Coal pillar damage dynamic monitoring system and method based on multi-physics field coupling and distributed optical fiber sensing

Abstract

The invention discloses a coal pillar damage dynamic monitoring system and method based on multi-physics coupling and distributed optical fiber sensing, and relates to the technical field of dynamic monitoring. Multi-physics coupling data of a coal pillar is collected in real time through a distributed optical fiber sensing network, and a health index of a coal pillar structure is obtained based on a trained time sequence multi-scale mapping model; the damage evolution rate of the coal pillar is calculated, and if the damage evolution rate is lower than a preset first threshold value, the sampling frequency of the distributed optical fiber sensing network is adjusted, and the data acquisition density of the distributed optical fiber sensing network is reduced; increasing the sampling frequency of the distributed optical fiber sensing network and constructing a trend prediction model to generate evolution prediction of the coal pillar damage if the coal pillar damage is not lower than the preset first threshold value but lower than the preset second threshold value; and if not, switching the distributed optical fiber sensing network to a high-frequency sampling mode, and generating the evolution acceleration of the coal pillar damage. The optimal configuration of the monitoring resources and the dynamic adjustment of the monitoring intensity are realized.

Description

Technical Field

[0001] The present invention relates to the field of dynamic monitoring technology, and in particular to a system and method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing. Background Art

[0002] Coal pillars are the coal structures that remain during the mining process. They serve as a key component supporting the roof and maintaining mine stability. In underground coal mining, coal pillars bear the important responsibility of supporting the weight of the overlying rock strata, preventing roof collapse, and protecting the safety of the mining space. With increasing mining depth and increasing mining intensity, the stress environment to which the coal pillars are subjected becomes increasingly complex. The coupling of multiple physical fields (such as stress, temperature, and acoustic fields) causes gradual damage to the internal structure of the coal pillars. In severe cases, this can lead to disasters such as coal pillar instability, rock bursts, and roof collapse. Therefore, real-time, accurate, and dynamic monitoring of coal pillar damage is necessary.

[0003] However, most existing monitoring systems are passive response types, lacking active and graded warning mechanisms. They are unable to provide differentiated monitoring strategies and warning plans for different degrees of damage, and mainly rely on the measurement of single physical field parameters, such as stress monitoring, displacement monitoring, or acoustic emission monitoring, which cannot fully reflect the complex damage evolution process inside the coal pillar.

[0004] To this end, the present invention proposes a coal pillar damage dynamic monitoring system and method based on multi-physical field coupling and distributed optical fiber sensing. Summary of the Invention

[0005] The present invention aims to address at least one of the technical problems existing in the prior art. To this end, it proposes a dynamic coal pillar damage monitoring system and method based on multi-physics field coupling and distributed fiber optic sensing. This system optimizes the allocation of monitoring resources and dynamically adjusts monitoring intensity, significantly improving the system's intelligence and resource utilization efficiency.

[0006] To achieve the above objectives, a dynamic monitoring method for coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing is proposed, which includes the following steps:

[0007] Step 1: Real-time acquisition of multi-physics coupling data of coal pillars through a distributed optical fiber sensing network;

[0008] Step 2: pre-training a time series multi-scale mapping model, and obtaining a coal pillar structural health index based on the multi-physics field coupling data and the time series multi-scale mapping model;

[0009] Step 3: Calculate the damage evolution rate of the coal pillar based on the coal pillar structural health index. If the damage evolution rate is lower than a preset first threshold, proceed to step 4. If the damage evolution rate is not lower than the preset first threshold but lower than a preset second threshold, proceed to step 5. If the damage evolution rate is not lower than the preset second threshold, proceed to step 6.

[0010] Step 4: Based on the multi-physics field coupling data and the damage evolution rate, an adaptive sampling dynamic strategy is used to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network;

[0011] Step 5: Based on the multi-physics field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight allocation algorithm is used to weightedly fuse the stress field data and temperature field data in the multi-physics field coupling data to construct a trend prediction model and generate an evolution prediction of coal pillar damage;

[0012] Step 6: Switch the distributed optical fiber sensing network to a high-frequency sampling mode, and use a nonlinear coupling analytical algorithm to collaboratively analyze the multi-physical field coupling data to generate the evolution acceleration of coal pillar damage.

[0013] The real-time acquisition of multi-physical field coupling data of coal pillars through a distributed optical fiber sensing network comprises the following steps:

[0014] The distributed optical fiber sensing network is partitioned and arranged according to the structural characteristics of the coal pillars, and then the distributed optical fiber sensing network collects stress field distribution tensors, temperature field distribution tensors, and electromagnetic field distribution tensors from different partitioned areas at a predetermined collection frequency to form the multi-physical field coupling data;

[0015] The obtaining of the coal pillar structure health index comprises the following steps:

[0016] Step 21: extracting time series features from the multi-physics field coupling data using a sliding window technique to construct a multi-physics field time series feature tensor;

[0017] Specifically, the multi-physics field time series feature extraction utilizes a sliding window technique to extract time-dimensional features from the stress, temperature, and electromagnetic field distribution tensors. Statistical features and frequency-domain features are then calculated for the stress, temperature, and electromagnetic field intensity within each sliding window. For stress field data, the stress gradient and principal stress direction are additionally calculated; for temperature field data, the temperature change rate and spatial temperature gradient are additionally calculated; and for electromagnetic field data, the electromagnetic field intensity change rate and spatial distribution features are additionally calculated.

[0018] The multi-physical field time series feature tensor finally obtained is organized in a four-dimensional tensor structure with the dimensions [T, S, F, C], where T represents the number of time windows, S represents the number of spatial monitoring points, F represents the total amount of extracted statistical features and frequency domain features, and C represents the number of physical field channels (stress field, temperature field, and electromagnetic field).

[0019] Step 22: Construct a time series multi-scale mapping model based on deep learning, wherein the time series multi-scale mapping model includes three units: a time series encoder, a spatial feature extractor, and a multi-scale fusion network;

[0020] The temporal encoder utilizes a bidirectional long short-term memory (LSTM) network structure, comprising three layers of BiLSTM units, each containing 128 hidden neurons. The encoder receives the multi-physics field temporal feature tensor as input to capture the long-term dependencies and short-term fluctuations of parameters across different physical fields over time. The forward propagation of the BiLSTM captures the impact of the current state on future states, while the backward propagation captures the impact of past states on the current state, forming a complete temporal feature representation.

[0021] The spatial feature extractor uses a graph convolutional network structure to construct a spatial graph structure from pre-set coal pillar monitoring points. Each node in this spatial graph corresponds to a coal pillar monitoring point, and the weights of the edges in the spatial graph are calculated based on the spatial distance between each coal pillar monitoring point and the correlation between physical field parameters. The graph convolutional network structure contains three graph convolution layers, with 64, 128, and 256 output channels per layer, respectively. This allows the spatial feature extractor to capture the spatial correlation and stress transfer characteristics between different regions of the coal pillar, enabling the mapping of microscopic monitoring data to macroscopic structural features.

[0022] The multi-scale fusion network utilizes a multi-scale feature fusion architecture with an attention mechanism. It consists of three parallel convolutional branches with kernel sizes of 3×3, 5×5, and 7×7, corresponding to different receptive fields. Each branch contains two convolutional layers, each with 64 output channels. Multi-scale features are adaptively weighted and fused using a channel-wise attention mechanism to generate a comprehensive feature representation. The output of the fusion network passes through two fully connected layers with a hidden layer size of 256, resulting in the final output layer representing the coal pillar structural health index.

[0023] Step 23: pre-collect historically acquired multi-physics field time series feature tensors and corresponding coal pillar structure state labels, and pre-train the time series multi-scale mapping model;

[0024] Step 24: inputting the multi-physics field time series feature tensor into the pre-trained time series multi-scale mapping model to generate a coal pillar structure health index;

[0025] The multi-physics field time series feature tensor is forward propagated through the time series encoder, spatial feature extractor, and multi-scale fusion network of the time series multi-scale mapping model. The time series encoder outputs a time series feature representation, and the spatial feature extractor outputs a spatial structure feature representation. The multi-scale fusion network fuses these time series and spatial structure features and maps them into a coal pillar structural health index. The coal pillar structural health index is output as structured data and contains at least the following health index information: an overall health index representing the overall stability of the coal pillar and a local damage index matrix.

[0026] Calculating the damage evolution rate of the coal pillar according to the coal pillar structural health index comprises the following steps:

[0027] Step 31: Use the sliding time window method to construct a time series analysis window and perform time series segmentation on the continuously collected coal pillar structure health index data;

[0028] Step 32: Calculate the time derivative of the overall health index to obtain the global damage evolution rate.

[0029] Step 33: Calculate the time derivative of the local damage index matrix to obtain the spatially distributed local damage evolution rate field;

[0030] Step 34: Combine the maximum values ​​of the global damage evolution rate and the local damage evolution rate field to obtain the damage evolution rate of the coal pillar.

[0031] The method of using an adaptive sampling strategy to adjust the sampling frequency of the optical fiber sensor network and reduce the data acquisition density of the distributed optical fiber sensor network includes the following steps:

[0032] Step 41: Dynamically adjust the boundaries of each partition area based on the coal pillar structure health index;

[0033] Step 42: Apply a frequency control algorithm based on the damage evolution rate to each partitioned area using a nonlinear mapping function to calculate an appropriate sampling frequency;

[0034] Step 43: Based on the appropriate sampling frequency, dynamically adjust the pulse laser emission frequency of the distributed optical fiber sensing network;

[0035] Step 44: dynamically optimizing the density of sampling points of the coal pillar based on an adaptive grid refinement algorithm, and adjusting the distribution density of the sampling points of the coal pillar according to the gradient distribution of the local damage evolution rate field of the coal pillar structure;

[0036] The method of weighting and fusing the stress field data and the temperature field data in the multi-physics field coupling data using a dynamic weight distribution algorithm to construct a trend prediction model and generate an evolution prediction of coal pillar damage includes the following steps:

[0037] Step 51: construct a dynamic weight distribution algorithm to adaptively adjust the weight coefficients of stress field data and temperature field data according to the coal pillar damage evolution rate and the time-varying characteristics of the physical field data;

[0038] Step 52: Based on the dynamically assigned weight coefficients, the stress field data and the temperature field data are weightedly fused using a tensor decomposition and reconstruction method to construct a multivariate time series prediction model and generate an evolution prediction of the coal pillar damage;

[0039] Switching the distributed optical fiber sensing network to a high-frequency sampling mode and using a nonlinear coupling analytical algorithm to collaboratively analyze the multi-physics field coupling data to generate an evaluation result of the evolution acceleration of coal pillar damage and the instability risk includes the following steps:

[0040] Step 61: Raise the sampling frequency of the distributed optical fiber sensor network to a preset high frequency state using a partition-differentiated frequency-raising strategy;

[0041] Step 62: Construct a nonlinear coupling analytical model consisting of three components: an inter-field mapping layer, a nonlinear coupling layer, and a damage evolution prediction layer, to perform a collaborative analysis of the multi-physics field coupling data collected during the high-frequency sampling process;

[0042] Step 63: Calculate the evolution acceleration of coal pillar damage based on the analytical results of the nonlinear coupling analytical model;

[0043] A dynamic monitoring system for coal pillar damage based on multi-physics field coupling and distributed fiber optic sensing is proposed. The system includes a multi-physics field data collection module, a health index generation module, an evolution level judgment module, and a dynamic monitoring module. Each module is electrically connected to the other.

[0044] The multi-physics field data collection module collects the multi-physics field coupling data of the coal pillar in real time through a distributed optical fiber sensing network, and sends the multi-physics field coupling data to the health index generation module and the dynamic monitoring module;

[0045] A health index generation module pre-trains a time series multi-scale mapping model, obtains a coal pillar structure health index based on the multi-physics field coupling data and the time series multi-scale mapping model, and sends the coal pillar structure health index to the evolution level judgment module;

[0046] an evolution level judgment module, which calculates the damage evolution rate of the coal pillar according to the coal pillar structural health index and sends the damage evolution rate to the dynamic monitoring module;

[0047] A dynamic monitoring module determines that if the damage evolution rate is lower than a preset first threshold, then based on the multi-physical field coupling data and the damage evolution rate, an adaptive sampling dynamic strategy is used to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network; if the damage evolution rate is not lower than the preset first threshold but lower than the preset second threshold, then based on the multi-physical field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight distribution algorithm is used to weightedly fuse the stress field data and temperature field data in the multi-physical field coupling data to construct a trend prediction model and generate an evolution prediction of coal pillar damage; if the damage evolution rate is not lower than the preset second threshold, the distributed optical fiber sensing network is switched to a high-frequency sampling mode, and a nonlinear coupling analytical algorithm is used to collaboratively analyze the multi-physical field coupling data to generate an evolution acceleration of coal pillar damage.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention implements a distributed fiber optic sensing network within and around the coal pillar, simultaneously sensing multiple physical field parameters such as stress and temperature fields. Using dedicated signal demodulation equipment, fiber optic sensing signals are collected in real time and converted into multi-physics field coupling data. Using a pre-trained time-series multi-scale mapping model, the collected multi-physics field coupling data is converted into a coal pillar structural health index, which comprehensively reflects the overall health of the coal pillar. The coal pillar damage evolution rate is then calculated in real time and compared with a preset threshold. The corresponding monitoring strategy is automatically selected. Under different monitoring strategies, adaptive sampling adjustment, dynamic weight fusion analysis, or nonlinear coupling analysis are performed to generate corresponding monitoring results and early warning information. This overcomes the limitations of single-physics field monitoring and enables collaborative analysis of multi-physics field coupling data. This improves the accuracy and comprehensiveness of coal pillar damage monitoring, optimizes the allocation of monitoring resources, enhances the system's prediction and early warning capabilities, and provides reliable technical support for coal mine safety production. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing in Example 1 of the present invention;

[0051] Figure 2 This is a module connection diagram of the coal pillar damage dynamic monitoring system based on multi-physical field coupling and distributed optical fiber sensing in Example 1 of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] Example 1

[0054] like Figure 1 As shown, the dynamic monitoring method of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing includes the following steps:

[0055] Step 1: Real-time acquisition of multi-physics coupling data of coal pillars through a distributed optical fiber sensing network;

[0056] Step 2: pre-training a time series multi-scale mapping model, and obtaining a coal pillar structural health index based on the multi-physics field coupling data and the time series multi-scale mapping model;

[0057] Step 3: Calculate the damage evolution rate of the coal pillar based on the coal pillar structural health index. If the damage evolution rate is lower than a preset first threshold, proceed to step 4. If the damage evolution rate is not lower than the preset first threshold but lower than a preset second threshold, proceed to step 5. If the damage evolution rate is not lower than the preset second threshold, proceed to step 6.

[0058] Step 4: Based on the multi-physics field coupling data and the damage evolution rate, an adaptive sampling dynamic strategy is used to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network;

[0059] Step 5: Based on the multi-physics field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight allocation algorithm is used to weightedly fuse the stress field data and temperature field data in the multi-physics field coupling data to construct a trend prediction model and generate an evolution prediction of coal pillar damage;

[0060] Step 6: Switch the distributed optical fiber sensing network to a high-frequency sampling mode, and use a nonlinear coupling analytical algorithm to collaboratively analyze the multi-physical field coupling data to generate the evolution acceleration of coal pillar damage.

[0061] The real-time acquisition of multi-physical field coupling data of coal pillars through a distributed optical fiber sensing network includes the following steps:

[0062] The distributed optical fiber sensing network is partitioned and arranged according to the structural characteristics of the coal pillars, and then the distributed optical fiber sensing network collects stress field distribution tensors, temperature field distribution tensors, and electromagnetic field distribution tensors from different partitioned areas at a predetermined collection frequency to form the multi-physical field coupling data;

[0063] Specifically, in the specific implementation process of the present invention, the distributed optical fiber sensing network is composed of a laser emitting unit, a multi-mode optical fiber sensing network and a signal demodulation and analysis unit.

[0064] The laser emitting unit generates narrow-linewidth laser pulses of three different wavelengths, one for monitoring stress, one for monitoring temperature, and one for monitoring electromagnetic fields. Each narrow-linewidth laser pulse is injected into a special optical fiber buried around the coal pillar via a wavelength division multiplexer. The stress field data is collected based on the principle of Brillouin scattering. For example, when an optical fiber is subjected to external stress, the frequency shift of its scattered light is linearly related to the stress. The temperature field data is collected based on the principle of Raman scattering. For example, the temperature is calculated by analyzing the intensity ratio of Stokes light to anti-Stokes light in the backscattered light. The electromagnetic field data is collected using the Faraday effect. For example, the electromagnetic field intensity is determined by measuring the rotation angle of the polarization state.

[0065] According to the structural characteristics of the coal pillar, the distributed fiber optic sensing network is arranged according to the three-level zoning strategy of "core area-transition area-edge area". The core area adopts a high-density grid layout, for example, the monitoring point spacing is 0.5 meters, to focus on monitoring the area where the coal pillar is most stressed; the transition area adopts a medium-density ring layout, for example, the monitoring point spacing is 1 meter, to monitor the stress transfer process; the edge area adopts a low-density radial layout, for example, the monitoring point spacing is 2 meters, to monitor the interaction between the coal pillar and the surrounding rock mass. The monitoring data of the three areas are collected at different sampling frequencies, for example, 10Hz in the core area, 5Hz in the transition area, and 1Hz in the edge area, that is, three narrow linewidth laser pulses of different wavelengths are sent at different frequencies to collect stress field data, temperature field data, and electromagnetic field data according to different frequencies;

[0066] The signal demodulation and analysis unit adopts coherent detection technology to receive and analyze the pulse signals of the three narrow-linewidth laser pulses in each area, and realizes the synchronous measurement of multi-physical field parameters of each spatial point through the corresponding pulse analysis algorithm, such as stress field analysis through the linear relationship between frequency shift and stress, temperature field calculation through the intensity ratio of Stokes light and anti-Stokes light, and electromagnetic field calculation through measuring the rotation angle of polarization state; forming stress field distribution tensor, temperature field distribution tensor and electromagnetic field distribution tensor, which together constitute the multi-physical field coupling data after timestamp matching and spatial coordinate matching;

[0067] Furthermore, the time-series multi-scale mapping model is used to convert the stress, temperature and electromagnetic signal changes sensed by the distributed optical fiber sensor at the micron scale into the structural response characteristics of the coal pillar at the meter scale. The coal pillar structural health index is a comprehensive numerical parameter that characterizes the overall stability and local damage degree of the coal pillar. This comprehensive numerical parameter is updated in real time as the coal pillar state changes.

[0068] The obtaining of the coal pillar structure health index comprises the following steps:

[0069] Step 21: extracting time series features from the multi-physics field coupling data using a sliding window technique to construct a multi-physics field time series feature tensor;

[0070] Specifically, the multi-physics field time series feature extraction uses a sliding window technique to extract temporal features from the stress field distribution tensor, the temperature field distribution tensor, and the electromagnetic field distribution tensor. In this embodiment of the present invention, the sliding window length of this technique is set to 60 seconds with a step size of 10 seconds to ensure that the temporal continuity characteristics of the coal pillar structural response are captured.

[0071] Then, for each sliding window, the corresponding statistical features and frequency domain features are calculated for stress, temperature, and electromagnetic field intensity. Specifically, the statistical features include mean, standard deviation, kurtosis, skewness, maximum, minimum, and quartiles; the frequency domain features are extracted using fast Fourier transform to extract power spectral density and dominant frequency components.

[0072] In a further preferred embodiment, for stress field data, stress gradient and principal stress direction are additionally calculated; for temperature field data, temperature change rate and spatial temperature gradient are additionally calculated; for electromagnetic field data, electromagnetic field intensity change rate and spatial distribution characteristics are additionally calculated.

[0073] The multi-physical field time series feature tensor finally obtained is organized in a four-dimensional tensor structure with the dimensions [T, S, F, C], where T represents the number of time windows, S represents the number of spatial monitoring points, F represents the total amount of extracted statistical features and frequency domain features, and C represents the number of physical field channels (stress field, temperature field, and electromagnetic field).

[0074] Step 22: Construct a time series multi-scale mapping model based on deep learning, wherein the time series multi-scale mapping model includes three units: a time series encoder, a spatial feature extractor, and a multi-scale fusion network;

[0075] Specifically, in the specific implementation process of the present invention, the time series encoder adopts a bidirectional long short-term memory network structure, including three layers of Bi LSTM units, each layer containing 128 hidden neurons. The time series encoder receives the multi-physics field time series feature tensor as input to capture the long-term dependencies and short-term fluctuation characteristics of the parameters of different physical fields over time. The forward propagation of the Bi LSTM captures the impact of the current state on the future state, and the backward propagation captures the impact of the historical state on the current state, forming a complete time series feature representation.

[0076] The spatial feature extractor uses a graph convolutional network structure to construct a spatial graph structure from pre-set coal pillar monitoring points. Each node in this spatial graph corresponds to a coal pillar monitoring point, and the weights of the edges in the spatial graph are calculated based on the spatial distance between each coal pillar monitoring point and the correlation between physical field parameters. The graph convolutional network structure contains three graph convolution layers, with 64, 128, and 256 output channels per layer, respectively. This allows the spatial feature extractor to capture the spatial correlation and stress transfer characteristics between different regions of the coal pillar, enabling the mapping of microscopic monitoring data to macroscopic structural features.

[0077] The multi-scale fusion network utilizes a multi-scale feature fusion architecture with an attention mechanism. It consists of three parallel convolutional branches with kernel sizes of 3×3, 5×5, and 7×7, corresponding to different receptive fields. Each branch contains two convolutional layers, each with 64 output channels. Multi-scale features are adaptively weighted and fused using a channel-wise attention mechanism to generate a comprehensive feature representation. The output of the fusion network passes through two fully connected layers with a hidden layer size of 256, resulting in the final output layer representing the coal pillar structural health index.

[0078] Step 23: pre-collect historically acquired multi-physics field time series feature tensors and corresponding coal pillar structure state labels, and pre-train the time series multi-scale mapping model;

[0079] Specifically, the pre-training of the temporal multi-scale mapping model uses a supervised learning approach. In the implementation of the present invention, the training data set consists of 10,000 sets of samples, each of which contains a 60-second multi-physics field temporal feature tensor and a corresponding structural state label. The structural state label is obtained through general expert evaluation and finite element analysis, and includes an overall stability score and a quantified value of the local damage degree. It is understood that the collected 60-second multi-physics field temporal feature tensor is obtained by extracting the multi-physics field coupling data collected during this time period using the feature extraction method of step 21;

[0080] In this example, the pre-training process utilizes a batch stochastic gradient descent optimization algorithm with a batch size of 32, an initial learning rate of 0.001, and a cosine annealing learning rate scheduling strategy. The loss function is designed as a weighted sum of mean squared error loss and structural correlation loss, where the structural correlation loss is used to maintain consistency between the prediction results and the physical model. An early stopping strategy is introduced during training, terminating training when the validation set loss stops decreasing for 10 consecutive epochs.

[0081] In a further preferred embodiment, to enhance the generalization capability of the temporal multi-scale mapping model, data augmentation techniques are employed during training, including random time window offset, Gaussian noise injection, and random masking. Regularization techniques, including L2 weight regularization and dropout (with a dropout rate of 0.2), are also introduced to prevent model overfitting.

[0082] Step 24: inputting the multi-physics field time series feature tensor into the pre-trained time series multi-scale mapping model to generate a coal pillar structure health index;

[0083] Specifically, the multi-physics field time series feature tensor is forward propagated through the time series encoder, spatial feature extractor, and multiscale fusion network of the time series multi-scale mapping model. The time series encoder outputs a time series feature representation, while the spatial feature extractor outputs a spatial structure feature representation. The multiscale fusion network fuses these time series and spatial structure features and maps them into a coal pillar structural health index.

[0084] In this embodiment, the coal pillar structural health index is output in the form of structured data, which includes at least the following health index information: an overall health index, such as a scalar value of 0-100, indicating the overall stability state of the coal pillar; a local damage index matrix, such as a matrix corresponding to the spatial distribution of monitoring points, indicating the degree of damage in each area; key risk area coordinates, such as the spatial location of identified high-risk areas; damage type classification results, such as categories including compression damage, shear damage, and tension damage; and a reliability assessment value, such as an indication of the confidence level of the prediction results.

[0085] To more easily assess the health of a coal pillar, this embodiment also provides a health grade assessment method. For example, an overall health index value between 90 and 100 in the coal pillar structural health index indicates good structural condition, 70-90 indicates minor damage, 50-70 indicates moderate damage, 30-50 indicates severe damage, and values ​​below 30 indicate imminent instability. High-value regions in the local damage index matrix correspond to critical areas of stress concentration or damage development within the coal pillar.

[0086] Furthermore, the calculation of the damage evolution rate of the coal pillar according to the coal pillar structural health index includes the following steps:

[0087] Step 31: construct a time series analysis window and perform time series segmentation on the continuously collected coal pillar structure health index data;

[0088] Specifically, a sliding time window method is used. The window length is set according to the size of the coal pillar and the geological conditions. For example, the window length can be set to 4 hours with a sliding step of 30 minutes. Within each time window, time series data of the overall health index H(t) and the local damage index matrix D(x, y, z, t) are collected.

[0089] Step 32: Calculate the time derivative of the overall health index to obtain the global damage evolution rate.

[0090] Specifically, the global damage evolution rate Vg is calculated by taking the negative time derivative of the overall health index H(t) using the central difference method: get.

[0091] Step 33: Calculate the time derivative of the local damage index matrix to obtain the spatially distributed local damage evolution rate field.

[0092] Specifically, the local damage evolution rate field V l(x, y, z) is calculated by the partial derivative of the local damage index matrix D(x, y, z, t) with respect to time:

[0093] Step 34: Combine the maximum values ​​of the global damage evolution rate and the local damage evolution rate field to obtain the damage evolution rate of the coal pillar.

[0094] Specifically, the comprehensive damage evolution rate V is calculated as a weighted combination of the global damage evolution rate Vg and the maximum value of the local damage evolution rate field, max(Vl): V = w1·Vg + w2·max(Vl), where w1 and w2 are weight coefficients, and w1+w2=1. The weight coefficients are dynamically adjusted according to the monitoring objectives. For example, when focusing on overall stability, w1 takes a larger value; when focusing on local instability risks, w2 takes a larger value.

[0095] Furthermore, the first and second thresholds are determined based on historical data and expert knowledge bases, respectively, for use in risk classification. Specifically, the first threshold V1 and the second threshold V2 are determined through statistical analysis and expert experience. First, based on historical monitoring data, the probability distribution function of the damage evolution rate is calculated. For example, in an embodiment of the present invention, the first threshold V1 is set to the 75th percentile of the distribution, indicating a medium risk level; the second threshold V2 is set to the 95th percentile of the distribution, indicating a high risk level.

[0096] What is certain is that the damage evolution rate represents the rate of change of the degree of damage to the coal pillar structure over time, in units of % / hour, reflecting the dynamic trend of the stability state of the coal pillar. A low rate indicates that the coal pillar is in a stable state or a slow deterioration stage; a medium rate indicates that the coal pillar has entered an accelerated deterioration stage and requires enhanced monitoring; a high rate indicates that the coal pillar may be about to become unstable and failure, and preventive measures need to be taken immediately. Compared with traditional monitoring indicators such as displacement and strain, the damage evolution rate has the characteristics of earlier warning time and greater adaptability. It can identify the accumulation process of microscopic damage before macroscopic damage occurs, providing a scientific basis for coal mine safety production. In addition, by analyzing the spatiotemporal distribution characteristics of the damage evolution rate, the damage extension path and instability mechanism inside the coal pillar can be identified, thereby providing a basis for support design and mining plan optimization.

[0097] Furthermore, the method of using an adaptive sampling strategy to adjust the sampling frequency of the optical fiber sensor network and reduce the data acquisition density of the distributed optical fiber sensor network includes the following steps:

[0098] Step 41: Dynamically adjust the boundaries of each partition area based on the coal pillar structure health index;

[0099] Specifically, during the implementation of the present invention, when the overall health index value of the coal pillar structure is higher than 0.85, it indicates that the coal pillar is healthy as a whole, and the core area accounts for 20% of the total monitoring area, the transition area accounts for 30%, and the edge area accounts for 50%; when the overall health index of the coal pillar structure is between 0.7 and 0.85, it indicates that the coal pillar is in sub-health, with the core area accounting for 30%, the transition area accounting for 40%, and the edge area accounting for 30%; when the overall health index of the coal pillar structure is lower than 0.7, it indicates that the coal pillar is in an unhealthy state, with the core area accounting for 40%, the transition area accounting for 40%, and the edge area accounting for 20%. Through this dynamic boundary adjustment mechanism, the system can adaptively allocate monitoring resources according to the overall state of the coal pillar, ensuring that key areas receive sufficient monitoring density.

[0100] Step 42: Apply a frequency control algorithm based on the damage evolution rate to each partitioned area to calculate the appropriate sampling frequency;

[0101] Specifically, the frequency control algorithm uses a nonlinear mapping function to convert the damage evolution rate into a sampling frequency adjustment coefficient. For example, for the core area, the sampling frequency adjustment coefficient K_core=2 -2 ×e -5×v_damage , where V_damage is the damage evolution rate; for the transition zone, the sampling frequency adjustment coefficient K_transition = 2 -3 ×e -4×v_damage ; For the edge area, the sampling frequency adjustment coefficient K_edge = 2 -4×e -3×v_damage . It can be understood that when the damage evolution rate is close to zero, the sampling frequency of each area will be significantly reduced, the core area will be reduced to 40% of the original frequency, the transition area will be reduced to 30% of the original frequency, and the edge area will be reduced to 20% of the original frequency; as the damage evolution rate increases, the sampling frequency adjustment coefficient gradually approaches 1, ensuring that the sampling density can be increased in time during the accelerated stage of damage development. Ultimately, the actual sampling frequency of each area is obtained by multiplying the original frequency by the corresponding adjustment coefficient, that is, f_actual = f_base × K, f_base is the original frequency set at the beginning of monitoring;

[0102] Step 43: Based on the appropriate sampling frequency, dynamically adjust the pulse laser emission frequency of the distributed optical fiber sensing network;

[0103] Specifically, the dynamic adjustment of the pulsed laser emission frequency is achieved through a programmable laser control unit. The control unit receives instructions from the frequency control algorithm and adjusts the emission frequencies of the three wavelength lasers used for stress field, temperature field and electromagnetic field monitoring.

[0104] In this embodiment, for stress field monitoring, the baseline sampling frequency is 10 Hz, with an adjustable range of 2 Hz to 15 Hz; for temperature field monitoring, the baseline sampling frequency is 5 Hz, with an adjustable range of 1 Hz to 8 Hz; and for electromagnetic field monitoring, the baseline sampling frequency is 3 Hz, with an adjustable range of 0.5 Hz to 5 Hz. The laser control unit uses phase-locked loop technology to ensure stability during frequency adjustment, with a frequency switching time of less than 50 milliseconds, ensuring continuous data acquisition. Furthermore, the laser power automatically adjusts with changes in the sampling frequency, reducing power output in low-frequency sampling mode to extend the laser's lifespan and reduce energy consumption.

[0105] Step 44: Implement dynamic optimization of the sampling point density of the coal pillar, and adjust the distribution density of the sampling points of the coal pillar according to the gradient distribution of the local damage evolution rate field of the coal pillar structure;

[0106] Specifically, the dynamic optimization of the sampling point density of the coal pillar is implemented based on an adaptive grid refinement algorithm. The adaptive grid refinement algorithm first calculates the spatial gradient field of the local damage evolution rate field in each partitioned area of ​​the coal pillar structure, and identifies the partitioned areas where the health index changes dramatically. In the partitioned areas where the gradient value is higher than the threshold Tg, generally set to 0.15 / meter by default, the original density sampling point distribution is maintained; in the partitioned areas where the gradient value is between 0.05 / meter and 0.15 / meter, the sampling point spacing is increased to 1.5 times the original; in the partitioned areas where the gradient value is lower than 0.05 / meter, the sampling point spacing is increased to 2 times the original. It can be understood that through the above-mentioned gradient-based sampling point density adjustment, it is possible to significantly reduce the data collection density in stable areas while ensuring the monitoring accuracy of key areas. The adaptive grid refinement algorithm re-evaluates the gradient distribution of the coal pillar structure health index at preset time periods and updates the sampling point distribution strategy accordingly.

[0107] Furthermore, the sampling frequency of the distributed optical fiber sensor network is increased.

[0108] The sampling frequencies of different partition areas are adjusted according to actual needs as follows:

[0109] In this embodiment, for the core area, the emission frequency of the narrow linewidth laser pulse is increased from 10Hz to 20Hz, the pulse width is kept unchanged at 10ns, and the peak power is increased by 20%, ensuring that a sufficient signal-to-noise ratio is maintained under high-frequency sampling conditions. The sampling frequency of the transition zone is increased from 5Hz to 10Hz, and the edge zone is increased from 1Hz to 5Hz, forming a gradient sampling strategy that decreases from the inside to the outside. The increase in sampling frequency is achieved by increasing the trigger frequency of the laser pulse, and at the same time, the data processing rate of the signal demodulation and analysis unit is increased accordingly, and a parallel computing architecture is adopted to ensure real-time processing capabilities. In the high-frequency sampling mode, the number of data points that can be obtained per second is 2-5 times that of the normal mode, which significantly improves the time resolution of the coal pillar damage evolution process.

[0110] The method of using a dynamic weight distribution algorithm to weightedly fuse the stress field data and temperature field data in the multi-physics field coupling data, constructing a trend prediction model, and generating an evolution prediction of coal pillar damage includes the following steps:

[0111] Step 51: construct a dynamic weight distribution algorithm to adaptively adjust the weight coefficients of stress field data and temperature field data according to the coal pillar damage evolution rate and the time-varying characteristics of the physical field data;

[0112] Specifically, the dynamic weight allocation algorithm is designed based on information entropy theory and Bayesian reasoning framework. First, the information entropy of stress field data and temperature field data is calculated to quantify the uncertainty and information content of each physical field data. The formula for calculating the information entropy of the stress field is Hs = -∑ iP(σi)log(P(σi)), where P(σi) is the probability distribution of stress value σi; the temperature field information entropy calculation formula is Ht=-∑ i P(Ti)log(P(Ti)), where P(Ti) is the probability distribution of temperature Ti. The higher the information entropy, the greater the uncertainty contained in the data and the richer the information content.

[0113] Subsequently, the mutual information between the stress and temperature field data and the damage evolution rate is calculated to quantify the contribution of each physical field data to damage prediction. The formula for calculating the stress field mutual information is I(S;D) = H(D) - H(D|S), where H(D) is the information entropy of the damage evolution rate and H(D|S) is the conditional entropy of the damage evolution rate under known stress field data. The formula for calculating the temperature field mutual information is I(T;D) = H(D) - H(D|T). A higher mutual information indicates a greater contribution of the physical field data to damage prediction.

[0114] Based on information entropy and information quantity, the dynamic weight allocation algorithm uses the Bayesian optimization method to calculate the weight coefficients of stress field and temperature field data. The calculation formula of stress field weight coefficient is: The calculation formula of temperature field weight coefficient is: The weight coefficients are dynamically updated over time, for example, with an update cycle of 10 seconds, to ensure that the weight assignments promptly reflect changes in the physical field data characteristics. In the early stages of coal pillar damage, the stress field data weight is typically high, for example, 0.7-0.8. As damage progresses, the temperature field data weight gradually increases, for example, to 0.4-0.5, reflecting the early warning value of temperature changes during the accelerated damage phase.

[0115] Step 52: Based on the dynamically assigned weight coefficients, perform weighted fusion on the stress field data and the temperature field data, construct a multivariate time series prediction model, and generate an evolution prediction of the coal pillar damage;

[0116] Specifically, the weighted fusion process employs a tensor decomposition and reconstruction method. First, the stress and temperature field data are represented as third-order tensors, where the three dimensions correspond to spatial coordinates, time series, and physical quantities, respectively. The stress field tensor is denoted as Ts, and the temperature field tensor is denoted as Tt. The weighted fusion tensor calculation formula is Tf = Ws × Ts + Wt × Tt.

[0117] The fused tensor Tf is converted to a low-dimensional representation through Tucker decomposition to extract the core structural features of the data. Tucker decomposition decomposes Tf into a core tensor G and three factor matrices U(1), U(2), and U(3). The decomposition formula is Tf ≈ G×1U(1)×2U(2)×3U(3), where ×1, ×2, and ×3 represent the 1-module tensor product, 2-module tensor product, and 3-module tensor product, respectively. The core tensor G captures the main variation patterns of the data, and the factor matrices represent the basis vectors of each dimension.

[0118] Based on the decomposition results, a long short-term memory network prediction model is constructed. The input of the multivariate long short-term memory network prediction model is the historical fusion tensor Tf in the sliding window, and the window length is 60 seconds, corresponding to 600-1200 data points under high-frequency sampling. The long short-term memory network prediction model contains two hidden layers, each with 128 neurons, and a regularization strategy with a dropout rate of 0.3 is used to prevent overfitting. The output of the long short-term memory network prediction model is the prediction of the evolution of coal pillar damage in the next 30 seconds, including the degree of damage and spatial distribution. It can be understood that by predicting the evolution of coal pillar damage in the next 30 seconds, the expansion trend of the risk area and the degree of damage can be intuitively displayed, so as to monitor the development trend of the health status of the coal pillar in real time and prevent further deterioration;

[0119] Furthermore, the distributed optical fiber sensing network is switched to a high-frequency sampling mode, and a nonlinear coupling analytical algorithm is used to collaboratively analyze the multi-physics field coupling data to generate an assessment result of the evolution acceleration of coal pillar damage and the instability risk, including the following steps:

[0120] Step 61: Raise the sampling frequency of the distributed optical fiber sensor network to a preset high frequency state using a partition-differentiated frequency-raising strategy;

[0121] Specifically, in this embodiment, the partition-differentiated frequency increase strategy increases the sampling frequency of the core area to 30 Hz, the transition area to 15 Hz, and the edge area to 5 Hz.

[0122] In actual implementation, the frequency increase is done in a step-by-step manner, with each increase never exceeding 50% of the current frequency to ensure system stability. In high-frequency sampling mode, the laser's pulse repetition frequency is increased, the pulse width is shortened to 100ns, and the spatial resolution is enhanced to 0.1 meters.

[0123] In the high-frequency sampling mode, the power consumption management unit of the distributed optical fiber sensor network automatically adjusts the power allocation strategy to allocate 80% of the system power to the core area sensors to ensure high-precision monitoring of key areas.

[0124] Step 62: Construct a nonlinear coupling analytical model consisting of three components: an inter-field mapping layer, a nonlinear coupling layer, and a damage evolution prediction layer, to perform a collaborative analysis of the multi-physics field coupling data collected during the high-frequency sampling process;

[0125] Specifically, the nonlinear coupling analytical model is constructed based on tensor decomposition and multi-field coupling theory;

[0126] The inter-field mapping layer maps the stress field tensor Tσ, the temperature field tensor TT, and the electromagnetic field tensor TE to a unified feature space. The mapping function uses a nonlinear projection operator and is expressed as:

[0127] F(Tσ,TT,TE)=Wσ·φσ(Tσ)+WT·φT(TT)+WE·φE(TE)+b;

[0128] Here, φσ, φT, and φE are nonlinear feature extraction functions for each physical field, implemented using a deep neural network; Wσ, WT, and WE are the corresponding weight matrices, and b is the bias vector. The weight matrices are dynamically adjusted using a physical constraint optimization algorithm to ensure that the mapping results conform to the physical laws of thermal, mechanical, and electric multi-field coupling.

[0129] The nonlinear coupling layer uses a tensor network structure to capture the high-order nonlinear interactions between multiple physical fields. The core algorithm is the improved Tucker decomposition, which expresses the coupling relationship between the three physical fields as: H(t) = G × 1Uσ × 2UT × 3UE;

[0130] Among them, G is the core tensor, which represents the intensity distribution of inter-field coupling; Uσ, UT, and UE are the modal matrices of each physical field, which represent the interaction mode within the field.

[0131] The damage evolution prediction layer constructs a state-space model of coal pillar damage evolution based on nonlinear dynamics theory. It outputs a time series of the coupling relationship between three physical fields and its first-order derivative, dH(t) / dt, expressed in a piecewise nonlinear form: dH(t) / dt = f(H(t), Tσ, TT, TE, t). Here, f(·) is the nonlinear state transfer function, representing the automatic switching of evolution modes at different damage stages to capture the characteristics of accelerated damage development and critical instability.

[0132] Step 63: Calculate the evolution acceleration of coal pillar damage based on the analytical results of the nonlinear coupling analytical model;

[0133] Specifically, the evolution acceleration of coal pillar damage is calculated based on the coupled relationship H(t) and its first-order derivative dH(t) / dt between the three physical fields of the coal pillar structure, as output by the damage evolution prediction layer. The evolution acceleration a(t) is defined as the second-order derivative of the coal pillar structural health index and is calculated using numerical differentiation methods.

[0134] It's understandable that the coal pillar damage evolution acceleration, as a second-order derivative, can capture changing trends in the coal pillar's structural state earlier than the first-order derivative, the damage evolution rate. When a coal pillar is stable, the damage evolution rate may remain low, making it difficult to detect potential risks in a timely manner. However, the evolution acceleration can, through changes in acceleration, indicate impending instability within the coal pillar's internal structure before the damage rate increases significantly. Therefore, when a coal pillar is already at risk of instability, it's even more important to capture these trends early, enabling more timely warnings and responses.

[0135] like Figure 2 As shown in FIG, a dynamic monitoring system for coal pillar damage based on multi-physics field coupling and distributed optical fiber sensing includes a multi-physics field data collection module, a health index generation module, an evolution level judgment module, and a dynamic monitoring module; wherein each module is electrically connected;

[0136] The multi-physics field data collection module collects the multi-physics field coupling data of the coal pillar in real time through a distributed optical fiber sensing network, and sends the multi-physics field coupling data to the health index generation module and the dynamic monitoring module;

[0137] A health index generation module pre-trains a time series multi-scale mapping model, obtains a coal pillar structure health index based on the multi-physics field coupling data and the time series multi-scale mapping model, and sends the coal pillar structure health index to the evolution level judgment module;

[0138] an evolution level judgment module, which calculates the damage evolution rate of the coal pillar according to the coal pillar structural health index and sends the damage evolution rate to the dynamic monitoring module;

[0139] A dynamic monitoring module determines that if the damage evolution rate is lower than a preset first threshold, then based on the multi-physical field coupling data and the damage evolution rate, an adaptive sampling dynamic strategy is used to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network; if the damage evolution rate is not lower than the preset first threshold but lower than the preset second threshold, then based on the multi-physical field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight distribution algorithm is used to weightedly fuse the stress field data and temperature field data in the multi-physical field coupling data to construct a trend prediction model and generate an evolution prediction of coal pillar damage; if the damage evolution rate is not lower than the preset second threshold, the distributed optical fiber sensing network is switched to a high-frequency sampling mode, and a nonlinear coupling analytical algorithm is used to collaboratively analyze the multi-physical field coupling data to generate an evolution acceleration of coal pillar damage.

[0140] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for dynamic monitoring of coal pillar damage based on multi-physics field coupling and distributed optical fiber sensing, characterized in that: The following steps are involved: Step 1: Real-time acquisition of multi-physics coupling data of coal pillars through a distributed optical fiber sensing network; Step 2: pre-training a time series multi-scale mapping model, and obtaining a coal pillar structural health index based on the multi-physics field coupling data and the time series multi-scale mapping model; Step 3: Calculate the damage evolution rate of the coal pillar based on the coal pillar structural health index. If the damage evolution rate is lower than a preset first threshold, proceed to step 4. If the damage evolution rate is not lower than the preset first threshold but lower than a preset second threshold, proceed to step 5. If the damage evolution rate is not lower than the preset second threshold, proceed to step 6. Step 4: Based on the multi-physics field coupling data and the damage evolution rate, an adaptive sampling dynamic strategy is used to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network; Step 5: Based on the multi-physics field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight allocation algorithm is used to weightedly fuse the stress field data and temperature field data in the multi-physics field coupling data to construct a trend prediction model and generate an evolution prediction of coal pillar damage; Step 6: Switch the distributed optical fiber sensing network to a high-frequency sampling mode, and use a nonlinear coupling analytical algorithm to collaboratively analyze the multi-physical field coupling data to generate the evolution acceleration of coal pillar damage.

2. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 1 is characterized in that: The real-time acquisition of multi-physical field coupling data of coal pillars through a distributed optical fiber sensing network comprises the following steps: The distributed optical fiber sensing network is partitioned and arranged according to the structural characteristics of the coal pillars. The distributed optical fiber sensing network then collects stress field distribution tensors, temperature field distribution tensors, and electromagnetic field distribution tensors from different partitioned areas at a predetermined acquisition frequency to form the multi-physical field coupling data.

3. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 2, characterized in that: The obtaining of the coal pillar structure health index comprises the following steps: Step 21: extracting time series features from the multi-physics field coupling data using a sliding window technique to construct a multi-physics field time series feature tensor; Step 22: Construct a time series multi-scale mapping model based on deep learning, wherein the time series multi-scale mapping model includes three units: a time series encoder, a spatial feature extractor, and a multi-scale fusion network; Step 23: pre-collect historically acquired multi-physics field time series feature tensors and corresponding coal pillar structure state labels, and pre-train the time series multi-scale mapping model; Step 24: Input the multi-physics field time series feature tensor into the pre-trained time series multi-scale mapping model to generate a coal pillar structural health index; the coal pillar structural health index is output in the form of structured data, which at least includes the following health index information: an overall health index, which indicates the overall stability state of the coal pillar; and a local damage index matrix, which indicates the degree of damage in each partition area.

4. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 3 is characterized in that: The multi-physics field time series feature extraction adopts sliding window technology to extract time dimension features of stress field distribution tensor, temperature field distribution tensor and electromagnetic field distribution tensor; then, the corresponding statistical features and corresponding frequency domain features are calculated for stress, temperature and electromagnetic field intensity within each sliding window; for stress field data, the stress gradient and principal stress direction are additionally calculated; for temperature field data, the temperature change rate and spatial temperature gradient are additionally calculated; for electromagnetic field data, the electromagnetic field intensity change rate and spatial distribution features are additionally calculated; The multi-physical field time series feature tensor finally obtained is organized in a four-dimensional tensor structure with the dimensions [T, S, F, C], where T represents the number of time windows, S represents the number of spatial monitoring points, F represents the total amount of extracted statistical features and frequency domain features, and C represents the number of physical field channels.

5. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 4 is characterized in that: The temporal encoder adopts a bidirectional long short-term memory network structure, including three layers of Bi LSTM units, each layer containing 128 hidden neurons; the temporal encoder receives the multi-physical field temporal feature tensor as input to capture the long-term dependency and short-term fluctuation characteristics of the parameters of different physical fields over time; The forward propagation of Bi LSTM captures the impact of the current state on the future state, and the back propagation captures the impact of the historical state on the current state, forming a complete temporal feature representation; The spatial feature extractor uses a graph convolutional network structure to construct pre-set coal pillar monitoring points into a spatial graph structure. Each node of the spatial graph structure corresponds to a coal pillar monitoring point, and the weight of the edges of the spatial graph structure is calculated based on the spatial distance between each coal pillar monitoring point and the correlation between physical field parameters. The graph convolutional network structure includes three layers of graph convolution, with the number of output channels of each layer being 64, 128, and 256, respectively. The spatial feature extractor thus captures the spatial correlation and stress transfer characteristics between different regions of the coal pillar, thereby achieving a mapping of microscopic monitoring data to macroscopic structural features. The multi-scale fusion network adopts a multi-scale feature fusion architecture with an attention mechanism, which includes three parallel convolution branches with convolution kernel sizes of 3×3, 5×5, and 7×7, corresponding to different receptive fields; each branch contains two convolution layers, and the number of output channels is 64; Multi-scale features are adaptively weighted and fused through the channel attention mechanism to generate a comprehensive feature representation; the output of the fusion network passes through two fully connected layers with a hidden layer size of 256, and the final output layer is the coal pillar structure health index.

6. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 5, characterized in that: Calculating the damage evolution rate of the coal pillar according to the coal pillar structural health index comprises the following steps: Step 31: Use the sliding time window method to construct a time series analysis window and perform time series segmentation on the continuously collected coal pillar structure health index data; Step 32: Calculate the time derivative of the overall health index to obtain the global damage evolution rate; Step 33: Calculate the time derivative of the local damage index matrix to obtain the spatially distributed local damage evolution rate field; Step 34: Combine the maximum values ​​of the global damage evolution rate and the local damage evolution rate field to obtain the damage evolution rate of the coal pillar.

7. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 6, characterized in that: The method of using an adaptive sampling strategy to adjust the sampling frequency of the optical fiber sensor network and reduce the data acquisition density of the distributed optical fiber sensor network includes the following steps: Step 41: Dynamically adjust the boundaries of each partition area based on the coal pillar structure health index; Step 42: Apply a frequency control algorithm based on the damage evolution rate to each partitioned area using a nonlinear mapping function to calculate an appropriate sampling frequency; Step 43: Based on the appropriate sampling frequency, dynamically adjust the pulse laser emission frequency of the distributed optical fiber sensing network; Step 44: Dynamically optimize the density of sampling points of the coal pillar based on the adaptive grid refinement algorithm, and adjust the distribution density of the sampling points of the coal pillar according to the gradient distribution of the local damage evolution rate field of the coal pillar structure.

8. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 7, characterized in that: The method of using a dynamic weight distribution algorithm to weightedly fuse the stress field data and temperature field data in the multi-physics field coupling data, constructing a trend prediction model, and generating an evolution prediction of coal pillar damage includes the following steps: Step 51: construct a dynamic weight distribution algorithm to adaptively adjust the weight coefficients of stress field data and temperature field data according to the coal pillar damage evolution rate and the time-varying characteristics of the physical field data; Step 52: Based on the dynamically assigned weight coefficients, the stress field data and the temperature field data are weightedly fused using the tensor decomposition and reconstruction method to construct a multivariate time series prediction model to generate an evolution prediction of coal pillar damage.

9. The method for dynamic monitoring of coal pillar damage based on multi-physical field coupling and distributed optical fiber sensing according to claim 8, characterized in that: Switching the distributed optical fiber sensing network to a high-frequency sampling mode and using a nonlinear coupling analytical algorithm to collaboratively analyze the multi-physics field coupling data to generate an evaluation result of the evolution acceleration of coal pillar damage and the instability risk includes the following steps: Step 61: Raise the sampling frequency of the distributed optical fiber sensor network to a preset high frequency state using a partition-differentiated frequency-raising strategy; Step 62: Construct a nonlinear coupling analytical model consisting of three components: an inter-field mapping layer, a nonlinear coupling layer, and a damage evolution prediction layer, to perform a collaborative analysis of the multi-physics field coupling data collected during the high-frequency sampling process; Step 63: Based on the analytical results of the nonlinear coupling analytical model, calculate the evolution acceleration of the coal pillar damage.

10. A coal pillar damage dynamic monitoring system based on multi-physics field coupling and distributed fiber optic sensing, which is used to implement the coal pillar damage dynamic monitoring method based on multi-physics field coupling and distributed fiber optic sensing as described in any one of claims 1 to 9, comprising a multi-physics field data collection module, a health index generation module, an evolution level judgment module, and a dynamic monitoring module; wherein, Each module is connected electrically; The multi-physics field data collection module collects the multi-physics field coupling data of the coal pillar in real time through a distributed optical fiber sensing network, and sends the multi-physics field coupling data to the health index generation module and the dynamic monitoring module; A health index generation module pre-trains a time series multi-scale mapping model, obtains a coal pillar structure health index based on the multi-physics field coupling data and the time series multi-scale mapping model, and sends the coal pillar structure health index to the evolution level judgment module; an evolution level judgment module, which calculates the damage evolution rate of the coal pillar according to the coal pillar structural health index and sends the damage evolution rate to the dynamic monitoring module; a dynamic monitoring module, which determines that if the damage evolution rate is lower than a preset first threshold, then, based on the multi-physics field coupling data and the damage evolution rate, uses an adaptive sampling dynamic strategy to adjust the sampling frequency of the distributed optical fiber sensing network to reduce the data acquisition density of the distributed optical fiber sensing network; If the damage evolution rate is not lower than a preset first threshold but lower than a preset second threshold, then based on the multi-physics field coupling data and the damage evolution rate, the sampling frequency of the distributed optical fiber sensing network is increased, and a dynamic weight allocation algorithm is used to perform weighted fusion on the stress field data and temperature field data in the multi-physics field coupling data, to construct a trend prediction model and generate an evolution prediction of the coal pillar damage; If the damage evolution rate is not lower than a preset second threshold, the distributed optical fiber sensing network is switched to a high-frequency sampling mode, and a nonlinear coupling analytical algorithm is used to collaboratively analyze the multi-physical field coupling data to generate an evolution acceleration of coal pillar damage.

Images