A mine roadway deformation data real-time analysis system based on edge computing

Abstract

The application discloses a kind of based on edge computing's mine roadway deformation data real-time analysis system, it is related to intelligent sensing and data analysis fusion technical field, including data acquisition module, roadway deformation initial data is collected by sensor, adopts finite element analysis method to simulate deformation field evolution process, obtain non-uniform field distribution characteristics and potential blind area redundant position, feature extraction module, from non-uniform field distribution characteristics extraction key stress distribution parameter, if parameter exceeds preset threshold, then mark as risk concentration area, determine monitoring position candidate set, signal quality evaluation module, for monitoring position candidate set, obtain the signal-to-noise ratio and integrity data quality index of sensor signal, priority grouping is carried out to monitoring point by clustering algorithm, obtain dynamic priority evaluation result;The based on edge computing's mine roadway deformation data real-time analysis system, realizes the intelligent, real-time and high reliability of mine roadway deformation monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent sensing and data analysis fusion, and particularly relates to a mine roadway deformation data real-time analysis system based on edge computing. BACKGROUND

[0002] The field of underground mine roadway deformation monitoring is directly related to production safety and efficient resource exploitation, and the key is to capture the real changes of the entire space through limited equipment to avoid local instability evolving into a major accident.

[0003] Current methods rely on pre-set experience or uniform point distribution, which makes it difficult to accurately reflect the deformation concentration area in complex geological environments, and resource allocation often coexists with the contradiction of redundancy and blind area. The deformation field distribution characteristics are jointly restricted by the roadway geometry, geological conditions and stress distribution law, and these factors are interwoven, making it difficult for a single location to represent the overall trend. The uneven analysis of geological conditions is requesting the solution content provided by the user to be reviewed, and the focus is on the optimization selection technology of roadway deformation monitoring. It will amplify the local differences in stress distribution, and then cause the deformation field to be highly non-uniformly expressed in space, and if the monitoring point is not aligned with these characteristic areas, the key deformation signal will be missed. For example, in a curved roadway, the upper rock layer is soft and the lower rock layer is hard, and the stress is concentrated on the inside of the turn, if the monitoring point is only arranged in the straight section, the inside deformation will suddenly accelerate when it accumulates to a certain extent, but it cannot be warned in advance due to the lack of direct observation, and eventually leads to the failure of the support structure. The data quality and coverage range of the existing monitoring points further exacerbate this problem, and data noise or gaps will weaken the recognition accuracy of the deformation field characteristics, forcing the network layout to fall into a cycle of frequent adjustment but difficult to converge. SUMMARY

[0004] The present application aims to provide a mine roadway deformation data real-time analysis system based on edge computing, which solves the problems existing in the prior art.

[0005] To achieve the above object, the present application provides the following technical scheme: a mine roadway deformation data real-time analysis system based on edge computing, comprising a data acquisition module, which acquires initial roadway deformation data through sensors, uses a finite element analysis method to simulate the deformation field evolution process, and obtains non-uniform field distribution characteristics and potential blind area redundant positions; a feature extraction module, which extracts key stress distribution parameters from the non-uniform field distribution characteristics, and if the parameters exceed a preset threshold, marks the area as a risk concentration area and determines a monitoring position candidate set; a signal quality evaluation module, which, for the monitoring position candidate set, obtains the signal-to-noise ratio and data quality indicators of the sensor signal, and groups the monitoring points in priority through a clustering algorithm to obtain a dynamic priority evaluation result; a layout optimization module, which, according to the dynamic priority evaluation result and the blind area redundant position, uses a genetic algorithm with coverage rate and resource consumption as the fitness function, generates a full-coverage network preliminary layout configuration through multi-point crossover and random mutation; an iterative adjustment module, which, through the preliminary layout configuration, simulates the coverage effect of the full-coverage network under the non-uniform field, and if the coverage rate is lower than a preset threshold, adjusts the monitoring point position through iteration to obtain an updated layout configuration; and a trend prediction module, which extracts deformation monitoring data flow from the updated layout configuration and uses a time series analysis method to predict the roadway deformation trend and determine the potential risk evolution path.

[0006] As can be seen from the above technical scheme, the present application has the following beneficial effects:

[0007] The present application can effectively overcome the problems of many monitoring blind areas, low data quality and slow point layout adjustment in the existing roadway deformation monitoring method, and realize dynamic, accurate and efficient monitoring of roadway deformation under complex geological conditions. By acquiring initial roadway deformation data through sensors and combining the finite element analysis method to simulate the deformation field evolution process, the non-uniform field distribution characteristics and potential blind area positions can be accurately obtained, thereby realizing the true restoration of the overall deformation trend. By extracting key stress parameters from the non-uniform field distribution and identifying high-risk concentration areas, the monitoring point layout is more targeted, significantly improving the monitoring coverage rate and risk identification accuracy. Through the clustering algorithm, the signal-to-noise ratio and data integrity of the monitoring signal are quality graded, effectively improving the data reliability and calculation efficiency. Based on the layout optimization mechanism of the genetic algorithm, the system can achieve an optimal balance between coverage rate and resource consumption, avoiding the coexistence of redundant point layout and monitoring blind areas. Combined with the time series analysis method, the deformation trend prediction and risk path deduction are realized, enabling the potential instability area to be warned in advance. At the same time, the system has the dynamic feedback and adaptive adjustment capability supported by the edge computing architecture, which can optimize the monitoring point and sensor resource allocation in real time according to the changes in geological conditions, and construct a closed-loop iterative monitoring system. In summary, through the synergistic effect of model calculation, data clustering, intelligent optimization and dynamic feedback, the present application realizes the intelligent, real-time and high-reliability of mine roadway deformation monitoring, and provides important technical support for the safe operation and efficient resource exploitation of underground mines. Attached Figure Description

[0008] Figure 1 This is a connection diagram of the real-time analysis system for mine roadway deformation data based on edge computing, as described in this invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] like Figure 1 As shown, this invention provides a technical solution: a real-time analysis system for mine roadway deformation data based on edge computing, comprising a data acquisition module that collects initial roadway deformation data through sensors, simulates the deformation field evolution process using finite element analysis, and obtains non-uniform field distribution characteristics and potential blind zone redundancy locations; a feature extraction module that extracts key stress distribution parameters from the non-uniform field distribution characteristics, and marks areas of concentrated risk as risk areas if the parameters exceed a preset threshold, thus determining a candidate set of monitoring locations; a signal quality assessment module that, for the candidate set of monitoring locations, obtains the signal-to-noise ratio and integrity data quality indicators of sensor signals, prioritizes monitoring points using a clustering algorithm, and obtains dynamic priority assessment results; and a layout optimization module that, based on the dynamic priority assessment results and blind zone redundancy locations, uses a genetic algorithm with coverage and resource consumption as fitness functions, and generates a layout through multi-point crossover and random mutation. The system comprises the following modules: a preliminary layout configuration for the full-coverage network; an iterative adjustment module, which simulates the coverage effect of the full-coverage network under non-uniform fields based on the preliminary layout configuration; a trend prediction module, which extracts deformation monitoring data streams from the updated layout configuration and uses time series analysis to predict the deformation trend of the tunnels and determine the evolution path of potential risks; a dynamic resource allocation module, which dynamically adjusts the allocation ratio of sensor resources based on the evolution path of potential risks and the real-time feedback mechanism; and a feedback closed-loop module, which integrates the signal-to-noise ratio and integrity data quality feedback loop through the optimized monitoring location set and uses a genetic algorithm with the minimum coverage threshold as a constraint to adjust the resource allocation ratio and obtain a stable operating configuration for the full-coverage monitoring network.

[0011] The system disperses data collection, feature extraction and analysis decision tasks to the roadway site node end through the edge computing architecture, thereby reducing the load of the central server. The data collection module collects original data such as stress and displacement of roadway deformation using a distributed sensor array, and establishes a deformation model through the finite element method to simulate the non-uniform field. The feature extraction module extracts key feature parameters through stress field gradient calculation and principal stress direction analysis, and determines the risk concentration area by threshold. The signal quality evaluation module performs clustering analysis based on signal-to-noise ratio and data integrity indicators to identify the optimal monitoring point and establish a dynamic priority. The layout optimization module takes genetic algorithm as the core, combined with blind area and redundant information to perform multi-point crossover and mutation operations, and constantly evolves to generate the optimal monitoring network layout. The iterative adjustment module verifies the coverage rate and redundancy of the full coverage network through simulation to ensure the efficiency and robustness of the layout. The trend prediction module uses time series models such as ARIMA or LSTM to predict future deformation trends and achieve early warning of potential risks. The resource dynamic allocation module adjusts sensor energy and communication resources according to real-time monitoring results to ensure priority coverage in high-risk areas. The feedback loop module establishes a dynamic optimization cycle to make resource allocation and network topology tend to be stable in multiple rounds of genetic evolution, achieving adaptive optimization of the monitoring network.

[0012] The system realizes real-time analysis and adaptive optimization of mine roadway deformation data by introducing edge computing and intelligent algorithms. Compared with traditional centralized processing systems, this scheme significantly reduces data transmission delay, improves monitoring response speed and system stability. The cooperative mechanism of finite element simulation and clustering evaluation can effectively identify potential risk areas and dynamically adjust the monitoring layout, improving coverage rate and resource utilization. The introduction of genetic algorithm enables the layout optimization and resource allocation process to have evolutionary learning ability, maintaining monitoring accuracy in complex geological environments. The time series prediction model further enhances the forward-looking analysis capability of risk trends, realizing closed-loop control of early warning and resource linkage. The overall system is superior to existing technologies in reliability, flexibility and intelligence, and has significant engineering application value and promotion potential.

[0013] The data acquisition module collects initial data of roadway deformation through sensors, simulates the evolution process of the deformation field by using the finite element analysis method, and obtains the non-uniform field distribution characteristics and potential blind area redundant positions. The initial data of roadway deformation is collected through sensors, the deformation parameter value is obtained from the initial data, the grid element is divided and the boundary condition is applied by using the finite element analysis method, the simulation grid is established, and the initial model of the deformation field is obtained. For the initial model of the deformation field, the evolution process parameters are input, the stress distribution is simulated by iterative calculation, the non-uniform field distribution evolution is simulated, and the distribution characteristic vector is determined. The potential blind area coordinates are extracted from the distribution characteristic vector, if the coordinates exceed the preset threshold, the redundant position points are marked, and the blind area redundant set is obtained. According to the blind area redundant set, the signal strength is integrated by using the weighted average method, the position layout scheme is optimized, the adjusted distribution characteristics are obtained, the evolution trend is analyzed by using the adjusted distribution characteristics, the time series method is used to track the changes, and the non-uniform field distribution characteristics and potential blind area redundant positions are obtained.

[0014] In the embodiment, the system is arranged with several sensors in the field according to the cross section, strike and inflection point position of the roadway, the sensor types include displacement sensor, strain sensor and pressure sensor, the field end acquires the initial data of roadway deformation according to fixed sampling period, the sampling period is suggested to be taken as integer minute level or second level, the default value is to collect 10 times per second, each collection records the time stamp, position number and corresponding deformation reading, after collecting a complete set, immediately execute data quality inspection at the edge end, the quality inspection steps are to eliminate obvious distorted readings, fill in the missing readings, remove duplicate readings of multiple sensors at the same time point, and supplement the data of missing time points in the linear interpolation manner of adjacent time points, so that each monitoring point has continuous records on the entire time axis; then obtain the deformation parameter values from the initial data, the deformation parameter values are defined as the current value, maximum value, minimum value, average value, change amplitude, change rate and change acceleration of displacement reading, strain reading and pressure reading, the change rate is obtained by difference between adjacent time points divided by time interval, the change acceleration is obtained by difference between adjacent rates divided by time interval, these calculations are executed point by point in time sequence at the edge end and saved as sequence; after completing parameter acquisition, the finite element analysis method is used to divide grid elements and apply boundary conditions, the grid division steps are firstly to import the roadway space contour and surrounding rock layering information, to disperse the roadway space into regular hexahedral or tetrahedral elements, the element length is selected according to the roadway scale and monitoring accuracy, the default is to divide the minimum characteristic size of the roadway by 20 as the initial edge length, if the edge length leads to element distortion in continuous grid check, it is automatically reduced to 0.8 times of the previous value until the check is passed; the boundary condition application steps are to apply normal constraint and tangential free condition on the outer boundary of the roadway, to apply ground stress and support reaction force on the stress side, the ground stress size is selected according to the in-situ stress range given by the mine area geological survey report, if the specific value is not given on site, the pressure average value of the historical stable period is taken as the initial value, the support reaction force is determined by looking up the table according to the corresponding relationship between the rated bearing capacity of the support member and the current deformation, after completing the boundary conditions, the simulation grid is established and the initial model of the deformation field is obtained;For the initial model of the deformation field, input the evolution parameters, including time step, material constitutive parameters and load evolution parameters. The time step is the time interval for each advance in the calculation, with a default value of 1 minute. The material constitutive parameters are determined from field material experiments or equipment nameplate data, including modulus values in the elastic range, yield point values, and residual strength values. If there are backfill or reinforcement materials, the corresponding parameters are entered. The load evolution parameters are defined as the increase and decrease amplitude of ground stress over time and the adjustment amplitude of support resistance, which are determined by the field operation plan and dynamic disturbance record table. The system iteratively calculates the stress distribution according to the evolution parameters. Each iteration advances by one time step, and the current stress, displacement, and strain updates are calculated at each grid element. After updating, convergence checks are performed on the entire grid. The convergence criterion is that the stress and displacement increments of all elements are simultaneously below their respective thresholds, with a default threshold of 1% of the initial peak value. If not met, the next iteration is continued. If met, the field distribution at that time step is output, and the next time step is entered. This process continues until the pre-set time range is completed. After completing a time range of iterations, the system forms a distribution feature vector for each grid element in the continuous time, consisting of stress extreme value, displacement extreme value, strain extreme value, stress gradient maximum value, and displacement gradient maximum value. Each item in the distribution feature vector is a specific numerical value, without symbols or letters. The system extracts potential blind area coordinates from the distribution feature vector. The criteria for identifying potential blind areas are insufficient monitoring signal coverage or significantly lower model response than the surrounding elements. The specific steps are as follows: superimpose the effective coverage radius of the sensors on the simulation grid, calculate how many sensors cover each grid element, and mark elements with coverage less than 1 as candidate blind areas. Also, calculate the stress and displacement response relative to the local neighborhood in all elements. If the ratio is less than 50% of the median in the neighborhood and the signal acquisition record shows that the missing packet ratio exceeds 10% at the same time point, it is marked as a low-response candidate blind area. Combine and remove duplicates of the two types of candidate blind areas to obtain the potential blind area coordinate set. The threshold determination method is as follows: the coverage threshold is 1, indicating that at least one sensor's effective coverage is required. The low-response threshold uses distribution statistics from the stable period in history. The median of the same element's response in the stable period is used as the reference. Responses with a ratio less than 50% to the reference are considered low-response. A fixed threshold of 10% is set for the missing packet ratio. This threshold is obtained from the continuous transmission test during communication link acceptance. If the on-site retest shows that the link quality has improved, update the threshold downward based on the 95% missing packet rate of the retest. When any of the potential blind area coordinates exceeds the above threshold, it is marked as a redundant position point. All redundant position points are aggregated into a blind area redundancy set.According to the blind area redundancy set, the system fuses multi-source sensor signals and integrates signal strength through weighted average method. The fusion process is as follows: firstly, the signal-to-noise ratio score and data integrity score of each sensor are calculated. The signal-to-noise ratio score is linearly mapped to 1 as the optimal value and 0 as the worst value after field noise calibration. The data integrity score is mapped to 0 to 1 according to the actual sampling success rate in the past week. Then the comprehensive weight is calculated as the product of the signal-to-noise ratio score and the data integrity score. The comprehensive weights of all sensors are added and each sensor's comprehensive weight is divided by the sum to obtain a weight distribution table with a weight sum of 1. Finally, the signal strength of all sensors in each grid cell is weighted and averaged according to the weight distribution table to obtain the fused signal strength at the cell level. After fusion, the system proposes layout optimization suggestions point by point according to the blind area redundancy set. Specifically, for the uncovered or low-response cells, the system prioritizes the allocation of additional sensors. For the over-dense cells, the system proposes point removal or relocation suggestions in order of decreasing fusion weight. The proposed scheme satisfies two rules: first, the number of covers for all cells after implementation of the scheme is not less than 1; second, the fused signal strength of all low-response cells after implementation of the scheme is not less than 50% of the fused strength of similar cells in the historical stable period. After executing the above rules, the adjusted distribution characteristics are obtained, which record the number of covers, the fused signal strength, the stress and displacement response extreme value and its time position of each cell. The system then tracks changes and analyzes evolution trends through time series method. The implementation process of the time series method is as follows: the stress, displacement and fused signal strength of each cell are smoothed by moving average to smooth high-frequency noise, with a default sliding window of 5 time steps. Then the change rate and acceleration of adjacent time steps are calculated. The trend is determined based on the change direction and amplitude of the past 60 time steps. If more than 45 time steps show the same change direction and the average change rate is more than twice the average rate in the historical stable period, it is determined as an upward or downward trend. The trend determination result and the adjusted distribution characteristics together generate the final output of the non-uniform field distribution characteristics and potential blind area redundancy positions. The sampling period is determined according to the monitoring real-time demand and device processing capacity. When the roadway disturbance is frequent, the sampling period is taken as seconds, and when the disturbance is slow, the sampling period is taken as minutes. The final value is determined by the maximum frequency during the trial period without packet loss. The grid cell edge length is determined according to the minimum characteristic size of the roadway geometry and the target resolution. The target is to make the cell edge length not more than one tenth of the minimum characteristic size. If the initial setting causes cell distortion, it will be automatically adjusted according to the scaling rule. The ground stress and support reaction value are adjusted to be consistent with the deformation during the trial period based on the geological survey report, support design file and historical stable data. When the three exist simultaneously, the field survey is given priority and the historical data is given secondary priority. The time step is as small as possible under the premise of ensuring that each step has observation data support, subject to the time resolution of the field data and the computing resources.The convergence threshold is 1% of the initial peak value to ensure numerical stability while avoiding computational delay caused by excessive iterations. The threshold is checked during the trial run by comparing the field distribution difference between the two adjacent steps and the measured deviation. If the deviation exceeds 5%, the threshold is lowered to 0.5%. The potential blind area coverage threshold is fixed at 1 to ensure the most basic coverage requirement. The low response threshold is based on the statistical results of the stable historical period to adapt to the differences in different geological sections. The packet loss rate threshold is initially 10% from the link acceptance test. In actual operation, the smaller of the 95th percentile packet loss rate in the last 7 days and the initial value is taken as the execution threshold. The mapping interval of signal-to-noise ratio score and data integrity score is fixed at 0 to 1. The mapping endpoints can be directly obtained through one-time noise calibration and continuous sampling statistics on site. The sliding average window and trend determination window are selected through historical data playback before going online to minimize the combination of false positive rate and false negative rate. If there is no historical data, the default values of 5 and 60 are used.

[0015] The feature extraction module extracts key stress distribution parameters from the non-uniform field distribution features. If the parameters exceed the preset threshold, the area is marked as a risk concentration area. The determination of the monitoring position candidate set includes obtaining key stress distribution parameters from non-uniform field distribution features, extracting intensity values from multi-source sensor signals through deformation data processing, integrating intensity values using a weighted average method to obtain a parameter feature vector. Threshold overrun judgment is performed on the parameter feature vector. If it exceeds the preset threshold, the risk area is marked, and the concentration area recognition result is determined. According to the concentration area recognition result, the blind area position calculation is fused. The coordinate value is extracted from the distribution features and the redundant points are judged. The position set optimization scheme is obtained, and the monitoring candidate set is obtained. The trend change tracking index is extracted from the monitoring candidate set. The multi-source signals are fused from the adjusted layout by integrating signal strength to obtain the risk concentration area monitoring distribution.

[0016] In the present embodiment, firstly, the non-uniform field distribution characteristics and the deformation intensity values of the co-located multi-source sensors are synchronously read at each grid cell or monitoring unit level and time-aligned. The time alignment is based on the system unified time, allowing matching within plus or minus one sampling period. If it exceeds, it is linearly interpolated according to the adjacent time to fill in the missing data. All cells have complete data at the same time point. Then, the key stress distribution parameters are calculated in each cell to form a parameter feature vector. The key stress distribution parameters include the current stress intensity, stress change rate, stress change acceleration, stress gradient intensity, and stress gradient change rate, a total of 5 items. The current stress intensity is directly taken from the cell stress value at the corresponding time point of the non-uniform field distribution characteristics. The stress change rate is the stress difference between the current time point and the previous time point divided by the time interval between the two points. The stress change acceleration is the difference in change rate between the current time point and the previous time point divided by the time interval between the two points. The stress gradient intensity is the stress difference between the current cell and all its adjacent cells, which is obtained by spatial distance weighted averaging. The stress gradient change rate is the gradient intensity difference between the adjacent two time points divided by the time interval. The above 5 items are kept to 3 decimal places after each calculation and written into the cell attribute table. To obtain a robust intensity integrated value, the module calculates the quality weight of the deformation intensity values of multiple sensors in the same cell and performs a weighted average. The quality weight is obtained by multiplying the signal-to-noise ratio score and the data integrity score. The signal-to-noise ratio score is determined by the ratio of effective signal to background noise of each sensor in the calibration environment, which is linearly mapped to 0 corresponding to the minimum value of the whole network and 1 corresponding to the maximum value to obtain a determined score. The data integrity score is calculated by the ratio of the number of successful samplings to the planned sampling number in the last fixed number of days and linearly mapped to 0 to 1 to obtain a determined score. The product of the two gives a quality weight between 0 and 1. The module takes the quality weight of each sensor as a coefficient to perform a weighted average of its deformation intensity values to obtain a cell-level intensity integrated value. This integrated value and the 5 key stress distribution parameters are concatenated in a fixed order to form a parameter feature vector. After the vector construction is completed, threshold overrun judgment and risk concentration area identification are performed. The threshold system is composed of parameter threshold, coverage threshold, redundancy threshold, quality threshold, and trend threshold. The parameter threshold is set to a determined value for each of the 5 key stress distribution parameters and is obtained in a partitioned calibration manner. Specifically, in a partition with the same surrounding rock type and support condition, a continuous and stable week is selected as the historical stable period. The median of each parameter in this period is calculated and multiplied by a safety factor to obtain the threshold. The initial value of the safety factor is 1.5. When the false positive rate is higher than 5% during the trial run, it is adjusted upwards by 0.1 steps to not more than 2. When the false negative rate is higher than 5%, it is adjusted upwards by 0.1, all adjustments are recorded in the parameter profile and take effect immediately; the module compares each unit's parameter feature vector with the corresponding threshold value item by item, and any item greater than or equal to the threshold value is a risk unit. Then, according to the adjacent rule of shared edges or shared faces, the module aggregates the continuous risk units into a risk cluster area in space and calculates the number of units, the outer boundary and the quality optimal center coordinates of the area. The quality optimal center coordinates are defined as the coordinates of the existing monitoring point with the maximum quality weight in the area or the geometric center coordinates of the area when there is no monitoring point. To fuse the risk identification results with the blind area information, the module reads the coverage number and the redundancy flag in each risk cluster area. The coverage number is the number of monitoring points that cover the unit at the same time. The lower limit of the coverage threshold is set to 1 to ensure minimum observability, and the upper limit is set to 3 to avoid resource stacking. The redundancy flag is true when the coverage number is greater than the upper limit and the minimum quality weight in the unit is less than 0.2. The value of 0.2 is determined during the trial operation period by the principle of minimizing false positives and false negatives at the same time and is fixed for use. The module generates a location set optimization scheme according to the blind area priority principle. For blind area units with a coverage number less than 1, a new point suggestion is generated at the center or the midpoint of the straight line connecting the nearest high-quality monitoring point. At the same time, the construction accessibility and power supply communication conditions are checked. If it is accessible, it is recorded as a new point. If it is not accessible, a move point suggestion is generated at the nearest accessible location from the blind area boundary. For units with a true redundancy flag, move or remove point suggestions are made according to the quality weight from small to large, and the coverage number is simulated before and after each suggestion to ensure that the coverage number is not less than 1 after the point is removed. All suggestions form the location set optimization scheme and are output as the monitoring candidate set. After obtaining the monitoring candidate set, the module extracts the trend change tracking indicators for each unit from the candidate set and generates the risk cluster area monitoring distribution. The trend change tracking indicators include the moving average, the change rate and the change acceleration. The window length of the moving average is fixed at 5 time steps to weaken high-frequency noise. The trend determination window length is fixed at 60 time steps to cover enough evolution periods. The trend threshold is composed of the same direction proportion threshold and the amplitude threshold. The same direction proportion threshold is fixed at 0.75, and the amplitude threshold is fixed at twice the average change rate of the historical stable period. The module counts the number of time steps consistent with the current overall change direction in each trend determination window and calculates the average change amplitude of the window. If the same direction proportion is not less than 0.75and the average variation amplitude is not lower than the above-mentioned threshold, the unit is marked as a continuous upward or downward trend, otherwise, it is marked as no significant trend, and the trend mark, the latest intensity integration value, the risk mark of whether each parameter is out of limit, and the corresponding candidate monitoring point coordinates in the position set optimization scheme are written into the risk set regional monitoring distribution as the output result of this time. In terms of parameter interpretation, the sampling period is determined by the device processing capacity and real-time requirement, and the maximum feasible frequency is selected in the trial operation period with the constraint of no packet loss, the unit size is determined by the minimum feature size of the roadway and the expected spatial resolution, and is fixed for use after the grid quality check passes, the length of the historical stable period is not less than one week and must meet the conditions of no alarm, no construction disturbance and link stability, the safety factor, the upper limit of coverage and the lower limit of quality are determined by the joint target minimization of false alarm and missed alarm in the trial operation calibration stage, and can only be adjusted according to the above-mentioned step after the on-line site review record is formed, the missing data processing threshold is set to be that the single-point missing data ratio is not higher than 10% and the continuous missing data is not more than 3 time steps, and if it exceeds, the monitoring point will be marked as fault in the current window and will be preferentially checked in the position optimization; through the above-mentioned deterministic value method and step-by-step reviewable calculation process, the feature extraction module can obtain the key stress distribution parameters from the non-uniform field distribution characteristics in each calculation period, form the parameter feature vector by using the weighted average method from the multi-source sensor signal integration intensity value, complete the risk area identification based on the explicit threshold, generate the position set optimization scheme combined with the blind area and redundancy judgment, and output the monitoring candidate set, and then extract the trend change tracking index in the fixed window and generate the risk set regional monitoring distribution accordingly.

[0017] The signal quality evaluation module obtains the signal-to-noise ratio and integrity data quality index of the sensor signal for the monitoring position candidate set, obtains the signal-to-noise ratio of the sensor signal from the monitoring position candidate set, extracts the noise interference value from the multi-source data through signal strength calculation, and obtains the signal-to-noise ratio quantitative index; fuses the integrity data quality index for the signal-to-noise ratio quantitative index, integrates the signal missing rate and continuity measure by using the weighted fusion method, and obtains the comprehensive quality evaluation vector; extracts the monitoring point feature from the comprehensive quality evaluation vector, processes the feature by grouping through the K-means clustering algorithm, and obtains the priority grouping set; calculates the dynamic adjustment factor for the priority grouping set, obtains the real-time update value from the signal change trend, adjusts the grouping boundary if the update value exceeds the preset threshold, and obtains the dynamic priority sequence; fuses the risk set regional coordinates from the dynamic priority sequence, integrates the position optimization data by using the coordinate mapping method, and obtains the dynamic priority evaluation result.

[0018] In the present embodiment, firstly, the signal-to-noise ratio quantitative index calculation is performed, the module establishes a sliding window with a length of 60 time steps for each monitoring point and eliminates abnormal readings outside the upper and lower 5% in the window, then the remaining readings are segmented in chronological order, the continuous segment with the lowest energy and not less than 10% of the window length is selected, the intensity median of the segment is calculated as the noise interference value, and the median of the signal intensity in the whole window is calculated as the effective signal intensity. When there is a device silent segment in the window and the length is not less than 10% of the window length, the intensity median of the silent segment is preferentially taken as the noise interference value. The signal-to-noise ratio quantitative index is defined as the deterministic ratio of the effective signal intensity to the noise interference value, and the signal-to-noise ratio quality score is obtained by normalizing the current minimum value in the candidate set to 0, the maximum value to 1, and the intermediate value to linear interpolation at equal intervals. The window length is 60 time steps to ensure the balance between statistical stability and real-time performance. The abnormal reading elimination ratio is 5% to suppress the influence of impulse noise and not to discard normal data. Secondly, the integrity data quality index calculation is performed, the module calculates the planned sampling number and the successful sampling number for each monitoring point in the same window. The signal loss rate is equal to the planned sampling number minus the successful sampling number divided by the planned sampling number and limited between 0 and 1. The continuity measure is equal to the length of the longest continuous missing segment in the window divided by the window length and limited between 0 and 1. Both are deterministic values and are written into the quality cache at the end of each window. In the third step, the weighted fusion is performed to form a comprehensive quality evaluation vector. The module combines the signal-to-noise ratio quality score, the signal loss rate, and the continuity measure after normalization. The combination rule is to give a fixed weight and output a comprehensive score and three original components to form an evaluation vector. The weight of the signal-to-noise ratio quality score is 0.6, the weight of the signal loss rate is 0.25, and the weight of the continuity measure is 0.15. The fusion method is to take the signal-to-noise ratio quality score as the plus item and the loss rate and continuity as the minus item. The comprehensive score is equal to the plus item minus the weighted sum of the two minus items and limited between 0 and 1. The above three weights are selected by minimizing the sum of false positives and false negatives in the trial operation stage and are fixedly written into the configuration file and do not automatically change in the running.05 step size adjustment and record version number; the fourth step performs priority clustering grouping, the module inputs the numerical standardization of the comprehensive quality evaluation vector of all candidate monitoring points into the K-means clustering algorithm, the initial number of clustering clusters K is 3 and is named as high priority, medium priority and low priority respectively, the initialization adopts random multiple mode and runs 10 times and selects the one with the minimum total distance within the group as the output, the iteration stop condition is that the distance of the center movement in the adjacent two times is less than the preset minimum threshold or the iteration number reaches 100 times, whichever comes first, after the clustering is completed, a priority grouping set is generated and four statistical quantities, including the in-group mean, variance, sample number and minimum comprehensive score, are calculated for each grouping, which are used for subsequent boundary adjustment; the fifth step performs dynamic adjustment factor calculation and grouping boundary update, the module calculates three relative change quantities between the previous window and the current window for each monitoring point in a trend window with a length of 60 time steps, the three quantities are the relative improvement of the signal-to-noise ratio quality score, the relative change of the signal loss rate and the relative change of the continuity measure, the calculation of the relative change quantity is to divide the difference between the current window and the corresponding index of the previous window by the reference value of the previous window and clip the result between -1 and 1, then the three relative change quantities are synthesized into a dynamic adjustment factor with fixed weights, the weights are 0.5 for the relative improvement of the signal-to-noise ratio, 0.3 for the relative change of the loss rate and 0.2 for the relative change of the continuity, the synthesized dynamic adjustment factor produces a determined value at each evaluation period and is compared with the threshold value, if the dynamic adjustment factor is greater than or equal to 0.2, it is determined that the quality trend of the monitoring point has improved significantly and it is moved to a higher priority grouping, if the dynamic adjustment factor is less than or equal to -0.2, it is determined that the quality trend has decreased significantly and it is moved to a lower priority grouping, if the value is between -0.2 and 0.2, the original grouping remains unchanged, the threshold values 0.2 and -0.2 are obtained in the trial run stage according to the joint minimization of false alarm rate and missed alarm rate and are allowed to be adjusted in the range of 0.1 to 0.4 with a step size of 0.05, each adjustment records the time, reason and new value; in order to maintain the stability and availability of the grouping, the module performs a minimum cost strategy when the boundary is moved, it preferentially moves only the trigger monitoring point and checks whether the number of samples in the grouping is less than 10% of the total number, if it will violate the lower limit, it changes to move the next monitoring point with the closest comprehensive score to the boundary until the lower limit is met or the moving stops; the sixth step performs risk concentration region coordinate fusion and position mapping, the module reads the risk concentration region coordinate set output by the previous step and the spatial coordinates of each monitoring point, maps the monitoring points to the corresponding risk region according to the spatial inclusion relationship, if a monitoring point is located in the overlapping range of multiple risk regions, a region weight factor is calculated and the priority of the point is amplified, the determination method of the region weight factor is to calculate the number of risk units of all risk regions where the monitoring point is located and calculate the proportion of these numbers in the total number of risk units, and the proportion is converted to a linear proportion between 1 and 1.5 between the factors are rounded to two decimal places, when the monitoring point only falls into a single area, the area proportion is converted into the corresponding factor and applied, when the monitoring point does not fall into any risk area, the factor is 1 and no amplification is made, the amplified priority takes effect in sequence sorting while retaining the original grouping for auditing; after all steps are completed, the module generates a dynamic priority sequence in the order of high priority first, medium priority second, and low priority last, and outputs the final values of the four indicators in the comprehensive quality evaluation vector for each monitoring point, i.e. signal-to-noise ratio quality score, signal loss rate, continuity measure and comprehensive score, outputs the dynamic adjustment factor value and its trigger direction, outputs the priority grouping and possible grouping change record, outputs the risk area number and area weight factor, and outputs the sorting number for resource scheduling, all of the above quantities are determined values and are permanently archived at each evaluation period; the parameter explanations and determination rules are summarized as follows, the window length of 60 time steps is used to balance the statistical confidence and real-time performance, the abnormal rejection proportion is 5% for each to stabilize the estimation noise, the noise segment length lower limit is 10% of the window length to ensure representativeness, the three fusion weights 0.6, 0.25, and 0.15 are used to emphasize the importance of signal-to-noise ratio and consider completeness, the clustering cluster number is 3 to form a clear three-level priority, the maximum iteration number is 100 to ensure convergence, the weight of the dynamic adjustment factor is 0.5, 0.3, and 0.2 to highlight the dominant role of the signal-to-noise ratio trend, the threshold upper limit is 0.4 and the lower limit is 0.1 to avoid excessive frequent boundary jitter, the grouping sample number lower limit is 10% to maintain statistical significance for each group, and the area weight factor range is 1-1.5 to establish a unified and limited priority amplification scale between different risk areas.

[0019] The layout optimization module uses a genetic algorithm to generate a full coverage network preliminary layout configuration based on the dynamic priority evaluation results and blind area redundant positions, taking coverage rate and resource consumption as fitness functions, through multi-point crossover and random mutation. The blind area redundant position coordinates are obtained from the dynamic priority evaluation results, the genetic algorithm is used to initialize the population, the initial fitness value is calculated through the coverage rate function, and the preliminary population set is obtained. The resource consumption index is integrated for the preliminary population set, the multi-point crossover operation is used to generate offspring individuals, and the post-crossover individual sequence is obtained. Random mutation parameters are extracted from the post-crossover individual sequence, and if the mutation rate exceeds the preset threshold, the individual gene position is adjusted, and the mutation optimization set is obtained. The position optimization data is integrated for the mutation optimization set, the individuals are sorted by fitness evaluation, and the sorted individual list is obtained. The network layout initial coordinates are integrated from the sorted individual list, the coordinate mapping method is used to generate a full coverage layout, and the preliminary layout configuration is obtained.

[0020] In the embodiment, first, the spatial coordinates of the candidate monitoring points, the risk level to which they belong, the quality weight, the blind area unit coordinates and the redundant point coordinates are read from the dynamic priority evaluation results and the blind area redundant position, and the roadway grid and the list of constructible positions are established. The grid side length is determined by the same value as the previous module and is fixed after system initialization. Then individual encoding and initial population generation are performed. Each individual is composed of several monitoring point coordinates in a fixed order and does not repeat. The initial population size is 100. The first batch of individuals in the population is generated by placing high-priority candidate points first and supplementing points near the blind area unit by the nearest distance. The second batch of individuals is generated by extracting from the list of constructible positions in order of quality weight from high to low and with the constraint of not covering repeated points. The third batch of individuals is generated by moving to the nearest uncovered unit near the redundant point by the shortest path. After the third batch is merged, if it is less than 100, it is supplemented according to the first batch rule in a loop. After initialization, the coverage rate is calculated and feasibility repair is performed. The coverage rate is defined as the proportion of grid unit numbers that meet the coverage condition to the total grid unit number. The coverage condition is that the distance from the center of the unit to at least one monitoring point is not greater than the effective coverage radius. The coverage radius is a determined value such as 15 meters determined by equipment acceptance test. If a unit is not covered, the candidate point with the shortest distance and constructible near the unit is selected and inserted into the individual until the unit is covered. When inserting, if the resource limit is reached, the non-critical point with the lowest quality weight is replaced. Then the resource consumption and comprehensive fitness are calculated. Resource consumption includes three items: the number of monitoring points, the expected energy consumption, and the concurrent occupation. The number of monitoring points is a determined integer of the number of coordinates in the individual. The expected energy consumption is the sum of the path distances from each monitoring point to the nearest convergence node recorded in meters. The path distance is accumulated along the center line of the roadway. The convergence node coordinates are provided by the construction drawing and are fixed at initialization. The maximum value of the number of monitoring points working at the same time in the same grid area is occupied. The area size is segmented and counted in fixed lengths of 50 meters. The comprehensive fitness combines the coverage rate as a bonus item and the three resource consumptions as a deduction item with fixed weights. The coverage rate weight is 0.7, the number of monitoring points weight is 0.15, the expected energy consumption weight is 0.1, and the concurrent occupation weight is 0.05, the results are linearly normalized to the range of 0-1 after combination and 3 decimal places are reserved, and the comprehensive fitness of any infeasible individual is directly set to 0 if it still cannot achieve full coverage after repair; after completing the fitness evaluation, parent selection and multi-point crossover are performed, the parent selection adopts a deterministic truncation method to reserve the top 50 individuals according to the comprehensive fitness from high to low, and then pairing is performed for crossover from the 50 individuals, each pair of individuals uses 3 crossover points, and the crossover point positions are rounded to 25%, 50% and 75% of the length of the individual, after crossover, the offspring are executed for repeated coordinate cleaning and coverage repair according to the rule of blind area priority, the cleaning rule is that if a repeated coordinate appears, the candidate point with the second highest quality weight in the risk area where the coordinate is located and not used is replaced, if there is no replaceable point, it is replaced on the adjacent grid according to the nearest available location; after completing the crossover, random mutation is performed, the default individual mutation rate is 0.1, when mutation is triggered, not less than 1 gene position is randomly selected in the individual for adjustment, mutation contains two types of operations and selects according to a probability of 50% each, one is coordinate replacement, which replaces the coordinate at this position with a candidate point in the same risk area with similar quality weight and not used, the other is coordinate micro-movement, which moves the coordinate to the nearest available location without conflict with existing points along the blind area boundary direction, immediately after mutation, coverage verification and engineering verification are performed, coverage verification is at least 1 time coverage of the entire grid, engineering verification is that the path distance from the monitoring point to the nearest power communication node does not exceed the allowed upper limit, which is determined according to the construction specification at initialization, for example, 120 meters, if either verification fails, the mutation is revoked and a candidate replacement or micro-movement is reselected at the same position, if it still fails for 3 times in a row, it remains unchanged; the proportion of individuals triggering mutation in the current generation is calculated, if the proportion exceeds 0.3, the default individual mutation rate is reduced to 0.08 in the next generation to suppress search shock, if it is lower than 0.05, it is increased to 0.12 in the next generation to enhance diversity, all adjustments record the time stamp and new value; after completing the mutation, the coverage rate, resource consumption and comprehensive fitness of all individuals are recalculated and sorted, the top 50 individuals with the highest comprehensive fitness are reserved as the parent generation of the next generation and elite preservation is performed, the top 5 individuals with the highest comprehensive fitness in the current generation are directly copied to the next generation to prevent degradation; the termination condition of the intergenerational cycle is that either of the two determined rules is satisfied, one is that the maximum number of generations is reached, for example, 100 generations, the other is that the optimal comprehensive fitness is improved by less than or equal to 0 for 10 consecutive generations.005, After termination, select the individual with the highest comprehensive fitness as the current solution; finally, perform coordinate mapping and preliminary layout generation, map all monitoring point coordinates in the current solution to the actual tunnel coordinate system and translate to the nearest accessible location along the shortest accessible path in the non-constructible area, after mapping, verify the coverage rate to 1 and verify that all engineering constraints are met, if there are individual units with insufficient coverage, insert points according to the blind area priority insertion principle until the coverage rate reaches 1 and recalculate the comprehensive fitness to confirm that the resource consumption has not significantly deteriorated, after confirmation, output the coordinate set of the individual, the corresponding grid number, coverage rate, monitoring point number, expected energy consumption, concurrent occupation, comprehensive fitness and generation number as the preliminary layout configuration of the full coverage network.

[0021] The iterative adjustment module simulates the coverage effect of the full coverage network in the non-uniform field based on the preliminary layout configuration, and if the coverage rate is lower than the preset threshold, the monitoring point position is adjusted iteratively to obtain an updated layout configuration, including obtaining non-uniform field simulation data from the preliminary layout configuration, calculating the current coverage rate by comparing the proportion of the monitoring point coverage area and the total area using the coverage effect evaluation method, extracting the iterative position adjustment parameter if the coverage rate is lower than the preset threshold, and obtaining the adjusted monitoring point coordinates; fusing the blind area redundant coordinates for the adjusted monitoring point coordinates, generating an optimization sequence using a weighted summation method through a resource consumption index balancing process, determining the random variation parameters in the sequence to update the position data; extracting the fitness function value from the updated position data, integrating the monitoring point optimization information using the position data as input using a multi-point crossover sequence method to obtain a crossover layout set; applying a coverage rate threshold to the crossover layout set to obtain variation parameters, generating a final configuration by extracting random values from the sequence, and obtaining an updated layout configuration.

[0022] In the embodiment, firstly, all monitoring point coordinates and their belonging candidate identifiers are read from the preliminary layout configuration, spatial coordinates, constructible flag and non-uniform field intensity level of grid cells are read from the non-uniform field simulation data, then an effective coverage radius of 15 meters determined by equipment acceptance is used to generate a coverage area for each monitoring point, and the ratio of the number of grid cells covered by at least one monitoring point to the total number of grid cells under the current non-uniform field is calculated as the current coverage rate, the preset coverage rate threshold is 0.98, which is determined by replaying historical data during the trial operation period, and the joint target is to minimize the number of uncovered cells and not to exceed 5% false alarm rate, after determination, the threshold is fixed and written into the configuration file, if the current coverage rate is not less than 0.98, the preliminary layout is directly output as the updated layout configuration, and the coverage rate and resource indicators are recorded, if the current coverage rate is less than 0.98, enter the iterative position adjustment loop, the maximum number of rounds is set to 10 and the maximum adjustment step is set to 100, and the first condition is used as the termination condition, the proportion of monitoring points allowed to move in a single round is 0.2 to control the disturbance range, the maximum moving distance of a single point is 10 meters to limit the displacement step size, the non-constructible avoidance distance is 2 meters to ensure the safety boundary, and the blind area priority weight is 0.6To give higher weight to blind area coverage when balancing resources, the above values are determined by grid search during the trial period to minimize the weighted sum of coverage missing times and resource consumption, and are written into the configuration. When entering each round, the module first constructs a blind area list and a redundancy list. The blind area list is a set of grid cells that are not covered in the coverage rate statistics, and the redundancy list is a set of cell sets where the coverage number is significantly greater than 1 and the quality weight is the lowest. Then, the blind area list is sorted by non-uniform field strength from high to low, and each blind area cell is processed one by one. When processing a blind area cell, the module searches for mobile monitoring points within a radius of 50 meters around it and sorts the candidate points by distance from the blind area center from near to far. The first candidate point that does not cause a new blind area in its original service range is selected as the moving object. If the first candidate point moves, it will cause an uncovered cell at the original position. Select the next candidate point. The moving direction is fixed as the shortest reachable path from the point to the blind area center. The moving distance is not more than 10 meters, and the distance from the new position to any non-constructible boundary is not less than 2 meters. After completing the single-point movement, the coverage state is recalculated on the local grid, and the blind area list and the number of moved steps are updated until the blind area is covered or the upper limit of the moving ratio in this round is reached. After completing the movement in this round, the resource consumption balancing process is performed to avoid excessive resource cost due to blind area elimination. Resource consumption is composed of three indicators: monitoring point number, estimated energy consumption, and concurrent occupancy. The monitoring point number is directly equal to the number of points in the current coordinate set. The estimated energy consumption is the sum of the path distances from all monitoring points to the nearest convergence node, measured in meters. The path is accumulated along the center line of the tunnel, and the convergence node coordinates are determined and fixed at initialization from the construction drawings. The occupancy is equal to the maximum number of monitoring points working simultaneously in a 50-meter-long area. The three indicators are normalized to 0 to 1 using the historical minimum and maximum values, and then the weighted sum is obtained using fixed weights. The weights are monitoring point number 0.5, estimated energy consumption 0.3, and concurrent occupancy 0.2. The weights are calibrated and fixed during the trial period with energy consumption and bandwidth constraints as boundary conditions and with the principle of not increasing false positives and false negatives. The balancing process evaluates each moved monitoring point to determine the coverage rate gain change and resource cost change if it is returned to the adjacent feasible position. If the resource cost increment is greater than the coverage rate gain multiplied by the blind area priority weight 0.6, then the return is executed. Otherwise, the moving result is retained. After balancing, an optimized sequence is formed and sorted by the contribution to coverage rate improvement from high to low for the next step of crossover and mutation. Then, the random mutation parameters are calculated and adaptive adjustment is performed. The default mutation rate is 0.1. When the coverage rate improvement in this round is less than 0.002, the mutation rate in the next round is increased to 0.12 to enhance exploration. When the coverage rate improvement in this round is greater than 0.01, the mutation rate in the next round is reduced to 0.08 to suppress oscillation. The mutation rate is limited to 0.05 to 0.2 range and record the time and new value at each change; then perform a multi-point crossover sequence method to integrate the position data to form a more optimal layout, specifically, a set of positions is extracted from the first 50% and the last 50% of the optimized sequence in pairs and 3 crossover switching positions are set on each pair of sets, the 3 positions are respectively 25%, 50% and 75% of the length of the set rounded down, the corresponding segments of the two sets are exchanged according to the crossover point to generate 2 candidate layouts, immediately after the exchange, repeat the coordinate cleaning to eliminate duplicate points and insert the minimum number of new coordinates according to the blind area priority rule to restore full coverage feasibility, all candidate layouts form a set of post-crossover layouts; on the post-crossover layout set, perform random variation according to the current mutation rate, at least 1 monitoring point is selected for each candidate layout to adjust the gene position, the adjustment operation includes coordinate replacement and coordinate micro-movement, each accounting for 50% probability, coordinate replacement is to replace the current coordinate with a candidate coordinate with similar quality weight from the same risk area and not used, coordinate micro-movement is to move not more than 5 meters to the nearest constructible position along the normal direction of the nearest blind area boundary, after the variation is completed, two checks are performed, coverage check and engineering check, the coverage check requires the coverage rate of the new layout to be not less than the coverage rate at the beginning of the round, the engineering check requires the path distance of each monitoring point to the nearest power communication node to be not more than 120 meters, either of the two checks does not pass, the variation is rolled back and a new set of coordinates is extracted for replacement or micro-movement at the same position, a maximum of 3 retries still fail, the original coordinates are kept and not adjusted; calculate the fitness function value for all layouts after variation and perform screening, the fitness is composed of coverage rate and resource cost and is linearly normalized to 0 to 1, the coverage rate is the weight of the plus item and is 0.7, the resource cost is the weight of the minus item and is 0.3, retain a number of layouts with the highest fitness to enter the threshold judgment, layouts with a fitness lower than the best fitness at the beginning of the round minus 0.02 are eliminated to avoid regression; finally, apply the coverage rate threshold judgment and the final configuration generation rule, if the coverage rate of at least 1 layout is not less than 0.98, select the layout with the highest fitness as the updated layout configuration and end the loop, if the coverage rate of all layouts is less than 0.98, randomly extract 1 from the top 10 layouts with the highest fitness as the reference layout for the next round and generate new initial position data from the sequence according to the current variation rate and random value to enter the next round, the loop is terminated when the maximum number of rounds 10 is reached or the coverage rate improvement is not higher than 0.002 for 3 consecutive rounds, and the layout with the highest fitness is output as the updated layout configuration; parameter explanation and determination rule: the coverage radius of 15 meters is derived from the equipment acceptance test report and is fixed at initialization, the coverage rate threshold of 0.98 is derived from the inflection point with the minimum number of uncovered units in historical data playback and is determined in combination with safety redundancy, the maximum number of rounds 10 and the maximum adjustment step number 100 are used to limit the calculation time and avoid overfitting, the single round movement ratio of 0.2 is used to ensure gradual convergence and avoid large disturbances, the maximum single point movement distance of 10 meters and the non-constructible avoidance distance of 2 meters are used to meet the construction accessibility and safety requirements, the blind area priority weight of 0.6 To provide explicit trade-off between resource cost and coverage gain, resource weight 0.5, 0.3, 0.2 to incorporate quantity, energy consumption and concurrency into the same scale, patch length 50 meters to count spatial scale of concurrency, engineering distance upper limit 120 meters to meet the boundary conditions of power supply and communication stability, fitness elimination difference 0.02 to prevent inferior solutions from entering the next step, mutation rate upper and lower limits 0.05 and 0.2 and adaptive step rule to keep balance between convergence and diversity.

[0023] The trend prediction module extracts the deformation monitoring data stream from the updated layout configuration, adopts a time series analysis method to predict the roadway deformation trend, and determines the potential risk evolution path. The deformation monitoring data stream is obtained from the updated layout configuration, the time series analysis method is adopted, the autoregressive integral moving average model is used to process the sequence correlation and stationarity by taking the data stream as the input, and the roadway deformation trend is obtained. The geological stability index is fused for the roadway deformation trend, the rock strength and stress distribution value are extracted from the preset geological database, the potential risk factors are integrated by using the weighted summation method, the potential risk factors are calculated by weighting the historical deformation records and environmental variables, and the risk evolution path is determined. The abnormal fluctuation sequence is extracted from the risk evolution path, the threshold comparison method is adopted to judge whether the fluctuation exceeds the preset threshold, and the alarm signal is generated to obtain the emergency response sequence. The equipment maintenance log is integrated for the emergency response sequence, the maintenance time and fault type are obtained from the monitoring equipment history record, the similarity of the log sequence and the response sequence is compared by the sequence matching process to identify the matching items, the optimization adjustment parameters are obtained, and the potential risk evolution path is determined.

[0024] In the embodiment, firstly, deformation monitoring data streams are extracted from the updated layout configuration by monitoring point, time alignment, deduplication and missing value filling are performed, time alignment is based on system unified time, matching within plus or minus one sampling period is allowed, linear interpolation filling is used for exceeding range, deduplication rule is to retain only one record with the highest quality weight when multiple records with the same timestamp appear, missing value threshold is 10%, when the missing value proportion of a monitoring point exceeds 10% in a window, the prediction result of the point in the window is marked as invalid and a low level quality prompt is triggered; then, steady-state preprocessing is performed on each data stream in a sliding window with a length of 60 time steps, first, stationarity test is performed, if not stationary, difference order candidate set 0 to 2 is tried one by one, and the lowest order after passing the stationarity test after difference is taken as the determined value, if none of them passes, still take 2 and increase the residual threshold in subsequent modeling to prevent false positives, preprocessing simultaneously uses a sliding average with a length of 5 time steps to remove high-frequency noise and removes abnormal spikes with 5% at the top and bottom by using the interquartile range, the above window length and quantile threshold are selected and fixed in the trial operation stage by replaying historical data and taking the maximum early warning hit rate and the false alarm rate not higher than 5% as the target; after preprocessing, enter the time series modeling and prediction stage, the module uses autoregressive integrated moving average model to model the stationary sequence of each monitoring point, the model order is selected from the grid of autoregressive candidate 0 to 5 and moving average candidate 0 to 5, the difference order takes the determined value, the evaluation criterion is the minimum value of the sum of Akaike information criterion score and Bayesian information criterion score, if there is a tie, the one with smaller root mean square error is better, training is performed at the end of each window and generates point prediction and interval prediction of 1 step, 5 steps and 30 steps, the update period is every 1 time step, the model forces the residual mean to be close to 0 during training and checks whether the autocorrelation of the residual is lower than 0.1, if not met, automatically increase the order of moving average and retrain until met or reach the upper limit of the candidate; the module writes the increments and slopes of each step prediction into the trend indicator table and generates trend labels at each monitoring point, the trend labels are output according to the following determination rule, if the prediction increments of at least 45 time steps in the past 60 time steps are positive and the average slope is not less than 2 times the average slope of the historical stable period, it is marked as continuous rise, if the prediction increments of at least 45 time steps are negative and the average slope is not higher than-2 times the average slope of the historical stable period, it is marked as continuous decline, and the rest is marked as basically stable, the historical stable period is defined as a section with no alarm and environmental disturbance record for 7 consecutive days, which is used to calculate the baseline slope and threshold and is fixed at the end of the trial run; after completing the single-point trend, the module fuses the geological stability index to obtain the unit-level risk score and determines the potential risk evolution path accordingly, the geological stability index reads the rock strength and stress distribution values from the preset geological database and normalizes them to 0 to 1 with the minimum and maximum values, the potential risk factors consist of three parts, namely the predicted deformation intensity index, the historical deformation fluctuation index and the environmental load index, the predicted deformation intensity index takes the maximum displacement prediction increment of the last 30 time steps after minimum-maximum normalization, the historical deformation fluctuation index takes the standard deviation of the observation increment of the last 60 time steps after minimum-maximum normalization, the environmental load index gives fixed influence coefficients according to the records of tunneling, blasting, pumping, rainfall and other records on the same day and normalizes the weighted sum to 0 to 1, the weights of the above three are 0.5, 0.3 and 0.2 respectively, the overall weight of the geological stability index is 0.4, and it is linearly combined with the total weight 0.6 of the above three to form the unit-level risk score, all weights are selected in the trial run stage to minimize the sum of false positives and false negatives and are fixed as configuration parameters; the module constructs a directed graph of all units on the grid at each time step according to the adjacency relationship and uses step-by-step expanding dynamic programming to search for the continuous path with the maximum cumulative score from the unit with the highest score to the adjacent units, the path length is 30 time steps, if there are multiple paths with the same score, the path containing more high-risk units is selected as the priority, the path obtained by searching is the potential risk evolution path and the path cumulative score and key node coordinates are recorded; the module then extracts the abnormal fluctuation sequence from the risk evolution path and generates the emergency response sequence, the abnormal fluctuation adopts a double-threshold judgment method, the first threshold is that the absolute residual error of the observation value and the corresponding prediction value exceeds 3 times the rolling standard deviation of the same window, the second threshold is that the single-step risk score is improved by not less than 0.15 and occurs continuously for not less than 3 time steps, if any threshold is met, an alarm signal is generated at that time step and written into the emergency response sequence, and the alarm level is mapped according to the overrun amplitude as 3 levels corresponding to high, 2 levels corresponding to medium and 1 level corresponding to low, the rolling standard deviation window length is 60 time steps.15threshold value is taken at 95% of the historical stable period and is fixed; after generating the emergency response sequence, the module integrates the device maintenance log to identify alarms related to device events and optimizes parameters accordingly, the device maintenance log obtains repair time and fault type from the monitoring device history record and sorts them by time to form a log sequence, the module performs sequence matching between the log sequence and the emergency response sequence within an alignment window of 90 time steps to calculate similarity, the similarity threshold is 0.75 and the time lag is allowed to be no more than 5 time steps, if both conditions are met, it is determined to be a matching item and two types of optimization are performed according to the matching type, the first type is model parameter optimization, when the matching is concentrated in a single monitoring point and the fault type is sensor anomaly, the differential order of the point is increased by 1 and the maximum candidate order of autoregression and moving average is decreased by 1 in the next window to reduce overfitting and enhance robustness, the second type is risk synthesis weight optimization, when the matching is dispersed in multiple monitoring points and the fault type is system maintenance, the geological stability index weight is temporarily reduced to 0.3 and the predicted deformation intensity index weight is temporarily increased to 0.6 in the next window, and after maintaining for 1 window, it is restored, all optimization actions record time, object and new value and take effect in the next period; the module finally outputs content including trend label of each monitoring point, point prediction and interval prediction of each time step, unit level risk score and spatial distribution, potential risk evolution path and cumulative score, emergency response sequence and alarm level, device log matching result and parameter optimization record generated therefrom, parameter description and determination method as follows, time alignment tolerance is 1 sampling period to accommodate communication jitter, window length 60, 30 and 90 are used to measure the time scale of short-term prediction, path planning and log matching respectively, differential order candidate 0 to 2 and model order candidate 0 to 5 are used to cover common dynamic characteristics on the premise of ensuring real-time, trend determination requires 45 time steps and 2 times slope to balance robustness and sensitivity, abnormal residual threshold 3 times to control false positives, risk jump threshold 0.15 and 3 consecutive time steps to exclude single pulse, log matching similarity threshold 0.75 and lag threshold 5 to ensure matching credibility, the values of the three risk factors and the weight of geological stability are used to establish a fixed proportion between data-driven and geological constraints.

[0025] The resource dynamic allocation module dynamically adjusts the sensor resource allocation proportion according to the potential risk evolution path and the real-time feedback mechanism. If the stress distribution deviates due to the change of the geological condition, the non-uniform field is simulated again to obtain an optimized monitoring position set. The geological condition change data is obtained from the risk evolution path, the change sequence is processed by using the real-time feedback mechanism, and the stress distribution deviation sequence is obtained. For the stress distribution deviation sequence, the field distribution value is extracted from the deviation sequence by non-uniform field simulation, and the difference between the distribution values is processed by using the difference calculation method. If the difference exceeds the preset threshold, the resource dynamic adjustment is triggered, and the sensor allocation proportion set is obtained. According to the sensor allocation proportion set, the path feedback integration data is integrated, the condition change response index is extracted from the preset geological database, and the monitoring position optimization parameter is determined. The deviation simulation triggering sequence is fused by using the monitoring position optimization parameter to obtain an optimized monitoring position set.

[0026] In the embodiment, first, the geological condition change data including the rock strength update value, the surrounding rock stress update value and the environmental load record are read from the potential risk evolution path step by step, and a change sequence is formed. The time alignment is based on the system unified time, which allows matching within 1 sampling period in positive and negative directions. If the tolerance is exceeded, linear interpolation is used to align adjacent time. Then, the stress distribution offset sequence is obtained by executing the deterministic processing flow of the real-time feedback mechanism. The flow is divided into three steps: deduplication, anomaly suppression and window aggregation. The deduplication rule is to keep only the record with the highest quality weight for the same timestamp. The anomaly suppression rule is to remove the peak values of 5% at the top and bottom in the sliding window with a length of 60 time steps, and replace them with the median of the adjacent two sides in the window. The window aggregation rule is to take the median of the change sequence in the sub-window with a length of 5 time steps to smooth the short-term fluctuations and output the stress distribution offset sequence. In the second step, the non-uniform field simulation is performed on the stress distribution offset sequence, and the unit difference degree is calculated. The non-uniform field simulation uses the consistent grid and boundary conditions of the previous module. The load increment and support change in each time step in the offset sequence are used as input to re-solve the field distribution values of the whole grid. Then, two types of difference values are calculated on each grid cell, and the unit difference degree is formed by combining them with a fixed weight. The first type is the absolute difference, which is the absolute amount of the difference between the current field distribution value and the latest stable distribution value. The second type is the neighborhood difference, which is the average amount of the distribution value difference between the cell and its adjacent cells. The combination weight is 0.7 for the absolute difference and 0.3 for the neighborhood difference. The obtained unit difference degree is a non-negative definite value in the range of 0 to positive infinity. In the third step, threshold judgment and trigger judgment are performed. The module calculates the maximum value, average value and unit proportion exceeding the threshold value of the unit difference degree in the whole grid. The difference threshold is determined by calculating the distribution of the unit difference degree in the historical stable period and taking the 95% corresponding value as the reference threshold. The reference threshold is multiplied by the safety factor 1.2 to obtain the execution threshold. The historical stable period is defined as a section with no alarm and zero construction disturbance for 7 consecutive days, and is fixed at the end of the trial run. If any unit difference degree in the current statistics is not less than the execution threshold or the unit proportion exceeding the threshold is not less than 10%, the resource dynamic adjustment is triggered and the proportion calculation is entered. Otherwise, the existing proportion is maintained and the reason for not triggering is recorded. In the fourth step, the sensor allocation proportion set is calculated and resource constraint review is performed. The module indexes each monitoring point with the monitoring location candidate set to allocate the basic proportion. The basic proportion is consistent with the dynamic priority, which is divided into three levels: high 0.5, medium 0.3 and low 0.2. When a monitoring point is down-weighted due to failure in the previous period, its basic proportion is directly assigned as 0.1 and returns to the original level after recovery. Then, the gain amount is calculated based on the unit difference degree and the difference degree weighted average of the monitoring point's coverage cell set. The weight of the cells in the coverage is normalized by the inverse of the cell area and the distance from the site, and is limited to 0 to 1. The new proportion is equal to the basic proportion plus the gain amount multiplied by the gain coefficient. The gain coefficient is initially 0.4 and is adjusted by 0.05 steps in the running based on the combined index of false positives and false negatives, but is limited to 0.2 to 0.6 between, the new ratio is normalized in the same area to ensure the proportion and in the area is 1, after the completion of the proportion update, three resource constraints are checked, the power consumption budget is taken as the upper limit of the total available power consumption of the area and the estimated occupation is estimated according to the rated power consumption and duty cycle of the site, the bandwidth budget is taken as the upper limit of the available bandwidth of the link and the estimated occupation is estimated according to the uplink data rate and the expected retransmission rate, and the concurrent upper limit is taken as the upper limit of the number of sites allowed to work simultaneously in the scheduling period, if any constraint exceeds the upper limit, the corresponding site proportion is recovered in proportion according to the unit difference from low to high order until all constraints are met and the recovery amount is recorded; the fifth step is to integrate the path feedback according to the sensor allocation proportion set and extract the condition change response index from the preset geological database to determine the monitoring position optimization parameter, the path feedback is linearly mapped to the risk segment level in the range of 0 to 1 according to the cumulative score of the risk evolution path, the high risk segment is 0.67 to 1, the medium risk segment is 0.33 to 0.67, and the low risk segment is 0 to 0.33, the rock strength level and stress distribution level are extracted from the geological database and normalized to 0 to 1, the module combines the proportion set and the risk segment level into three types of device parameters, namely the sampling rate factor, the transmission power factor and the duty cycle factor, the sampling rate factor is taken between 1.5 to 2 in the high risk segment, between 1 to 1.5 in the medium risk segment, and between 0.5 to 1 in the low risk segment, and is linearly positioned in the interval according to the proportion set size, the transmission power factor is taken as 1.2 in the high risk segment, 1 in the medium risk segment, and 0.8 in the low risk segment, the duty cycle factor is taken as 1.2 in the high risk segment, 1 in the medium risk segment, and 0.8 in the low risk segment, and the rock strength level and stress distribution level and the unit difference are taken as 0.2 and 0.2 and 0.6 Weighted to get position weight, which is used to sort candidate points and filter monitoring location optimization parameter set in the same risk segment; the sixth step generates the result of the position set by fusing the monitoring location optimization parameter offset simulation trigger sequence, the module selects candidate points from high to low according to position weight in each risk segment and gives them sampling rate factor, transmit power factor and duty cycle factor, while performing two hard constraint checks, namely coverage constraint and construction constraint, the coverage constraint requires that all grid cells be covered by at least one station, the construction constraint requires that the path distance from the station to the nearest power communication node be no more than 120 meters and the minimum distance from the station to the non-constructible boundary be no less than 2 meters, if coverage is insufficient during selection, continue to add points in the corresponding risk segment according to position weight until the coverage constraint is met, if the resource constraint is exceeded, first backtrack and remove the newly added point with the lowest position weight, if it is still insufficient, perform a micro-shift on the station with the second lowest position weight, the micro-shift moves no more than 10 meters along the direction of the nearest blind area or the center line of the channel and performs coverage and construction double checks again, and outputs the optimized monitoring location set after all constraints are met; the seventh step forms and archives the output content, which includes the coordinates, final allocation ratio, sampling rate factor, transmit power factor, duty cycle factor, coverage range, position weight, risk segment number and resource occupation details of each station, and includes the area-level power consumption ratio, bandwidth ratio, concurrency ratio and record of whether to trigger recovery, while recording the difference threshold, trigger condition, gain coefficient and any parameter fine-tuning history of this period for tracing back; the parameters and thresholds are explained as follows, the time alignment tolerance of 1 sampling period is used to compatible communication jitter, the sliding window length of 60 and the sub-window length of 5 and the 5% of spike rejection are used to balance between robustness and real-time performance, the weights of 0.7 and 0.3 of the unit difference degree are used to emphasize the absolute offset while considering spatial continuity, the 95% of the difference threshold and the safety factor of 1.2 are used to control the trigger sensitivity and to calibrate during the trial period by playing back the data to minimize the sum of false positives and false negatives, the basic ratios of 0.5, 0.3 and 0.2 are used for the resource starting point consistent with the dynamic priority, the gain coefficient value range of 0.2 to 0.6 is used to limit the single round ratio change amplitude, the upper bounds of the three resources are determined by equipment acceptance and link test and are fixed at initialization, the risk segment interval and the factor interval endpoints of the three types of equipment are selected and remain unchanged during the trial period to maximize the early warning hit rate and keep the energy consumption within the budget, the position weight ratios of 0.6, 0.2 and 0.2 are used to set a fixed trade-off between data-driven and geological constraints, the 120 meters, 2 meters and 10 meters of micro-shift upper limit in the construction constraint come from the construction specification and safety distance.

[0027] The feedback closed loop module integrates signal-to-noise ratio and integrity data quality feedback cycle through the optimized monitoring position set, adopts genetic algorithm with the minimum coverage rate threshold as the constraint to adjust the resource allocation ratio, and obtains the full-coverage monitoring network stable operation configuration, specifically including obtaining signal-to-noise ratio integrated data from the monitoring position optimization, adopting integrity feedback processing data sequence to obtain genetic algorithm adjustment input set; for the genetic algorithm adjustment input set, the resource allocation ratio is fused through the coverage rate constraint to determine the full-coverage network parameters, wherein the genetic algorithm takes the input set as the initial population and iteratively optimizes the parameters through selection, crossover and mutation operations; according to the full-coverage network parameters, the stable operation configuration index is obtained, and if the index is lower than the threshold constraint setting, the dynamic adjustment mechanism is triggered to obtain the configuration stable verification sequence; through the configuration stable verification sequence, the data quality cycle response is integrated, the multi-sensor fusion index is extracted from the preset geological response library to obtain the real-time data verification set, wherein the preset geological response library is pre-established based on historical geological condition change data; for the real-time data verification set, the threshold constraint setting is fused with the configuration stable verification sequence to generate the full-coverage monitoring network stable operation configuration.

[0028] In the embodiment, firstly, the signal-to-noise ratio integrated data and the integrity data are extracted from the monitoring position optimization result, and a time synchronization sequence is established; the time alignment is based on the system unified time, and matching within 1 sampling period is allowed; if it exceeds, linear interpolation is adopted for supplement; the signal-to-noise ratio integrated data is weighted median of intensity value in a window with a length of 60 time steps according to quality weight, and the ratio is calculated based on the median of the window noise; then, the ratio is linearly standardized to obtain the signal-to-noise ratio score by mapping the minimum value of the candidate point at the current time to 0 and the maximum value to 1; the integrity data includes two items of missing rate and continuity measure; the missing rate is equal to the planned sampling number in the window minus the successfully sampled number, and then divided by the planned sampling number and limited to 0 to 1; the continuity measure is equal to the length of the longest continuous missing segment in the window divided by the window length and limited to 0 to 1; the abnormal value processing excludes the peaks of 5% at the top and bottom in the same window, and replaces them with the adjacent median; then, the three indexes are combined into a comprehensive quality score according to the fixed weight; the weight is 0.6 for the signal-to-noise ratio score, 0.25 for the missing rate, and 0.15 for the continuity measure; wherein, the missing rate and the continuity are involved in the calculation as penalty items, and the result is linearly clipped to 0 to 1; finally, the comprehensive quality score of each monitoring point and the three original indexes are written into the genetic algorithm adjustment input set; when entering the genetic optimization stage with coverage constraint, the module encodes the individual into a resource allocation proportion vector based on the input set; the vector length is equal to the number of monitoring points; each component is the resource proportion of the point in the current period, and is normalized at the individual level to make the sum of each component equal to 1; the initial population size is 100; 40 individuals are allocated with a basic proportion from high to low according to the comprehensive quality score, and a uniform random disturbance not more than 0.02 is injected; the basic proportion is consistent with the dynamic priority and is divided into three grades: high 0.5, medium 0.3, and low 0.2; 30 individuals are randomly initialized in proportion to the blind area weight and the historical alarm frequency; 30 individuals are initialized according to uniform distribution to enhance diversity; the fitness function adopts a weighted target combination of coverage hard constraint and quality priority; the coverage hard constraint threshold is 0.98; if the coverage rate of the individual is less than 0.98 after coverage evaluation, the fitness is set to 0 to be directly eliminated; after meeting the coverage constraint, the fitness is calculated according to three items of comprehensive quality improvement, energy consumption penalty, and concurrency penalty, and is linearly normalized to 0 to 1; the comprehensive quality improvement is equal to the average of the comprehensive quality score improvement in the next 60 time steps after adopting the proportion; the energy consumption penalty is minimized by summing the path distance to the nearest convergence node after multiplying each point proportion; the concurrency penalty is minimized by the peak value of the sum of the proportions of each point in a 50-meter area; the weights of the three items are 0.7, 0.2, and 0.1 respectively; the selection operator adopts deterministic truncation to reserve the first 50 individuals and fills the parent pairing in the roulette way; the crossover operation adopts 3-point crossover; the 3 crossover positions are 25%, 50%, and 75% of the vector length, and are rounded down; after crossover, the offspring is executed for proportion normalization and coverage repair; the coverage repair rule is to preferentially improve the proportion of adjacent high-quality monitoring points in the uncovered unit until the coverage rate is not less than 0.98.98; random variation to select no less than 1 component in offspring as proportional fine-tuning at individual variation rate 0.1, each fine-tuning amplitude does not exceed 0.05 and keeps vector sum 1, if the proportion of individuals triggering variation in this generation exceeds 0.3, the next generation variation rate is set to 0.08, if less than 0.05, the next generation variation rate is set to 0.12, the variation rate is limited to 0.05 to 0.2; elite preservation directly copies the top 5 individuals in the current generation to the next generation to avoid degradation, the iteration termination condition is to reach 100 generations or the optimal fitness improvement of 10 consecutive generations is not higher than 0.005, the first to reach is taken as the criterion, after termination, the individual with the highest fitness is selected as the resource allocation proportion set in the full coverage network parameters and linear mapping is given to the recommended values of 3 operation parameters, the sampling rate factor interval is 0.8 to 2, the transmission power factor interval is 0.8 to 1.2, the duty cycle factor interval is 0.8 to 1.2 and 3 decimal places are reserved; the module then calculates the stable operation configuration index and judges whether to trigger dynamic adjustment, the stable operation configuration index includes 4 items of coverage stability index, quality stability index, energy consumption stability index and reconfiguration frequency, the coverage stability index takes the minimum value of the coverage rate in the last 60 time steps, the quality stability index takes the 10% of the comprehensive quality score in the last 60 time steps, the energy consumption stability index takes the inverse of the 90% of the unit time energy consumption in the last 60 time steps after 0 to 1 mapping, the reconfiguration frequency takes the inverse of the number of configuration updates triggered in the last 7 days after 0 to 1 mapping, the above 4 items are compared with the threshold to synthesize the stability score, the threshold is that the coverage stability index is not less than 0.98, the quality stability index is not less than 0.7, the energy consumption stability index is not less than 0.6, and the reconfiguration frequency is not higher than 1 time per day, the unmet items participate in deduction according to the weight 0.4, 0.3, 0.2 and 0.1, if the stability score is less than 0.8, the dynamic adjustment mechanism is triggered and the configuration stability verification sequence is generated; in the dynamic adjustment mechanism, the module integrates the configuration stability verification sequence with the data quality feedback cycle and extracts multi-sensor fusion indicators from the pre-set geological response library to construct a real-time data verification set, the pre-set geological response library is pre-built by historical geological condition change data and maintained by mining section, the real-time data verification set includes consistency index, proportional configuration residual and stability index time series, the consistency index is in the range of 0 to 1, indicating the consistency degree of multi-sensor response to the same unit, the proportional configuration residual is the time series of the difference between the measured quality score and the predicted quality score after pushing according to the current proportion; the threshold fusion stage is constrained by 3 threshold values for judgment, the lower limit of consistency is 0.7, the upper limit of residual is 3 times the corresponding window rolling standard deviation, and the lower limit of stability score is 0.8, when the consistency of any monitoring point is less than 0.7 or the residual exceeds 3 times or the whole network stability score is less than 0.8, fine-grained adjustment is performed in the fixed order until the threshold is met or the maximum adjustment step 100 is reached, one of which is to recover and transfer from the point with the lowest consistency to the point with the highest consistency in the 50-meter area where the point is located at a proportion of 0.02, the other is to reduce the sampling rate factor of the problem point by no more than 0.1and simultaneously up-regulate the adjacent high-consistency point sampling rate factor by no more than 0.1 each time, and thirdly when the residual exceeds the upper limit for 3 consecutive time steps, up-regulate the transmit power factor of the point by no more than 0.05 each time until no higher than 1.2 and simultaneously up-regulate the duty cycle factor by no more than 0.05 each time until no higher than 1.2, and immediately after each step of adjustment, recalculate the coverage rate and the 4 stability indicators and write back to verify the sequence; when all threshold values are met or the maximum adjustment step number is reached, stop adjusting and generate a full-coverage monitoring network stable operation configuration, output the final resource allocation ratio, sampling rate factor, transmit power factor, duty cycle factor, coverage range, 4 stability indicators, whether to trigger dynamic adjustment, and when the genetic optimization algebra, optimal fitness, eliminated individual number and elite retention number, etc. Audit information; The determination method of parameters and threshold values is as follows, the coverage rate threshold 0.98 is derived from the compromise inflection point of the number of uncovered units and false alarm rate in historical data playback, the comprehensive quality weights 0.6, 0.25 and 0.15 are obtained by grid search to minimize the sum of false positives and false negatives during the trial operation period and are fixed, the fitness weights 0.7, 0.2 and 0.1 are used to ensure that quality improvement is prioritized while taking into account energy consumption and concurrency, the window length 60 and 7 days are used to cover short-term fluctuations and weekly-scale stability assessment at the same time, 5% of the abnormal data is used to suppress sharp peak interference, the crossover point position, population size 100, mutation rate range 0.05 to 0.2, elite number 5, maximum generation 100 and convergence criterion 10 generations not higher than 0.005 are selected in the trial operation stage with convergence speed and solution stability as the target and are fixed when online, the stability score threshold 0.8, the consistency lower limit 0.7, the residual 3 times and the reconfiguration upper limit once a day are determined in the trial operation stage with the maximum early warning hit rate and no increase in operation and maintenance cost as the target.

[0029] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A real-time analysis system for mine roadway deformation data based on edge computing, characterized in that, include: The data acquisition module collects initial data on tunnel deformation through sensors and uses the finite element analysis method to simulate the evolution of the deformation field, thereby obtaining the characteristics of non-uniform field distribution and potential blind zone redundancy locations. The feature extraction module extracts key stress distribution parameters from the non-uniform field distribution features. If the parameters exceed a preset threshold, they are marked as risk concentration areas, and a candidate set of monitoring locations is determined. The signal quality assessment module acquires the signal-to-noise ratio and integrity data quality indicators of sensor signals for the candidate set of monitoring locations, and uses a clustering algorithm to prioritize the monitoring points to obtain dynamic priority assessment results. The layout optimization module uses a genetic algorithm with coverage and resource consumption as fitness functions, based on the dynamic priority evaluation results and the redundant positions of blind spots, to generate a preliminary layout configuration of the full coverage network through multi-point crossover and random mutation. The iterative adjustment module simulates the coverage effect of the full coverage network under a non-uniform field through the initial layout configuration. If the coverage rate is lower than the preset threshold, the monitoring point positions are adjusted iteratively to obtain the updated layout configuration. The trend prediction module extracts deformation monitoring data streams from the updated layout configuration, uses time series analysis to predict roadway deformation trends, and determines potential risk evolution paths. The resource dynamic allocation module dynamically adjusts the sensor resource allocation ratio based on the potential risk evolution path and real-time feedback mechanism. If changes in geological conditions cause a shift in stress distribution, the non-uniform field is re-simulated to obtain an optimized set of monitoring locations, including: Geological condition change data are obtained from the risk evolution path, and the change sequence is processed by a real-time feedback mechanism to obtain the stress distribution migration sequence; For the stress distribution migration sequence, the field distribution value is extracted from the migration sequence through non-uniform field simulation and the difference between the distribution values ​​is processed by the difference calculation method. If the difference exceeds the preset threshold, the resource dynamic adjustment is triggered to obtain the sensor allocation ratio set. Based on the sensor allocation ratio set, the path feedback data is integrated, and the condition change response index is extracted from the preset geological database to determine the optimal parameters for the monitoring location. An optimized set of monitoring locations is obtained by fusing the offset simulation trigger sequence with the optimized parameters of the monitoring location; The feedback closed-loop module, through an optimized set of monitoring locations, integrates signal-to-noise ratio and integrity data quality feedback loops. Using a genetic algorithm constrained by a minimum coverage threshold, it adjusts resource allocation ratios to obtain a stable operating configuration for the full-coverage monitoring network, including: Signal-to-noise ratio integrated data is obtained from the monitoring location optimization, and the data sequence is processed using integrity feedback to obtain the genetic algorithm to adjust the input set; The genetic algorithm adjusts the input set and determines the full-coverage network parameters by fusing the resource allocation ratio through coverage constraints. The genetic algorithm uses the input set as the initial population and iteratively optimizes the parameters by selecting crossover and mutation operations. Based on the full coverage network parameters, obtain stable operation configuration indicators, determine if the indicators are lower than the threshold constraint setting, trigger the dynamic adjustment mechanism, and obtain the configuration stability verification sequence; By configuring a stable verification sequence and integrating the data quality cyclic response, multi-sensor fusion indicators are extracted from the preset geological response database to obtain a real-time data verification set. The preset geological response database is pre-established based on historical geological condition change data. For the real-time data validation set, a threshold constraint is used to set a stable validation sequence for the fusion configuration, and a stable operation configuration for the full-coverage monitoring network is generated.

2. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The data acquisition module collects initial data on roadway deformation through sensors, and uses the finite element analysis method to simulate the deformation field evolution process, obtaining the non-uniform field distribution characteristics and potential blind zone redundancy locations, including: Initial data of tunnel deformation is collected by sensors, deformation parameter values ​​are obtained from the initial data, and the finite element analysis method is used to divide the grid elements and apply boundary conditions to establish a simulation grid and obtain the initial model of the deformation field. For the initial model of the deformation field, input the evolution process parameters, and simulate the evolution of the non-uniform field distribution by iteratively calculating the stress distribution to determine the distribution characteristic vector; The coordinates of potential blind zones are extracted from the distribution feature vector. If the coordinates exceed a preset threshold, redundant location points are marked to obtain a set of redundant blind zones. Based on the blind zone redundancy set, multi-source sensor signals are fused and the signal strength is integrated through a weighted average method to optimize the location layout scheme and obtain the adjusted distribution characteristics. By adjusting the distribution characteristics, the evolution trend is analyzed and the changes are tracked using time series methods to obtain the non-uniform field distribution characteristics and potential blind zone redundancy locations.

3. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The feature extraction module extracts key stress distribution parameters from the non-uniform field distribution features. If the parameters exceed a preset threshold, they are marked as risk concentration areas. The candidate set of monitoring locations includes: Key stress distribution parameters are obtained from the non-uniform field distribution characteristics. Intensity values ​​are extracted from multi-source sensor signals through deformation data processing. The intensity values ​​are integrated using a weighted average method to obtain the parameter feature vector. For the parameter feature vector, a threshold over-limit judgment is used. If it exceeds the preset threshold, the risk area is marked to determine the concentrated area identification result. Based on the results of centralized area identification, blind spot location calculation is performed, coordinate values ​​are extracted from distribution features and redundant points are identified, and an optimized location set is obtained to get a monitoring candidate set. Trend change tracking indicators are extracted from the monitoring candidate set, and multi-source signals are integrated from the adjusted layout through signal strength integration to obtain the monitoring distribution of risk-concentrated areas.

4. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The signal quality assessment module, for the candidate set of monitoring locations, obtains the signal-to-noise ratio and integrity data quality indicators of the sensor signals, and uses a clustering algorithm to prioritize the monitoring points to obtain dynamic priority assessment results, including: The signal-to-noise ratio (SNR) of sensor signals is obtained from the candidate set of monitoring locations. Noise interference values ​​are extracted from multi-source data by calculating the signal strength to obtain the quantitative index of SNR. To integrate the signal-to-noise ratio quantization index with the integrity data quality index, a weighted fusion method is used to integrate the signal missing rate and continuity measure to obtain a comprehensive quality assessment vector. Features of monitoring points are extracted from the comprehensive quality assessment vector, and the features are grouped using the K-means clustering algorithm to obtain a priority group set; A dynamic adjustment factor is calculated for the priority group set, and the real-time update value is obtained from the signal change trend. If the update value exceeds the preset threshold, the group boundary is adjusted to obtain a dynamic priority sequence. By fusing the coordinates of risk-concentrated areas from the dynamic priority sequence and integrating location optimization data using a coordinate mapping method, dynamic priority assessment results are obtained.

5. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The layout optimization module, based on the dynamic priority evaluation results and blind zone redundancy locations, uses a genetic algorithm with coverage and resource consumption as fitness functions to generate a preliminary layout configuration for the full coverage network through multi-point crossover and random mutation, including: The coordinates of the redundant blind zone locations are obtained from the dynamic priority evaluation results. The population is initialized using a genetic algorithm, and the initial fitness value is calculated using the coverage function to obtain the preliminary population set. To address the resource consumption balance index of the initial population fusion, a multi-point crossover operation is used to generate offspring individuals, resulting in a crossover sequence. Random mutation parameters are extracted from the crossover individual sequences. If the mutation rate exceeds a preset threshold, the individual gene loci are adjusted to obtain the mutation optimization set. For the mutation optimization set, the location optimization data is integrated, and the fitness of the ranked individuals is evaluated to obtain the ranked list of individuals; The initial coordinates of the network layout are merged from the sorted list of individuals, and a full-coverage layout is generated using a coordinate mapping method to obtain the initial layout configuration.

6. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The iterative adjustment module simulates the coverage effect of a full-coverage network under a non-uniform field through initial layout configuration. If the coverage rate is lower than a preset threshold, the updated layout configuration is obtained by iteratively adjusting the monitoring point positions, including: Non-uniform field simulation data are obtained from the initial layout configuration. The coverage effect evaluation method is used to calculate the current coverage rate by comparing the coverage area of ​​the monitoring point with the total area. If the coverage rate is lower than the preset threshold, the iterative position adjustment parameters are extracted to obtain the coordinates of the monitoring point after adjustment. For the redundant coordinates in the blind zone of the adjusted monitoring point coordinate fusion, an optimization sequence is generated by weighted summation through the resource consumption index balancing process, and the random variation parameters in the sequence are determined to update the location data. The fitness function value is extracted from the updated location data. The multi-point crossover sequence method is used to integrate the monitoring point optimization information with the location data as input to obtain the crossover layout set. The application of coverage threshold judgment to the cross-layout set is used to obtain the mutation parameters. Random values ​​are extracted from the sequence to generate the final configuration, resulting in the updated layout configuration.

7. The real-time analysis system for mine roadway deformation data based on edge computing according to claim 1, characterized in that: The trend prediction module extracts deformation monitoring data streams from the updated layout configuration, uses time series analysis to predict roadway deformation trends, and determines potential risk evolution paths, including: Deformation monitoring data streams are obtained from the updated layout configuration. Time series analysis is used to process the correlation and stationarity of the sequence using an autoregressive integral moving average model with the data stream as input, and the deformation trend of the roadway is obtained. To address the deformation trend of the tunnel, geological stability indicators are integrated. These indicators extract rock strength and stress distribution values ​​from a pre-set geological database and integrate potential risk factors using a weighted summation method. The potential risk factors are then weighted by historical deformation records and environmental variables to determine the risk evolution path. Abnormal fluctuation sequences are extracted from the risk evolution path, and an alarm signal is generated if the fluctuation exceeds a preset threshold using a threshold comparison method, thus obtaining an emergency response sequence. For emergency response sequences, equipment maintenance logs are integrated. The equipment maintenance logs obtain maintenance time and fault type from the historical records of the monitored equipment. The similarity between the log sequence and the response sequence is compared through a sequence matching process to identify matching items, obtain optimization and adjustment parameters, and determine potential risk evolution paths.

Images