 5, construct a deep reinforcement learning prediction model to predict the strain increment in the future time period.

[0017] Preferably, after Step 7, it further includes:

[0018] Step 8. Apply the aging deformation prediction model to the support design of mine roadways, the optimization of the advancing speed of coal mining faces, and the prevention and control of rock bursts.

[0019] Preferably, in step 1:

[0020] Select a depth of 100 m inside the advanced abutment pressure area of the coal mining face as the sampling position, and drill a coal core with a drill pipe having a diameter of 200 mm;

[0021] The dimensions of the specimen in terms of width × length × height are correspondingly 50 mm × 50 mm × 100 mm. The width is in the x-direction, the length is in the y-direction, and the height is in the z-direction. The dimensional error of the specimen in each direction is not greater than ±0.5 mm.

[0022] Preferably, in step 2:

[0023] There are at least 9 installation cavities in the same mining roadway, and at least 3 specimens of each type are installed;

[0024] The dimensions of the installation cavity are added with a safety margin based on the dimensions of the specimen. The depth, width, and height of the installation cavity respectively satisfy the following formulas:

[0025]

[0026] Wherein, L chamber , W chamber , H chamber are respectively the depth, width, and height of the installation cavity; L sample , W sample , H sample are respectively the length, width, and height of the specimen, δ is the installation margin, L max After determining the width and height, its depth direction is determined by the achievable safe excavation depth, L max not less than 1 m;

[0027] The specimen installation cavity closest to the coal mining face is at least 400 m away from the coal mining face.

[0028] The system compares the real-time collected loading force data with the target pre-tightening force. After calculating the error, a control signal is generated through the PID controller u k , and this signal drives the hydraulic actuator to adjust the pressure parameters so that the loading force gradually approaches the target value:

[0029] (4).

[0030] Preferably, step 4 includes fine-tuning control, and the fine-tuning control is as follows: ​

[0031] When the error is small, capture the subtle changes in the stress error at a higher sampling frequency and calculate:

[0032] (5)

[0033] Update the PID parameters using an adaptive algorithm based on the online error and the change trend:

[0034] (6)

[0035] Where σ target,x , σ target,y and σ target,z are the target stress values of the specimen in the three directions of x, y, and z, σ x , σ y、 σ z are the actual stress values of the specimen in the three directions of x, y, and z, α p , α i and α d are the adaptive factors, f k is the system dynamic prediction function.

[0036] Preferably, in step 5, the data acquisition method and the signal processing for each sensor are as follows:

[0037] Let the true signal of the sensor be S j ( t ), the measurement noise be N j ( t ), and let the total number of sensors be N . Then the continuous output signal of the sensor can be described as:

[0038] (7)

[0039] Unify the sampling period of all sensors to T s , k which is used to represent the sampling point index during discrete sampling, i.e., the k th sampling point. Then the discrete sampling time t k can be defined as:

[0040] (8)​

[0041] Collect data from each sensor at each sampling moment to obtain discrete sampling data:

[0042] (9)

[0043] To ensure the synchronization of multi-channel acquisition, it is required that the timestamps of each channel satisfy:

[0044] (10)

[0045] ξ is a very small allowable deviation.

[0046] After that, the data collected from each channel at time t k is composed into a data vector:

[0047] (11)

[0048] To eliminate high-frequency noise, first perform low-pass filtering on the data vector X k , and set the impulse response of the filter to h n . Through convolution operation, the suppression of high-frequency noise is achieved, and the filtered signal data is obtained as:

[0049] (12)

[0050] where L is the filter length. For the data after low-pass filtering, perform discrete wavelet transform. The discrete wavelet transform decomposes the signal into different scales and time positions, W i is the wavelet coefficient, which reflects the frequency components of the signal in the local area. According to the set threshold λ perform soft threshold processing on the wavelet coefficients, and the criterion is:

[0051] (13)

[0052] The parameter λ is determined according to the noise level and the number of samples n as follows:

[0053] (14)

[0054] Reconstruct the signal through the inverse wavelet transform to obtain the denoised signal as . Perform the inverse wavelet transform on the processed wavelet coefficients to obtain the denoised signal as:

[0055] ​ (15)

[0056] To further improve data smoothness and estimation accuracy, a Kalman filter is used to perform state estimation on the denoised signal. In the formula, where A is the state transition matrix:

[0057] (16)

[0058] Error covariance prediction:

[0059] (17)

[0060] Among them, Q is the process noise covariance.

[0061] Set H as the observation matrix, R as the measurement noise covariance, and the Kalman gain is calculated as:

[0062] (18)

[0063] Then the updated state estimate is:

[0064] (19)

[0065] Update the error covariance as:

[0066] (20)

[0067] To reduce the transmission bandwidth requirement, after preprocessing, the denoised signal is data-compressed. The transform coding method is used for data compression, and then the data is converted into a bit stream through adaptive quantization and entropy coding. Denote the compression function as Q , then the final preprocessing output vector data is:

[0068] (21)

[0069] In summary, the overall operation model of the preprocessing module is:

[0070] (22)

[0071] After the preprocessed data vector Y[k] is stored by the data acquisition unit, it is transmitted in the form of a data packet. The data packet is defined as:

[0072] (23)

[0073] The entire data stream is represented as a continuous data set D:

[0074] (24)

[0075] For a multi-channel configuration, each unit m generates a data set D m , containing its respective timestamp and preprocessed data. Let the data set of the m th unit be denoted as:

[0076] (25)

[0077] The data integration system performs time alignment and fusion on the data of each unit to form an overall data set, providing a unified data source for global monitoring and subsequent analysis:

[0078] (26)

[0079] When constructing data transmission, a low-latency transmission network is selected to cooperate in constructing a data transmission model, and the data feedback delay is less than 50 ms:

[0080] (27)

[0081] The final feedback data is constructed at the ground control center, representing the data received by the ground side at time t , and its form is:

[0082] (28).

[0083] Preferably, step 6, constructing the time-dependent deformation model includes:

[0084] Construct the accumulated strain as follows:

[0085] (29)

[0086] where ε ij 0 is the initial strain, is the creep strain rate, A ij is the material constant reflecting the basic creep characteristics in different directions, σ ij ( t ) is the corresponding component of the stress tensor, expressed as σ x ( t ), σ y ( t ) σ z ( t );n ij is the stress sensitivity index; O ij is the activation energy, J is the gas constant, T C ( t ) is the temperature; g ( RH ( t )) is the humidity modulation function, D ij ( t ) is the anisotropic damage variable, m ij is the damage inhibition parameter;

[0087] The creep strain rate of the specimen in the ij direction is as follows:

[0088] (30)

[0089] where, B ij is the damage rate constant; p , q , r are the sensitivity indices describing the influence of stress and strain on damage accumulation; O D,ij is the damage activation energy; h ( RH ( t )) is the humidity modulation function.

[0090] Preferably, in step 7, based on the downhole transmission data comprehensive dataset , a deep reinforcement learning prediction model is constructed, which consists of the following modules:

[0091] First, perform a convolution operation on to extract local spatial features:

[0092] (31)

[0093] Then use LSTM to process the time series features output by the convolutional layer to capture the dynamic evolution of time-dependent deformation:

[0094] (32)

[0095] Map the LSTM output to the prediction target and output the strain increment Δ ε pre ( t k + Δt ) is predicted as follows:

[0096] (33)

[0097] Among them, θ CNN represents the weight parameters of the convolutional neural network layer; θ LSTM represents the weight parameters of the long short-term memory network layer; θ FC represents the weight parameters of the fully connected layer (output layer);

[0098] The deep reinforcement learning prediction model uses the difference between the actually measured strain increment Δ ε true ( t k + Δt ) and the prediction result to construct a reward function:

[0099] (34)

[0100] This reward function constitutes a feedback signal for online updating of model parameters to achieve adaptive adjustment and finally form a closed loop.

[0101] Preferably, the stress sensing devices corresponding to the top and bottom of the specimen completely cover the top and bottom surfaces of the specimen.

[0102] The present invention can truly reflect the influence of mining disturbance: By installing specimens in the undisturbed area during the implementation of the deep roadway surrounding rock project, the present invention can monitor the time-dependent deformation characteristics of the roadway surrounding rock under the change of mining disturbance stress in real time, so as to truly reflect the influence of mining disturbance on the behavior of rock mass.

[0103] The present invention can comprehensively capture anisotropic characteristics. By using multi-directional specimens and corresponding monitoring equipment, the anisotropic mechanical behavior of the roadway surrounding rock can be fully reflected. By using multi-sensor technology, key parameters such as stress, strain, displacement, and environmental temperature and humidity inside the rock are collected in real time at a high sampling rate, and the data of each sensor are ensured to be synchronously collected under a unified time reference, so as to obtain comprehensive and accurate rock mass state information.

[0104] The present invention adopts automatic monitoring and data analysis: The system realizes automatic data collection and remote transmission, greatly improving the monitoring efficiency and data reliability, and providing solid data support for on-site real-time decision-making.

[0105] The prediction model of the present invention has strong practicability: The time-dependent deformation prediction model constructed based on on-site monitoring data can accurately predict the deformation trend of the rock mass in the future for a period of time, and provide a scientific basis and early warning information for the safety support of the deep roadway surrounding rock, with high practical value.

[0106] The engineering application value of the present invention is high: through real-time data collection, online prediction and intelligent early warning, the system can discover potential risks in advance and achieve refined management. It provides a scientific basis for roadway surrounding rock control and safety support, as well as the health monitoring of roadway surrounding rock, helps to optimize the filling support plan for roadway surrounding rock, and improves the overall strength of roadway surrounding rock. Brief Description of the Drawings

[0107] Figure 1 is a flow chart of the present invention.

[0108] Figure 2 is a structural schematic diagram of the specimen.

[0109] Figure 3 is an installation schematic diagram of the specimen.

[0110] Figure 4 is a schematic diagram of the stress sensing device and the strain acquisition device.

[0111] Figure 5 is a structural schematic diagram of the hydraulic loading device and the data acquisition system.

[0112] In the figure, 1 - coal seam, 2 - upper hydraulic loading device, 3 - specimen, 4 - x-axis hydraulic loading device, 5 - lower hydraulic loading device, 6 - hydraulic device, 7 - data integration system, 8 - installation cavity, 9 - y-axis hydraulic loading device, 10 - ground control center, 11 - stress sensing device, 12 - deformation monitoring gauge, 13 - telescopic arm, 14 - extraction roadway. Detailed Embodiment

[0113] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0114] The present invention discloses a monitoring and prediction model for anisotropic time-dependent deformation of roadway surrounding rock under mining disturbance, as Figure 1 shown, and specifically includes the following steps:

[0115] Step 1, in-situ specimen preparation:

[0116] Select a depth of 100 m inside the advanced abutment pressure area of the coal mining face as the sampling location to ensure that the surrounding rock of the mined roadway is not disturbed by mining, so as to truly reflect the original state of the surrounding rock of the deep roadway. Use a drill pipe with a diameter of 200 mm for borehole sampling. This diameter not only ensures a sufficient coal core cross-sectional size but also maintains the integrity of the anisotropic structure inside the sample, providing sufficient materials for subsequent multi-directional specimen preparation. The drilled samples are promptly pre-treated, and temperature, humidity, and other environmental parameters are recorded to ensure that the samples are maintained in the in-situ state as much as possible during storage. Use suitable wrapping materials (such as anti-static films or thermal insulation materials) to seal and package the coal cores to prevent additional damage caused by changes in the external environment during transportation and storage. According to the natural bedding structure of the roadway surrounding rock, the coal cores are pre-treated and divided, as Figure 2 shown. Using high-precision numerically controlled cutting equipment, cut and sample along three directions: horizontal bedding at 0°, vertical bedding at 90°, and inclined bedding at 45° respectively, to form standard cuboid specimens of uniform size, with dimensions of 50 mm × 50 mm × 100 mm, and the error of each dimension of the specimen is not greater than ±0.5 mm. The sampling quantity in each direction must meet the statistical requirements to ensure the significance and representativeness of the data.

[0117] Step 2: Construction of the specimen installation cavity: To restore the stress state of the roadway surrounding rock under real mining disturbances, in this step, a specimen installation cavity is constructed in front of the mining face to accommodate the standard specimens prepared above and simulate the in-situ stress environment on-site.

[0118] Specifically, as Figure 3 shown, an installation cavity 8 is constructed in the coal seam 1 of the return airway 14 in the space area not disturbed by mining, ensuring that the position where the installation cavity is located can truly reflect the on-site stress field and fracture distribution characteristics in the later stage. To ensure that the specimens can fully restore the in-situ stress state in the installation cavity, the size of the installation cavity 8 should add a safety margin based on the specimen size. Theoretically, the depth, width, and height of the installation cavity should satisfy:

[0119] (1)

[0120] Among them, L chamber , W chamber , H chamber are the depth, width, and height of the installation cavity respectively; L sample , W sample , H sample are the length, width, and height of the specimen respectively (in this invention, the specimen size is 50 mm × 50 mm × 100 mm); δ is the installation margin, adjusted according to the monitoring equipment;L max After determining the width and height, the safe excavation depth that can be achieved in the depth direction is determined according to the actual site conditions, generally not less than 1 m. In practical applications, to ensure structural stability and construction process requirements, the size of the installation cavity is usually slightly larger than the above theoretical value. During the construction process, the roughness of the groove wall is strictly controlled, and the surface of the cavity is machined or manually trimmed to ensure flatness and rounded corners at the ends, thereby reducing local stress concentration. In order to obtain the anisotropic characteristics of the roadway surrounding rock, at least 9 specimen installation cavities should be arranged in the same roadway to carry out on-site monitoring experiments on the anisotropic time-dependent deformation of the deep roadway surrounding rock under actual mining disturbances. The specimen installation cavity closest to the working face should be more than 400 m away from the working face to facilitate the excavation of the installation cavity and the stability of the installation and stress recovery of the specimens.

[0121] Step 3: Specimen installation and monitoring system layout: This step aims to precisely install and pre-tighten load a standard specimen using six hydraulic loading devices with telescopic arms to achieve stress recovery and real-time monitoring in all directions (x, y, z axes). For this purpose, a step-by-step installation and adjustment scheme is adopted to comprehensively pre-tighten the specimen in the in-situ environment and ensure uniform and coaxial loading. Among them, facing the space cavity, the y-axis is defined as the depth direction of the space cavity; the z-axis is defined as the up and down direction of the specimen, where the bottom end of the specimen corresponds to the lower end of the z-axis and the upper end of the specimen corresponds to the upper end of the z-axis; the x-axis is defined as the left and right direction of the specimen. A total of six hydraulic loading devices are configured, and each device is equipped with a telescopic arm to facilitate flexible adjustment of the specimen position and loading direction during installation. The upper and lower z-axis hydraulic loading devices are respectively installed at the lower and upper ends of the installation cavity, and telescopic adjustment is used to apply balanced loading synchronously. The two x-axis hydraulic loading devices 4 on both sides are respectively installed on the left and right sides of the specimen, and telescopic adjustment is used to apply balanced loading synchronously. The y-axis hydraulic loading device 9: One of them is fixed at the center of the y-axis in the installation cavity, and the other is installed inside and outside the cavity and is equipped with a support device for the final pre-tightening adjustment in the y-axis direction. The six hydraulic loading devices are used in combination, enabling precise adjustment of the position during specimen installation, achieving multi-axial coaxial pre-tightening, and greatly improving the installation accuracy and loading uniformity. As Figure 4 shown, there is a stress sensing device 11 at the end of the hydraulic loading device, which is internally embedded with a fiber Bragg grating stress sensor to measure the three-directional stresses σ x , σ y , σ z . The stress sensing device is a cube, and the loading surface area of the cube in the z-axis direction is not less than the bottom surface area of the specimen (50 mm × 50 mm), and the loading surface areas of the cube in the y and x-axis directions are not less than the bottom surface area of the specimen (50 mm × 100 mm). On the telescopic arm 13 of the hydraulic loading device, a deformation monitor 12 is installed to monitor its axial displacement and deformation.

[0122] As Figure 5 shown, the specific process of specimen installation is as follows: First, on the y-axis inside the specimen installation cavity 8, fix the y-axis hydraulic loading device 9 at the exact center of the cavity wall to ensure that it can accurately collect deep stress data. Install the lower hydraulic loading device 5 at the lower end of the specimen installation cavity 8. Place the bottom end of the specimen 3 (corresponding to the lower end of the z-axis) on the lower hydraulic loading device 5 so that the bottom of the specimen is coaxial with the loading surface of the device. Put the z-axis upper hydraulic loading device 2 with the bottom end of the fixed specimen 3 into the installation cavity together. Adjust the telescopic arm of the lower hydraulic loading device 5 to achieve precise coaxiality between the specimen 3 and the aforementioned y-axis hydraulic loading device 9 in the y-axis direction. Install two x-axis hydraulic loading devices 4 on the left and right sides of the specimen respectively. Adjust the telescopic arms on both sides simultaneously to make them move synchronously, achieving balanced preloading of the specimen in the x-axis direction, ensuring consistent loading on both left and right sides, and preventing lateral displacement caused by unbalanced loading. Install a hydraulic loading device at the upper end of the specimen, and also use the telescopic arm to dock with the upper surface of the specimen. Adjust the upper hydraulic loading device to ensure that it works together with the lower hydraulic loading device to uniformly preload the specimen in the z-axis direction. Install a hydraulic device 6 at the open end of the specimen installation cavity 8, supplemented by an external support device to ensure the stability of the device during adjustment. Adjust this device to achieve the final preloading with the specimen in the y-axis direction. Finally, apply the initial preloading force through the telescopic arms of all six hydraulic devices, making the preloading forces in the six directions consistent, but the preloading force should not be too large.

[0123] In addition, add miniature temperature and humidity sensors inside the cavity to monitor the local temperature in real time. Install a data integration system 7 underground. Each sensor is connected to the data acquisition unit (DAQ) through a dedicated interface. This unit supports high sampling rate and multi-channel synchronous acquisition, ensures the unity of data timestamps, and finally aggregates the data into the data integration system. The data integration system is connected to the underground communication and sensing network, and finally transmits the data back to the ground control center 10 to achieve safe and effective monitoring.

[0124] During this process, a preloading force feedback system is constructed. By real-time monitoring the force sensors on each hydraulic loading system, real-time control is achieved, and the parameters of the hydraulic device are adjusted in a timely manner to control the loading error within 5% to ensure balanced preloading in each axial direction. Specifically, set the sampling frequency of the stress sensor to T s , and at each sampling moment k collect the actual loading force recorded currently as F actual k , and denote the target preloading force as F target , then the error e k can be calculated as follows:

[0125] ​​ (2)

[0126] Adopt a closed-loop control algorithm based on PID control to generate a regulation instruction u k and timely adjust the parameters of the hydraulic device (such as oil pressure, flow rate, etc.) to compensate for errors:

[0127] (3)

[0128] Among them, K p is the proportional gain, which controls the amplification of the instantaneous error; K i is the integral gain, which accumulates past errors to eliminate the steady-state deviation; K d is the derivative gain, which predicts the future error trend to suppress oscillations. To cope with the nonlinearity and uncertainty of the on-site environment, the pre-tightening force feedback system introduces an adaptive correction algorithm. The system compares the real-time collected loading force data with the target pre-tightening force. After calculating the error, a regulation signal is generated through the PID controller u k , and this signal drives the hydraulic actuator to adjust the pressure parameter, so that the loading force gradually approaches the target value:

[0129] (4)

[0130] Step 4, in-situ test environment construction and in-situ stress recovery: According to the on-site exploration data, determine the target stress values of the specimen in the x, y, and z directions, which are respectively denoted as σ target,x , σ target,y and σ target,z , and the actual stress values are respectively σ x , σ y、 σ z . Adopt an independent and synchronous loading scheme in each direction. Through a unified hydraulic loading system, ensure that the pre-tightening force is applied by the loading devices on each axis at the same time, and realize independent regulation, truly simulating the in-situ stress state. To ensure the loading stability, a fine-tuning control theory is constructed:

[0131] When the error is small, capture the subtle changes of the stress error at a high sampling frequency and calculate:

[0132] (5)

[0133] Update the PID parameters using the adaptive algorithm according to the online error and change trend:​​

[0134] (6)

[0135] Wherein α p 、 α i and α d are adaptive factors, f k is the system dynamic prediction function. This fine-tuning strategy ensures that when the system approaches the target stress state, the loading can be fine-tuned in a smooth and highly accurate manner, further reducing errors.

[0136] Step 5, data acquisition: Through a multi-sensor data acquisition platform, multi-dimensional parameter data of the specimen during the loading process are obtained in real time and long term. It includes stress data of the specimen collected in real time by stress sensors in the x, y, and z directions σ x ( t )、 σ y ( t ) 、 σ z ( t ), strain data recorded by strain gauges installed on the telescopic arm of the hydraulic loading device ε ij ( t ) and displacement data d i ( t ), local temperature T C 、humidity RH and other environmental data are collected to correct the thermal stress effect and other environmental factors during the accelerated test process.

[0137] The present invention specifically constructs a data storage and transmission model for storing and transmitting on-site monitoring data. It includes stress data σ x ( t ), σ y ( t ), 、 σ z ( t ), strain data ε ij ( t ) and displacement data d i ( t ), temperature T C 、humidity RH。 ​Through the acquisition unit (DAQ), high sampling rate and multi-channel synchronous acquisition are achieved, ensuring the unity of data timestamps. Finally, the data is aggregated in the data integration system and transmitted back to the ground control center to achieve safe and effective monitoring. The model is as follows:

[0138] For each sensor (such as stress, strain, displacement, temperature sensor, humidity sensor), let its true signal be S j ( t ), and the measurement noise be N j ( t ). Let the total number of sensors be N , then the continuous output signal of the sensors can be described as:

[0139] (7)

[0140] Unify the sampling period of all sensors to T s , k which is used to represent the sampling point index during discrete sampling, that is, the k th sampling point. Then the discrete sampling time t k can be defined as:

[0141] (8)

[0142] At each sampling moment, data is collected from each sensor to obtain discrete sampling data:

[0143] (9)

[0144] To ensure the synchronization of multi-channel acquisition, it is required that the timestamps of each channel satisfy:

[0145] (10)

[0146] ξ is a very small allowable deviation.

[0147] After that, the data collected from each channel at time t k is composed into a data vector:

[0148] (11)

[0149] To improve data quality and facilitate subsequent transmission, this system has a multi-level preprocessing module built into the data acquisition unit. To eliminate high-frequency noise, first, the data vector X k is subjected to low-pass filtering. Set the impulse response of the filter to h n ​, through convolution operation, the suppression of high-frequency noise is achieved, and the filtered signal data obtained is:

[0150] (12)

[0151] Among them, L is the filter length. This operation can adopt FIR filter design to make the system have a stable frequency-domain response. For the data after low-pass filtering, discrete wavelet transform is performed. Wavelet transform decomposes the signal into different scales and time positions, W i Wavelet coefficients reflect the frequency components of the signal in the local area. According to the set threshold λ Soft threshold processing is performed on the wavelet coefficients, and the specific criterion is:

[0152] (13)

[0153] Parameter λ is usually determined according to the noise level (estimated by the median absolute deviation) and the number of samples n as follows:

[0154] (14)

[0155] By inverse wavelet transform the signal is reconstructed, and the denoised signal obtained is W ( k ). Performing inverse wavelet transform on the processed wavelet coefficients, the denoised signal obtained is:

[0156] (15)

[0157] To further improve the data smoothness and estimation accuracy, a Kalman filter is used to perform state estimation on the denoised signal. Among them, A is the state transition matrix:

[0158] (16)

[0159] Error covariance prediction:

[0160] (17)

[0161] Among them, Q is the process noise covariance.

[0162] Set H as the observation matrix, R as the measurement noise covariance, and the Kalman gain is calculated as:

[0163] ​ (18)

[0164] Then the updated state estimate is as follows:

[0165] (19)

[0166] The updated error covariance is:

[0167] (20)

[0168] This step adaptively smooths the acquired data, providing a signal with a higher signal-to-noise ratio for subsequent data compression and transmission. To reduce the transmission bandwidth requirement, the denoised signal is data-compressed after preprocessing. The transform coding method is used for data compression, and then the data is converted into a bit stream through adaptive quantization and entropy coding. Denote the compression function as Q , then the final preprocessing output vector data is:

[0169] (21)

[0170] In summary, the overall operation model of the preprocessing module can be expressed as:

[0171] (22)

[0172] After multi-stage preprocessing, not only the signal quality is significantly improved, but also the data volume is greatly reduced, facilitating subsequent low-latency transmission.

[0173] After the preprocessed data vector Y[k] is stored by the data acquisition unit, it is transmitted in the form of data packets. The data packet is defined as:

[0174] (23)

[0175] Therefore, the entire data stream can be represented as a continuous data set D:

[0176] (24)

[0177] For the multi-channel configuration, each unit m generates a data set D m , containing its own timestamp and preprocessed data. Denote the data set of the m th unit as:

[0178] (25)

[0179] The data integration system performs time alignment and fusion on the data of each unit, and can form an overall data set, providing a unified data source for global monitoring and subsequent analysis:

[0180] (26)

[0181] This process involves timestamp correction and multi-sensor data fusion algorithms to ensure high consistency in time and content for each data packet.

[0182] After integrating and storing the data in the downhole system, it is transmitted through the downhole communication network. To achieve real-time monitoring, when constructing the data transmission, a low-latency transmission network should be selected to cooperate in constructing the data transmission model, and the data backhaul latency is less than 50 ms:

[0183] (27)

[0184] Therefore, the final backhaul data is constructed at the ground control center, indicating the data received by the ground side at time t , and its form is:

[0185] (28)

[0186] By correcting the time offset, it is ensured that the data has the correct time reference during analysis.

[0187] Step 6: Construct the time-dependent deformation model: Based on the comprehensive data set transmitted back to the ground control center in Step 5 , which includes stress data σ x ( t ), σ y ( t ), σ z ( t ), strain data ε ij (t), displacement data d i ( t ), and environmental parameters temperature T C ( t ) and humidity RH ( t ), construct an anisotropic time-dependent deformation prediction model for the surrounding rock of deep roadway under real mining disturbance. First, combining the measured parameters of the real downhole environment, using a combination of damage mechanics and creep theory, construct the cumulative strain as:

[0188] (29)

[0189] Among them, ε ij 0 is the initial strain, is the creep strain rate; A ij is a material constant that reflects the basic creep characteristics in different directions; σ ij ( t ) represents the corresponding component of the stress tensor and can be expressed as σ x ( t ), σ y ( t ) σ z ( t ); n ij is the stress sensitivity index; O ij is the activation energy, J is the gas constant, T C ( t ) is the temperature; g ( RH ( t )) is the humidity modulation function, which can be a linear or exponential function and describes the influence of humidity on the creep rate; To characterize the influence of material damage on the creep process, an anisotropic damage variable D ij ( t ) is introduced to describe the degree of local structural degradation of the material; m ij describes the damage inhibition effect.

[0190] Under the framework of damage mechanics and creep theory, this method proposes a coupling model. The creep strain rate of the specimen in the ij direction is described by the following equation:

[0191] (30)

[0192] where, B ij is the damage rate constant; p 、 q 、 r are the sensitivity indices that describe the influence of stress and strain on damage accumulation; O D,ij is the damage activation energy; h ( RH ( t )) is the humidity modulation function. This model considers the coupling effect of stress, accumulated strain and the remaining intact rate on the damage rate.

[0193] Step 7: Construct an aging deformation prediction model: Based on the comprehensive dataset of downhole transmission data , a deep reinforcement learning prediction model is constructed to predict the strain increment in the future time period. The model mainly consists of the following modules:

[0194] First, perform a convolution operation on to extract local spatial features:

[0195] (31)

[0196] Then, use LSTM to process the time series features output by the convolutional layer to capture the dynamic evolution of time-dependent deformation:

[0197] (32)

[0198] Map the LSTM output to the prediction target and output the strain increment Δ in the future time period ε pre ([[]]END]] t k + Δt ) The prediction is as follows:

[0199] (33)

[0200] In the above three formulas, θ CNN represents the weight parameters of the convolutional neural network layer; θ LSTM represents the weight parameters of the long short-term memory network layer; θ FC represents the weight parameters of the fully connected layer (output layer).

[0201] The model uses the difference between the actually measured strain increment Δ ε true ([[]]END]] t k + Δt ) and the prediction result to construct a reward function:

[0202] (34)

[0203] This reward function constitutes a feedback signal for online updating of the model parameters to achieve adaptive adjustment, and finally forms a closed loop.

[0204] Step 8: Apply the prediction model to the prediction of time-dependent deformation, such as the control and safety support of surrounding rock in deep roadway and the health monitoring of roadway surrounding rock. By comparing the prediction results with the real-time monitoring data, the system pre-sets a safety threshold for strain increment. When the predicted future strain increment reaches or exceeds this safety threshold, the system automatically triggers an early warning. The early warning signal will be transmitted to the ground control center through the underground communication network and further sent to the on-site monitoring terminal. When the prediction shows a sharp increase in strain and there is a potential risk of instability, the system can recommend timely addition of support structures or local pressure relief treatment to reduce the risk of sudden accidents. In addition, the system also supports automatic recording of early warning events and related parameters, providing data basis for subsequent accident investigation and safety assessment. After receiving the underground data and early warning information, the ground control center generates an early warning report to help decision-makers timely understand the safety status of the surrounding rock in deep roadway and organize on-site emergency disposal.

[0205] Of course, the present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and deformations according to the present invention, but these corresponding changes and deformations should all fall within the protection scope of the appended claims of the present invention.

Claims

1. The anisotropic time-dependent deformation monitoring and prediction model of tunnel surrounding rock under mining disturbance is characterized by: Model building includes the following steps: Step 1, in-situ sample preparation, select the tunnel surrounding rock that is not disturbed by mining in the advanced support pressure area of ​​the coal mining face, obtain coal samples by coring, cut along the horizontal bedding 0°, vertical bedding 90°, and inclined bedding 45° of the coring coal samples, and obtain three kinds of samples. The three samples have the same size and are rectangular; Step 2: constructing a sample installation cavity, constructing an installation cavity in a surrounding rock area of ​​the mining tunnel that is not disturbed by mining in front of the coal mining face, wherein the size of the installation cavity is larger than the sample; Step 3: Sample installation and monitoring system arrangement: Place the sample in the installation cavity, arrange a hydraulic loading device at the open end of the installation cavity, support the sample and the five side walls of the installation cavity through the hydraulic loading device, and the hydraulic loading device is supported on the end surface of the sample, finally forming a three-axis support; The loading end of the hydraulic loading device is provided with a stress sensing device, a fiber Bragg grating stress sensor is embedded in the stress sensing device, a strain gauge is arranged on the telescopic arm of the hydraulic loading device, a temperature sensor and a humidity sensor are arranged in the installation cavity, the stress sensor, the strain gauge, the temperature sensor and the humidity sensor are all connected to a data acquisition device, and the data acquisition device is connected to a ground control center; The hydraulic loading device is controlled by a ground control center; Step 4: In-situ test environment construction and in-situ stress recovery: According to the field exploration data, the target value of the in-situ ground stress of the sample is determined, which is spatially represented as three directions: x, y, and z. A preload is applied to the sample through a hydraulic loading device to simulate the in-situ stress state; Step 5: Data collection: Under the disturbance of mining, real-time data from stress sensors, strain gauges, temperature sensors, and humidity sensors are collected to correct thermal stress effects and other environmental factors during the accelerated test; Step 6: construct a time-dependent deformation model based on the data obtained in step 5; Step 7: Construct a time-dependent deformation prediction model. Based on the data obtained in step 5, construct a deep reinforcement learning prediction model to predict the strain increment in the future period.

2. The anisotropic time-dependent deformation monitoring and prediction model of tunnel surrounding rock under mining disturbance according to claim 1 is characterized in that: After step 7, also include: Step 8: Apply the time-dependent deformation prediction model to mine tunnel support design, coal mining face advancement speed optimization, and rock burst prevention and control.

3. The anisotropic time-dependent deformation monitoring and prediction model of tunnel surrounding rock under mining disturbance according to claim 1 is characterized in that: In step 1: The sampling location was selected at a depth of 100m inside the advanced support pressure zone of the coal mining face, and the coal core was drilled using a drill rod with a diameter of 200mm. The dimensions of the sample are 50mm × 50mm × 100mm in width × length × height, with the width being in the x direction, the length being in the y direction, and the height being in the z direction. The dimensional error of the sample in each direction is no more than ±0.5mm.

4. The anisotropic time-dependent deformation monitoring and prediction model of tunnel surrounding rock under mining disturbance according to claim 1 is characterized in that: In step 2: There are at least 9 installation cavities in the same mining tunnel, and at least 3 of each type of specimen are installed; The size of the installation cavity is based on the size of the specimen with a safety margin added, and the depth, width, and height of the installation cavity satisfy the following formulas: ； in, L chamber , W chamber , H chamber They are the depth, width and height of the installation cavity respectively; L sample , W sample , H sample are the length, width and height of the specimen respectively, δ For installation allowance, L max After determining the width and height, the depth direction is determined by the achievable safe excavation depth. L max Not less than 1 m; The specimen installation cavity closest to the coal mining face is at least 400 m away from the coal mining face.

5. The anisotropic time-dependent deformation monitoring and prediction model of tunnel surrounding rock under mining disturbance according to claim 1 is characterized in that: The stress sensing devices corresponding to the top and bottom of the specimen completely cover the top and bottom surfaces of the specimen.

Images