Multi-source data driven comprehensive evaluation method and system for waterproof performance of coal seam floor

Abstract

The invention discloses a multi-source data driven comprehensive evaluation method and system for the water insulation performance of a coal seam floor, and relates to the technical field of water insulation performance evaluation.Time sequence pressure values of all monitoring points are mapped to a unified two-dimensional space grid, distance weighting processing is carried out based on disturbance diffusion characteristics, and the water insulation performance of the coal seam floor is evaluated. And a disturbance stress field layer with both time resolution and space coverage is constructed. The modeling mode breaks through a traditional stress analysis method based on single-point or single-linear interpolation, so that disturbance response has continuous expression ability on the regional scale, subsequent dynamic identification and structure situation awareness are facilitated, and the method is particularly suitable for a coal seam floor under irregular point distribution and heterogeneous formation conditions. By constructing disturbance layer sequences of different time points and introducing a disturbance change rate function to capture disturbance intensity changes between adjacent time points, the whole process of'dynamic attenuation-rebound-interference 'of a disturbance field is monitored.

Description

Technical Field

[0001] This invention relates to the field of water-tightness performance evaluation technology, specifically to a multi-source data-driven comprehensive evaluation method and system for the water-tightness performance of coal seam floor. Background Technology

[0002] The water-tightness of the coal seam floor is a key concern, especially in areas with multiple water systems and fractured zones. Floor instability and leakage can trigger water and mud inrushes, goaf flooding, and even endanger the lives of underground workers. Therefore, continuous time-series monitoring and trend analysis of the stress field response of the floor under mining disturbances are crucial for improving the reliability of dynamic assessments and guiding scheduling decisions. The construction and indexation of the "disturbance residual field" provides a key transition mechanism from raw sensor signals to comprehensive stability assessments for this monitoring.

[0003] Current practices for assessing the water-tightness of coal seam floors generally rely on static analysis of data from single moments or sections, such as borehole hydrostatic tests, empirical methods for water inrush coefficients, or instantaneous stress monitoring to determine local risks. While these methods offer some guidance in local or short-term contexts, they fail to reflect the "attenuation trend" and "cumulative residual response" of the floor structure under continuous mining disturbances. Many water inrush events do not occur at the moment of maximum disturbance but are induced before residual stress is fully released and the disturbance field stability has recovered. This indicates that traditional single-point, static index assessments lack the ability to observe the "disturbance recovery process."

[0004] The root cause of these deficiencies lies in the lack of a mechanism that can comprehensively analyze the temporal evolution trend of the floor disturbance field in conjunction with the spatial stress residual distribution. The disturbance to the floor caused by mining activities is continuous and non-uniform, and its response process is not linear, often influenced by the superposition of structural interfaces, fault zones, and aquifers. Without quantitative analysis of the "residualness" and "recovery rate" of the disturbance field, the system is prone to the following problems: First, over-reliance on local outliers leads to misjudgments of risk levels; second, neglecting the hysteresis of disturbances results in misjudging stability before the disturbance trend fully converges, leading to missed detection of potential water inrush points; third, the inability to distinguish the differences in recovery capabilities of disturbances in different regions results in the failure to implement differentiated interventions based on regional variations during scheduling. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-source data-driven comprehensive evaluation method and system for the water-impermeable performance of coal seam floor, solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a multi-source data-driven comprehensive evaluation system for the water-impermeable performance of coal seam floor, comprising a multi-source pressure data acquisition and processing module, a disturbance residual stress field layer reconstruction module, a disturbance field temporal difference extraction module, a disturbance trend modeling and fitting module, a disturbance stability index calculation module, and a water-impermeable performance classification and risk layer output module.

[0007] The multi-source pressure data acquisition and processing module acquires raw pressure data (yPa) through sensors installed in the coal seam floor area, and performs preprocessing to obtain time-series pressure values ​​(Pa).

[0008] The perturbation residual stress field layer reconstruction module projects the time series data of time series pressure values ​​Pa onto a two-dimensional spatial grid, and reconstructs the perturbation stress field function Rf(x, y, tk) at each specified time point tk, forming a continuous perturbation layer;

[0009] The disturbance field temporal difference extraction module compares the continuous disturbance layers frame by frame, extracts the residual amplitude of the disturbance field change, and obtains the disturbance change curve ΔRres.

[0010] The disturbance trend modeling and fitting module smooths and models the acquired disturbance change curve ΔRres to form a continuous disturbance trend curve Rtr.

[0011] The disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr to calculate the disturbance residual time series stability index Ψres.

[0012] The waterproof performance grading and risk layer output module binds the acquired disturbance residual time-series stability index Ψres with the spatial region to form a dynamic waterproof performance grading layer Πres(x,y) and makes decisions.

[0013] Preferably, the multi-source pressure data acquisition and processing module includes a raw pressure data structure construction and anomaly removal unit and a normalization standard conversion unit for disturbance response characteristics;

[0014] The raw pressure data structure construction and anomaly removal unit collects pressure data and obtains raw pressure data yPa by using pressure sensors installed at various monitoring points in the coal seam floor area;

[0015] The raw pressure data yPa of each monitoring point is processed by using the non-center offset slope scanning method to obtain the local slope anomaly rate Ys, and compared with the preset anomaly rate threshold Tys to process the data of the monitoring point.

[0016] The local slope anomaly rate Ys is obtained using the following formula:

[0017] ;

[0018] In the formula, Ys(tk) represents the local slope anomaly rate at time point tk, yPa(tk) represents the original pressure data at time point tk, pyPa[tk-ts, tk+ts] represents the local average pressure value, specifically the average pressure value within each ts time step before and after tk, and ts represents the time point.

[0019] When the local slope anomaly rate Ys > the anomaly rate threshold Tys, the monitoring point is identified as an anomaly and is cleared.

[0020] When the local slope anomaly rate Ys ≤ the anomaly rate threshold Tys, it indicates that the monitoring point is normal and is retained.

[0021] Preferably, the normalization standard conversion unit of the disturbance response characteristics performs dimensional unification processing on the original pressure values ​​of different measuring points. By using the disturbance sensitive-equilibrium normalization mechanism, the original pressure data yPa of each monitoring point is uniformly converted into time-series pressure values ​​Pa with consistent response.

[0022] The amplitude of the local disturbance response at the monitoring point is extracted to obtain the disturbance response amplitude Ω;

[0023] The disturbance response amplitude Ω is obtained by the following formula:

[0024] ;

[0025] In the formula, Ωi represents the disturbance response amplitude of the i-th monitoring point, yPai(t) represents the original pressure data of the i-th monitoring point at time t, pyi represents the average pressure value of the i-th monitoring point, and max(t) represents the maximum value at time t.

[0026] The original pressure data yPa is normalized based on the obtained disturbance response amplitude Ω to obtain the time-series pressure value Pa.

[0027] The time-series pressure value in Pa is obtained using the following formula:

[0028] ;

[0029] In the formula, Pai(t) represents the time-series pressure value of the i-th monitoring point at time t, and es represents a non-zero constant.

[0030] Preferably, the disturbance residual stress field layer reconstruction module includes a spatial pressure projection and mesh interpolation unit and a disturbance diffusion fitting and dynamic layer reconstruction unit;

[0031] Spatial pressure projection and grid interpolation units map temporal pressure values ​​Pa to a two-dimensional geographic grid according to spatial location, constructing the original grid pressure field at each time point;

[0032] First, the pressure values ​​Pa at all monitoring points (xi, yi) are projected onto a unified spatial grid (x, y).

[0033] Next, based on spatial distance and disturbance attenuation characteristics, a local diffusion function with the monitoring point as the core is introduced to perform weighted modeling of the target grid points, and the disturbance stress field function Rf is reconstructed at each specified time point tk;

[0034] The perturbation stress field function Rf is obtained through the following formula:

[0035] ;

[0036] In the formula, Rf(x, y, tk) represents the perturbation stress field function of the spatial grid (x, y) at time point tk, N represents the total number of monitoring points, Pa(i, tk) represents the time-series pressure value of the i-th monitoring point at time point tk, the coordinates of the i-th monitoring point are (xi, yi), exp represents the exponential function, Di(x, y) represents the Euclidean distance from the i-th monitoring point to the target spatial grid (x, y), and Ld represents the perturbation attenuation scale constant;

[0037] The perturbation diffusion fitting and dynamic layer reconstruction unit constructs a static perturbation field layer sequence based on multiple time points, and uses a time evolution modeling strategy to construct a continuous perturbation layer;

[0038] Define the perturbation rate of change function Vrf for each spatial grid (x, y) to capture the dynamic evolution of the perturbation field;

[0039] The perturbation rate of change function Vrf is obtained by the ratio of the difference between the perturbation stress field function Rf at time point tk and the perturbation stress field function Rf at time point tk-1 to the time interval.

[0040] Dynamic evolution is obtained through matching in the following ways:

[0041] When the rate of change of the disturbance function Vrf is greater than the preset disturbance threshold Tvr, it indicates that an anomaly has occurred at the time point, and it is not included as part of the continuous disturbance layer and is removed.

[0042] When the perturbation rate of change function Vrf ≤ the preset perturbation threshold Tvr, it indicates that the time point is normal and is retained to form a continuous perturbation layer {Rf(x, y, t1), Rf(x, y, t2), ...}.

[0043] Preferably, the perturbation field time series difference extraction module includes a layer difference profile generation unit and a perturbation residual time series construction unit;

[0044] The layer difference profile generation unit compares the acquired continuous perturbation layers frame by frame, and extracts the perturbation change curve ΔRres of the perturbation difference on the entire region A based on the perturbation change rate function Vrf.

[0045] The disturbance variation curve ΔRres is obtained using the following formula:

[0046] ;

[0047] In the formula, ΔRres(tk) represents the perturbation change curve at time point tk, the perturbation change rate function Vrf(x,y,tk) represents the perturbation change rate function at the spatial grid (x,y) at time point tk, A represents the spatial area of ​​the monitoring area, and d represents the derivative sign;

[0048] The perturbation residual time series construction unit uses a sliding window mechanism to construct the perturbation change curve ΔRres over a continuous time period into a complete perturbation evolution curve sequence DR;

[0049] The formula for the perturbation evolution curve sequence DR is as follows: {ΔRres(t1), ΔRres(t2), ..., ΔRres(tk)}

[0050] The perturbation change curves ΔRres at adjacent time points are analyzed to calculate the higher-order differential sensitivity index Sdyn.

[0051] The higher-order differential sensitivity index Sdyn is obtained by taking the absolute value of the difference between the perturbation change curve ΔRres at time point tk and the perturbation change curve ΔRres at time point tk-1.

[0052] The obtained high-order differential sensitivity index Sdyn is compared with the preset sensitivity threshold Tdyn to determine the interference state.

[0053] When the higher-order differential sensitivity index Sdyn ≥ sensitivity threshold Tdyn, it indicates that the disturbance change curve suddenly increases, an anomaly occurs, and a disturbance state is initiated.

[0054] When the higher-order differential sensitivity index Sdyn < the sensitivity threshold Tdyn, it indicates that the disturbance change curve is stable and there is no abnormal state.

[0055] Preferably, the disturbance trend modeling and fitting module includes a disturbance residual smoothing and denoising analysis unit and a multi-parameter nonlinear trend fitting unit;

[0056] The disturbance residual smoothing and denoising analysis unit uses a window-weighted smoothing algorithm to smooth the disturbance change curve ΔRres at each time point to obtain the smoothed disturbance residual value nRre;

[0057] The smoothed perturbation residual value nRre is obtained by the following formula:

[0058] ;

[0059] In the formula, nRre(tk) represents the smoothed perturbation residual value at time point tk, w represents the half width of the sliding window, j represents the index change, ΔRres(tj) represents the perturbation change curve at time point tj, and ηj represents the weighting factor, which specifically represents the weighting coefficient of the perturbation value corresponding to time point tj. It is usually related to the perturbation stationarity or the difference with the center point. The higher the value, the higher the confidence level.

[0060] The formula for obtaining the weighting factor ηj is as follows: ηj = exp(-λ × |ΔRres(tj) - ΔRres(tk) |);

[0061] In the formula, exp represents the exponential function, and λ represents the disturbance amplitude sensitivity coefficient.

[0062] Preferably, the multi-parameter nonlinear trend fitting unit performs nonlinear trend modeling on the smoothed perturbation residual value nRre to generate a differentiable trend function Rtr;

[0063] The formula for obtaining the trend function Rtr is as follows: Rtr(t) = a1×exp(-a2×t)+a3;

[0064] In the formula, Rtr(t) represents the trend function at time t, a1 represents the initial disturbance amplitude, a2 represents the disturbance decay rate, a3 represents the long-term disturbance residual steady-state value, and exp represents the exponential function;

[0065] The obtained trend function Rtr is differentiated and analyzed. When the value of the derivative is negative, it indicates that the disturbance trend is gradually decreasing over time. The absolute value represents the rate at which the disturbance subsides per unit time.

[0066] The obtained trend function Rtr is combined with the smoothed perturbation residual value nRre to calculate the fitting error evaluation value Efit.

[0067] The fitting error evaluation value Efit is obtained using the following formula:

[0068] ;

[0069] In the formula, K represents the total number of time points, and Rtr(tk) represents the trend function at time point tk;

[0070] Analyze the obtained fitting error evaluation value Efit to obtain the perturbation trend of the current data;

[0071] When the fitting error evaluation value Efit ≤ 0, it indicates that the current data is stable;

[0072] When the fitting error evaluation value Efit > 0, it indicates that the current data has an irregular perturbation trend, and an instability factor needs to be introduced in the stability evaluation.

[0073] Preferably, the disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr within a fixed period to calculate the disturbance residual time series stability index Ψres.

[0074] The residual temporal stability index Ψres is obtained using the following formula:

[0075] ;

[0076] In the formula, Ψres(t) represents the disturbance residual time series stability index at time t, Tw represents the integration window length, and d represents the integration sign;

[0077] The obtained residual temporal stability index Ψres of the disturbance is analyzed to determine the state of the disturbance;

[0078] The disturbance state is obtained by matching in the following way:

[0079] When the residual temporal stability index Ψres ≤ 0.015, it indicates that the disturbance state remains unchanged and the structure tends to be stable.

[0080] When 0.015 < perturbation residual time-series stability index Ψres ≤ 0.05, it indicates that the perturbation state is in a substeady state and there is micro-perturbation activity.

[0081] When 0.05 < the disturbance residual time-series stability index Ψres, it indicates that the disturbance state is active and the structure is unstable.

[0082] Preferably, the waterproof performance grading and risk layer output module binds the disturbance residual time series stability index Ψres to the spatial region, obtaining the stability index time series from each monitoring point;

[0083] The formula is as follows: Ψres,i(tk), i=1,2,...,N;

[0084] In the formula, Ψres,i(tk) represents the disturbance residual time series stability index of the i-th monitoring point at time point tk, and N represents the total number of monitoring points;

[0085] Based on the ideas of spatial attenuation and distance weighting, a continuous stability field function is constructed by using the diffusion projection method on the stability index time series, and the dynamic water-tightness performance classification layer Πres(x,y) is obtained.

[0086] The dynamic waterproofing performance grading layer Πres(x,y) is obtained using the following formula:

[0087] ;

[0088] In the formula, Di(x,y) represents the Euclidean distance from the i-th monitoring point to the target spatial grid (x,y), and Ls represents the diffusion scale parameter;

[0089] The obtained dynamic waterproof performance classification layer Πres(x,y) is mapped and compared with the preset classification threshold Tres to divide each region into different waterproof levels.

[0090] The division method is as follows:

[0091] When the dynamic water-tightness performance grading layer Πres(x,y) ≤ grading threshold Tres, it indicates the first water-tightness level, a highly stable state; the decision is: normal tunneling, and mining is allowed in the goaf.

[0092] When the grading threshold Tres < dynamic waterproofing performance grading layer Πres(x,y) ≤ grading threshold Tres × 1.5, it indicates the second waterproofing level, a medium stable state; the decision is: reduce the rate and increase support; retest after 24 hours.

[0093] When the grading threshold Tres×1.5 < the dynamic waterproofing performance grading layer Πres(x,y), it indicates the first waterproofing level, a low stability state; the decision is: stop the operation and install a waterproof wall or reinforce with grouting.

[0094] A multi-source data-driven comprehensive evaluation method for the water-tightness performance of coal seam floor includes the following steps:

[0095] Step 1: The multi-source pressure data acquisition and processing module acquires raw pressure data (yPa) through sensors installed in the coal seam floor area, and performs preprocessing to obtain time-series pressure values ​​(Pa).

[0096] Step 2: The perturbation residual stress field layer reconstruction module projects the time series data of the time series pressure value Pa onto a two-dimensional spatial grid, and reconstructs the perturbation stress field function Rf(x, y, tk) at each specified time point tk, forming a continuous perturbation layer;

[0097] Step 3: The disturbance field temporal difference extraction module compares the continuous disturbance layers frame by frame, extracts the residual amplitude of the disturbance field change, and obtains the disturbance change curve ΔRres.

[0098] Step 4: The disturbance trend modeling and fitting module smooths and models the acquired disturbance change curve ΔRres to form a continuous disturbance trend curve Rtr.

[0099] Step 5: The disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr to calculate the disturbance residual time series stability index Ψres.

[0100] Step Six: The waterproof performance grading and risk layer output module binds the acquired disturbance residual time-series stability index Ψres with the spatial region to form a dynamic waterproof performance grading layer Πres(x,y), and makes a decision.

[0101] This invention provides a multi-source data-driven comprehensive evaluation method and system for the water-tightness performance of coal seam floor, which has the following beneficial effects:

[0102] (1) During system operation, by mapping the time-series pressure values ​​of each monitoring point to a unified two-dimensional spatial grid and performing distance-weighted processing based on the disturbance diffusion characteristics, a disturbance stress field layer with both time resolution and spatial coverage is constructed. This modeling method breaks through the traditional stress analysis method based on single-point or single-linear interpolation, enabling the disturbance response to have continuous expression capabilities at the regional scale, which is convenient for subsequent dynamic identification and structural situation perception, and is particularly suitable for coal seam floor under irregular point layout and heterogeneous strata conditions.

[0103] By constructing a sequence of perturbation layers at different time points and introducing a perturbation rate of change function to capture the changes in perturbation intensity between adjacent time points, the entire process of "dynamic decay-bounce-interference" in the perturbation field was monitored. In particular, by combining perturbation rate of change thresholds with layer retention and removal, a set of perturbation layers retaining only the true evolution process was constructed, effectively removing spurious changes caused by sensing delays and abrupt interference. This mechanism makes the evolution of the perturbation field no longer a passive observation, but a controllable, identifiable, and fittable dynamic behavior.

[0104] (2) By using a window-weighted smoothing algorithm in the perturbation residual smoothing and denoising analysis unit, and introducing a weight function with the perturbation amplitude difference as the kernel, the interference of local abnormal fluctuations and instantaneous spikes in the monitoring data on trend judgment is effectively reduced. When the original perturbation change curve ΔRres has significant non-stationarity and abrupt disturbances, this processing method ensures that the input of the trend model is more stable, continuous and controllable, and solves the fundamental problem of "biasedness caused by noise" in traditional fitting.

[0105] This module introduces three parameters—initial disturbance value, decay rate, and residual steady-state value—through a nonlinear exponential function model to construct a continuous and differentiable disturbance trend function. This function not only reflects the full-cycle trend of disturbance evolution but also supports the quantitative evaluation of disturbance decay speed through first-order derivative analysis, truly realizing the leap from "outcome observation" to "process analysis," and laying the model foundation for dynamic stability analysis and future disturbance prediction.

[0106] (3) The stability index values ​​of each monitoring point are mapped to a two-dimensional spatial grid through a diffusion projection mechanism to form a dynamic stability field covering the entire bottom plate area, and a "dynamic water-tightness performance classification layer" is constructed. This method realizes the information dimension enhancement from "point measurement data" to "regional visualization", which significantly enhances the spatial perception capability of risk monitoring. Based on the construction of the stability field, a classification threshold is set and the stability field is compared with it in segments, thereby automatically generating risk zones of the corresponding level, and giving different operation instructions and risk countermeasures in combination with the tunneling status. This mechanism enables the system to have an integrated linkage capability from data identification, spatial positioning to operation decision-making, which is closer to the practical needs of the engineering site.

[0107] (4) The disturbance residual index is bound to a specific spatial grid region, and a spatial distribution map Πres(x,y) of Ψres values ​​is output in the form of a dynamic layer to realize the coupled expression of disturbance stability and geographical region. At the same time, this layer can be used to automatically compare with preset classification thresholds, quickly divide risk level areas, and provide regionalized visual decision support for tunneling safety. In the data preprocessing stage, a non-central slope anomaly scanning mechanism is introduced to identify and remove data mutation anomalies, which improves the stability and reliability of subsequent disturbance evolution modeling. At the same time, the disturbance response amplitude Ω is introduced as a unified dimensional standard to realize the standard conversion of response characteristics between different measurement points, which provides a basic guarantee for multi-source data fusion. Attached Figure Description

[0108] Figure 1 This is a flowchart illustrating the multi-source data-driven comprehensive evaluation system for the water-tightness performance of coal seam floor according to the present invention.

[0109] Figure 2 This is a schematic diagram illustrating the steps of a multi-source data-driven comprehensive evaluation method for the water-tightness performance of coal seam floor according to the present invention. Detailed Implementation

[0110] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0111] Example 1

[0112] This invention provides a multi-source data-driven comprehensive evaluation system for the water-tightness performance of coal seam floor. Please refer to [link / reference]. Figure 1It includes a multi-source pressure data acquisition and processing module, a disturbance residual stress field layer reconstruction module, a disturbance field time series difference extraction module, a disturbance trend modeling and fitting module, a disturbance stability index calculation module, and a water-tightness performance classification and risk layer output module.

[0113] The multi-source pressure data acquisition and processing module acquires raw pressure data (yPa) through sensors installed in the coal seam floor area, and performs preprocessing to obtain time-series pressure values ​​(Pa).

[0114] The perturbation residual stress field layer reconstruction module projects the time series data of time series pressure values ​​Pa onto a two-dimensional spatial grid, and reconstructs the perturbation stress field function Rf(x, y, tk) at each specified time point tk, forming a continuous perturbation layer;

[0115] The disturbance field temporal difference extraction module compares the continuous disturbance layers frame by frame, extracts the residual amplitude of the disturbance field change, and obtains the disturbance change curve ΔRres.

[0116] The disturbance trend modeling and fitting module smooths and models the acquired disturbance change curve ΔRres to form a continuous disturbance trend curve Rtr.

[0117] The disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr to calculate the disturbance residual time series stability index Ψres.

[0118] The waterproof performance grading and risk layer output module binds the acquired disturbance residual time-series stability index Ψres with the spatial region to form a dynamic waterproof performance grading layer Πres(x,y) and makes decisions.

[0119] In this embodiment, by constructing a disturbance residual stress field layer and a disturbance trend change curve, the concept of temporal evolution of the disturbance process is introduced, overcoming the drawback of traditional evaluation methods that rely on a single point value at a certain moment to judge stability. By gradually establishing a disturbance layer sequence, residual trend, and stability integral chain, the system achieves continuous tracking of the base plate disturbance decay process, making the evaluation closer to the actual evolution process of the base plate structure.

[0120] This invention emphasizes the ability to perceive the overall spatial disturbance structure through gridded construction of the disturbance field and layer difference extraction mechanism. This effectively identifies whether a disturbance tends to stabilize over a large area, avoiding over-response or risk misguidance due to local fluctuations, and enhancing the system's robustness in judgment. The system introduces the disturbance residual temporal stability index Ψres, quantifying the previously experience-dependent "stable-unstable" binary boundary logic into a continuously expressible stability numerical range. This index introduces temporal memory capabilities through a sliding integral mechanism, possessing comparability and discriminability of time trends, providing a reliable data-driven basis for subsequent scheduling.

[0121] The final output of the system, the "water-tightness performance grading layer," is a spatial fusion representation of the disturbance stability results, solving the problem that traditional evaluations often fail to translate results into regional operational levels. This layer, combining a spatial diffusion model with a regional grading mechanism, not only achieves a graphical and visual representation of water-tightness capabilities but also directly provides zoning decision support for the scheduling system, enabling the formulation of operational instructions on "which areas can continue mining and which areas need to be avoided or grouted."

[0122] The system architecture of this invention is a closed-loop modular system, with each module possessing independent functions and clear data flow, making it suitable for integrated deployment and real-time operation. From data acquisition to preprocessing, modeling, and judgment output, a complete chain is formed, significantly reducing manual intervention and helping to improve the level of intelligent risk identification in underground mining, while also adapting to the high-frequency online update requirements under dynamic mining conditions.

[0123] Example 2

[0124] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically: the multi-source pressure data acquisition and processing module includes a raw pressure data structure construction and anomaly removal unit and a normalization standard conversion unit for disturbance response characteristics;

[0125] The raw pressure data structure construction and anomaly removal unit collects pressure data and obtains raw pressure data yPa by using pressure sensors installed at various monitoring points in the coal seam floor area;

[0126] The raw pressure data yPa of each monitoring point is processed by using the non-center offset slope scanning method to obtain the local slope anomaly rate Ys, and compared with the preset anomaly rate threshold Tys to process the data of the monitoring point.

[0127] The local slope anomaly rate Ys is obtained using the following formula:

[0128] ;

[0129] In the formula, Ys(tk) represents the local slope anomaly rate at time point tk, yPa(tk) represents the original pressure data at time point tk, pyPa[tk-ts, tk+ts] represents the local average pressure value, specifically the average pressure value within each ts time step before and after tk, and ts represents the time point.

[0130] When the local slope anomaly rate Ys > the anomaly rate threshold Tys, the monitoring point is identified as an anomaly and is cleared.

[0131] When the local slope anomaly rate Ys ≤ the anomaly rate threshold Tys, it indicates that the monitoring point is normal and is retained.

[0132] The normalization standard conversion unit of the disturbance response characteristics performs dimensional unification processing on the original pressure values ​​of different measuring points. By using the disturbance sensitive-equilibrium normalization mechanism, the original pressure data yPa of each monitoring point is uniformly converted into time-series pressure values ​​Pa with consistent response.

[0133] The amplitude of the local disturbance response at the monitoring point is extracted to obtain the disturbance response amplitude Ω;

[0134] The disturbance response amplitude Ω is obtained by the following formula:

[0135] ;

[0136] In the formula, Ωi represents the disturbance response amplitude of the i-th monitoring point, yPai(t) represents the original pressure data of the i-th monitoring point at time t, pyi represents the average pressure value of the i-th monitoring point, and max(t) represents the maximum value at time t.

[0137] The original pressure data yPa is normalized based on the obtained disturbance response amplitude Ω to obtain the time-series pressure value Pa.

[0138] The time-series pressure value in Pa is obtained using the following formula:

[0139] ;

[0140] In the formula, Pai(t) represents the time-series pressure value of the i-th monitoring point at time t, and es represents a non-zero constant.

[0141] In this embodiment, by introducing a non-center offset slope scanning method, this module not only reads and processes the pressure value of each monitoring point, but also further constructs a local trend change model on the time axis. This method can dynamically identify abnormal abrupt changes in the pressure time series, especially "pseudo-abnormal high values" formed under blasting, equipment interference, or instantaneous stress impact, thereby effectively improving the purity of the data source. Compared with the traditional elimination method that relies on absolute thresholds, this implementation considers the local trend background, reducing the probability of mistakenly eliminating real disturbance points.

[0142] To address the amplitude differences at different monitoring points due to variations in burial depth, surrounding rock properties, or sensor sensitivity, this embodiment introduces a normalized scale for each monitoring point based on the disturbance response amplitude. Compared to conventional methods using maximum or range normalization, this mechanism focuses more on the responsiveness of each point within its actual historical disturbance range, ensuring that the normalized time-series pressure values ​​possess a unified scale without obscuring the original response characteristics. This achieves a data standard of "comparable physical responses and unified disturbance trends."

[0143] The time-series pressure value Pa, after slope anomaly cleaning and normalized scaling, possesses good temporal continuity, scale uniformity, and physical response, and can be directly used as the input basis for the perturbation layer reconstruction module. This structural transformation opens up the mapping path from raw data to model data, avoiding the uncertainty brought about by manually setting classification intervals and empirical segmentation standards, and enabling subsequent steps such as perturbation residuals and trend modeling to be based on a unified and standardized data system.

[0144] The entire module is built without altering the existing sensor deployment. It overcomes physical limitations such as sampling noise, response differences, and monitoring blind spots entirely through data structure reconstruction, processing optimization, and algorithm-driven approaches. This soft logic-driven improvement path provides a lightweight solution for upgrading legacy monitoring systems, possessing strong practicality and engineering application value.

[0145] Example 3

[0146] This embodiment is an explanation based on Embodiment 2. Please refer to it. Figure 1 Specifically: the disturbance residual stress field layer reconstruction module includes a spatial pressure projection and mesh interpolation unit and a disturbance diffusion fitting and dynamic layer reconstruction unit;

[0147] Spatial pressure projection and grid interpolation units map temporal pressure values ​​Pa to a two-dimensional geographic grid according to spatial location, constructing the original grid pressure field at each time point;

[0148] First, the pressure values ​​Pa at all monitoring points (xi, yi) are projected onto a unified spatial grid (x, y).

[0149] Next, based on spatial distance and disturbance attenuation characteristics, a local diffusion function with the monitoring point as the core is introduced to perform weighted modeling of the target grid points, and the disturbance stress field function Rf is reconstructed at each specified time point tk;

[0150] The perturbation stress field function Rf is obtained through the following formula:

[0151] ;

[0152] In the formula, Rf(x, y, tk) represents the perturbation stress field function of the spatial grid (x, y) at time point tk, N represents the total number of monitoring points, Pa(i, tk) represents the time-series pressure value of the i-th monitoring point at time point tk, the coordinates of the i-th monitoring point are (xi, yi), exp represents the exponential function, Di(x, y) represents the Euclidean distance from the i-th monitoring point to the target spatial grid (x, y), and Ld represents the perturbation attenuation scale constant;

[0153] The perturbation diffusion fitting and dynamic layer reconstruction unit constructs a static perturbation field layer sequence based on multiple time points, and uses a time evolution modeling strategy to construct a continuous perturbation layer;

[0154] Define the perturbation rate of change function Vrf for each spatial grid (x, y) to capture the dynamic evolution of the perturbation field;

[0155] The perturbation rate of change function Vrf is obtained by the ratio of the difference between the perturbation stress field function Rf at time point tk and the perturbation stress field function Rf at time point tk-1 to the time interval.

[0156] Dynamic evolution is obtained through matching in the following ways:

[0157] When the rate of change of the disturbance function Vrf is greater than the preset disturbance threshold Tvr, it indicates that an anomaly has occurred at the time point, and it is not included as part of the continuous disturbance layer and is removed.

[0158] When the perturbation rate of change function Vrf ≤ the preset perturbation threshold Tvr, it indicates that the time point is normal and is retained to form a continuous perturbation layer {Rf(x, y, t1), Rf(x, y, t2), ...}.

[0159] The perturbation field temporal difference extraction module includes a layer difference profile generation unit and a perturbation residual time series construction unit;

[0160] The layer difference profile generation unit compares the acquired continuous perturbation layers frame by frame, and extracts the perturbation change curve ΔRres of the perturbation difference on the entire region A based on the perturbation change rate function Vrf.

[0161] The disturbance variation curve ΔRres is obtained using the following formula:

[0162] ;

[0163] In the formula, ΔRres(tk) represents the perturbation change curve at time point tk, the perturbation change rate function Vrf(x,y,tk) represents the perturbation change rate function at the spatial grid (x,y) at time point tk, A represents the spatial area of ​​the monitoring area, and d represents the derivative sign;

[0164] The perturbation residual time series construction unit uses a sliding window mechanism to construct the perturbation change curve ΔRres over a continuous time period into a complete perturbation evolution curve sequence DR;

[0165] The formula for the perturbation evolution curve sequence DR is as follows: {ΔRres(t1), ΔRres(t2), ..., ΔRres(tk)}

[0166] The perturbation change curves ΔRres at adjacent time points are analyzed to calculate the higher-order differential sensitivity index Sdyn.

[0167] The higher-order differential sensitivity index Sdyn is obtained by taking the absolute value of the difference between the perturbation change curve ΔRres at time point tk and the perturbation change curve ΔRres at time point tk-1.

[0168] The obtained high-order differential sensitivity index Sdyn is compared with the preset sensitivity threshold Tdyn to determine the interference state.

[0169] When the higher-order differential sensitivity index Sdyn ≥ sensitivity threshold Tdyn, it indicates that the disturbance change curve suddenly increases, an anomaly occurs, and a disturbance state is initiated.

[0170] When the higher-order differential sensitivity index Sdyn < the sensitivity threshold Tdyn, it indicates that the disturbance change curve is stable and there is no abnormal state.

[0171] In this embodiment, by mapping the temporal pressure values ​​of each monitoring point to a unified two-dimensional spatial grid and performing distance-weighted processing based on the disturbance diffusion characteristics, a disturbance stress field layer with both time resolution and spatial coverage is constructed. This modeling method breaks through the traditional stress analysis methods based on single-point or unilinear interpolation, enabling the disturbance response to have continuous expression capabilities at the regional scale, facilitating subsequent dynamic identification and structural situation awareness, and is particularly suitable for coal seam floors under irregular point layout and heterogeneous strata conditions.

[0172] By constructing a sequence of perturbation layers at different time points and introducing a perturbation rate of change function to capture the changes in perturbation intensity between adjacent time points, this embodiment achieves the monitoring of the entire process of "dynamic decay-bounce-interference" in the perturbation field. In particular, by combining the perturbation rate of change threshold for layer retention and removal, a set of perturbation layers that retains only the true evolution process is constructed, thereby effectively removing spurious changes caused by sensing delays and abrupt interference. This mechanism makes the evolution of the perturbation field no longer a passive observation, but a controllable, identifiable, and fittable dynamic behavior.

[0173] This embodiment introduces a "higher-order difference sensitivity index" by continuously recording the disturbance change curve to reflect whether the disturbance change trend has abruptly increased. This mechanism compensates for the slow response of traditional average residual analysis to sharp trend reversals, effectively identifying critical point jumps caused by the accumulation of small disturbances, predicting potential trends before abnormal states occur, and providing a data basis for setting early response windows in the scheduling system.

[0174] By constructing a perturbation evolution sequence using a sliding window approach, the system's ability to stably identify perturbation trends is enhanced, while also providing a continuous and clean input data source for subsequent module fitting. Compared to the traditional method that relies on the difference between two points, this sequence mechanism exhibits higher resistance to fluctuations, spurious changes, and time sensitivity, laying a solid foundation for overall perturbation trend fitting and stability integration.

[0175] Example 4

[0176] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 1 Specifically: the disturbance trend modeling and fitting module includes a disturbance residual smoothing and denoising analysis unit and a multi-parameter nonlinear trend fitting unit;

[0177] The disturbance residual smoothing and denoising analysis unit uses a window-weighted smoothing algorithm to smooth the disturbance change curve ΔRres at each time point to obtain the smoothed disturbance residual value nRre;

[0178] The smoothed perturbation residual value nRre is obtained by the following formula:

[0179] ;

[0180] In the formula, nRre(tk) represents the smoothed perturbation residual value at time point tk, w represents the half width of the sliding window, j represents the index change, ΔRres(tj) represents the perturbation change curve at time point tj, and ηj represents the weighting factor.

[0181] The formula for obtaining the weighting factor ηj is as follows: ηj = exp(-λ × |ΔRres(tj) - ΔRres(tk) |);

[0182] In the formula, exp represents the exponential function, and λ represents the disturbance amplitude sensitivity coefficient.

[0183] The multi-parameter nonlinear trend fitting unit models the nonlinear trend of the smoothed perturbation residual value nRre, generating a differentiable trend function Rtr.

[0184] The formula for obtaining the trend function Rtr is as follows: Rtr(t) = a1×exp(-a2×t)+a3;

[0185] In the formula, Rtr(t) represents the trend function at time t, a1 represents the initial disturbance amplitude, a2 represents the disturbance decay rate, a3 represents the long-term disturbance residual steady-state value, and exp represents the exponential function;

[0186] The obtained trend function Rtr is differentiated and analyzed. When the value of the derivative is negative, it indicates that the disturbance trend is gradually decreasing over time. The absolute value represents the rate at which the disturbance subsides per unit time.

[0187] The obtained trend function Rtr is combined with the smoothed perturbation residual value nRre to calculate the fitting error evaluation value Efit.

[0188] The fitting error evaluation value Efit is obtained using the following formula:

[0189] ;

[0190] In the formula, K represents the total number of time points, and Rtr(tk) represents the trend function at time point tk;

[0191] Analyze the obtained fitting error evaluation value Efit to obtain the perturbation trend of the current data;

[0192] When the fitting error evaluation value Efit ≤ 0, it indicates that the current data is stable;

[0193] When the fitting error evaluation value Efit > 0, it indicates that the current data has an irregular perturbation trend.

[0194] In this embodiment, a window-weighted smoothing algorithm is used in the perturbation residual smoothing and denoising analysis unit, and a weight function with the perturbation amplitude difference as the kernel is introduced. This effectively reduces the interference of local abnormal fluctuations and instantaneous spikes in the monitoring data on trend judgment. When the original perturbation change curve ΔRres has significant non-stationarity and abrupt disturbances, this processing method ensures that the input of the trend model is more stable, continuous, and controllable, solving the fundamental problem of "biasedness caused by noise" in traditional fitting.

[0195] This module introduces three parameters—initial disturbance value, decay rate, and residual steady-state value—through a nonlinear exponential function model to construct a continuous and differentiable disturbance trend function. This function not only reflects the full-cycle trend of disturbance evolution but also supports the quantitative evaluation of disturbance decay speed through first-order derivative analysis, truly realizing the leap from "outcome observation" to "process analysis," and laying the model foundation for dynamic stability analysis and future disturbance prediction.

[0196] Traditional perturbation trends often rely on "visual fitting" or "image slope perception" for judgment, lacking quantitative control over the rationality of the trend. This embodiment establishes a quantitative index system for trend judgment by calculating the fitting error value Efit between the trend function and the smoothed perturbation residual. It also introduces error polarity as the judgment boundary for the stable state of the perturbation, giving the system clear judgment criteria and intervention triggering basis, which helps in the automated decision-making and execution of subsequent control logic.

[0197] Example 5

[0198] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 1Specifically: the disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr within a fixed period to calculate the disturbance residual time series stability index Ψres.

[0199] The residual temporal stability index Ψres is obtained using the following formula:

[0200] ;

[0201] In the formula, Ψres(t) represents the disturbance residual time series stability index at time t, Tw represents the integration window length, and d represents the integration sign;

[0202] The obtained residual temporal stability index Ψres of the disturbance is analyzed to determine the state of the disturbance;

[0203] The disturbance state is obtained by matching in the following way:

[0204] When the residual temporal stability index Ψres ≤ 0.015, it indicates that the disturbance state remains unchanged and the structure tends to be stable.

[0205] When 0.015 < perturbation residual time-series stability index Ψres ≤ 0.05, it indicates that the perturbation state is in a substeady state and there is micro-perturbation activity.

[0206] When 0.05 < the disturbance residual time-series stability index Ψres, it indicates that the disturbance state is active and the structure is unstable.

[0207] The waterproofing performance grading and risk layer output module binds the disturbance residual time-series stability index Ψres to the spatial region, obtaining the stability index time series from each monitoring point;

[0208] The formula is as follows: Ψres,i(tk), i=1,2,...,N;

[0209] In the formula, Ψres,i(tk) represents the disturbance residual time series stability index of the i-th monitoring point at time point tk, and N represents the total number of monitoring points;

[0210] Based on the ideas of spatial attenuation and distance weighting, a continuous stability field function is constructed by using the diffusion projection method on the stability index time series, and the dynamic water-tightness performance classification layer Πres(x,y) is obtained.

[0211] The dynamic waterproofing performance grading layer Πres(x,y) is obtained using the following formula:

[0212] ;

[0213] In the formula, Di(x,y) represents the Euclidean distance from the i-th monitoring point to the target spatial grid (x,y), and Ls represents the diffusion scale parameter;

[0214] The obtained dynamic waterproof performance classification layer Πres(x,y) is mapped and compared with the preset classification threshold Tres to divide each region into different waterproof levels.

[0215] The division method is as follows:

[0216] When the dynamic water-tightness performance grading layer Πres(x,y) ≤ grading threshold Tres, it indicates the first water-tightness level, a highly stable state; the decision is: normal tunneling, and mining is allowed in the goaf.

[0217] When the grading threshold Tres < dynamic waterproofing performance grading layer Πres(x,y) ≤ grading threshold Tres × 1.5, it indicates the second waterproofing level, a medium stable state; the decision is: reduce the rate and increase support; retest after 24 hours.

[0218] When the grading threshold Tres×1.5 < the dynamic waterproofing performance grading layer Πres(x,y), it indicates the first waterproofing level, a low stability state; the decision is: stop the operation and install a waterproof wall or reinforce with grouting.

[0219] In this embodiment, the disturbance residual time-series stability index is introduced as an integral expression index of the continuous trend curve. It can accumulate and judge the total intensity of the disturbance fluctuation range within the sliding time window, so that the stability of the structural response is transformed from "qualitative judgment" to a numerical quantification process of "segment computability and state comparison", which solves the actual bottleneck of "being able to see but not judge" the disturbance trend.

[0220] This embodiment maps the stability index values ​​of each monitoring point to a two-dimensional spatial grid through a diffusion projection mechanism, forming a dynamic stability field covering the entire foundation area and constructing a "dynamic water-tightness grading layer." This method achieves information dimensionality enhancement from "point measurement data" to "regional visualization," significantly improving the spatial perception capability of risk monitoring. Based on the stability field construction, this embodiment sets grading thresholds and compares the stability field with them in segments, thereby automatically generating risk zones of corresponding levels and providing different operation instructions and risk countermeasures based on the tunneling status. This mechanism enables the system to have integrated linkage capabilities from data recognition and spatial positioning to operational decision-making, making it closer to the practical needs of engineering sites.

[0221] This embodiment not only enables spatial grading of the current disturbance stability state, but also supports tracking the temporal evolution trend of floor disturbance stability. Combined with the tunneling timeline and monitoring frequency settings, the system possesses the capability for "daily controllable and periodic early warning" trend monitoring of floor disturbances, providing continuous and stable structural support criteria for dynamic tunneling.

[0222] Example 6

[0223] A multi-source data-driven comprehensive evaluation method for the water-tightness performance of coal seam floor; please refer to [reference needed]. Figure 2 Specifically, it includes the following steps:

[0224] Step 1: The multi-source pressure data acquisition and processing module acquires raw pressure data (yPa) through sensors installed in the coal seam floor area, and performs preprocessing to obtain time-series pressure values ​​(Pa).

[0225] Step 2: The perturbation residual stress field layer reconstruction module projects the time series data of the time series pressure value Pa onto a two-dimensional spatial grid, and reconstructs the perturbation stress field function Rf(x, y, tk) at each specified time point tk, forming a continuous perturbation layer;

[0226] Step 3: The disturbance field temporal difference extraction module compares the continuous disturbance layers frame by frame, extracts the residual amplitude of the disturbance field change, and obtains the disturbance change curve ΔRres.

[0227] Step 4: The disturbance trend modeling and fitting module smooths and models the acquired disturbance change curve ΔRres to form a continuous disturbance trend curve Rtr.

[0228] Step 5: The disturbance stability index calculation module performs a sliding integral on the acquired disturbance trend curve Rtr to calculate the disturbance residual time series stability index Ψres.

[0229] Step Six: The waterproof performance grading and risk layer output module binds the acquired disturbance residual time-series stability index Ψres with the spatial region to form a dynamic waterproof performance grading layer Πres(x,y), and makes a decision.

[0230] In this embodiment, by constructing a complete disturbance trend modeling chain (including layer reconstruction, temporal difference extraction, trend function fitting and sliding integral calculation), the base plate disturbance is transformed from "isolated fragment data" into "continuous evolution curve", realizing the dynamic perception and calculation of the entire process of base plate disturbance stability, and can more accurately identify the residual structural response characteristics after mining disturbance.

[0231] This method binds the residual disturbance index to a specific spatial grid region and outputs a spatial distribution map Ψres(x,y) of the Ψres values ​​in the form of a dynamic layer, achieving a coupled expression of disturbance stability and geographical region. Simultaneously, this layer can be used to automatically compare with preset classification thresholds, quickly delineating risk level areas and providing regionalized visual decision support for tunneling safety. In the data preprocessing stage, this method introduces a non-central slope anomaly scanning mechanism to identify and remove abrupt data mutations, improving the stability and reliability of subsequent disturbance evolution modeling. Furthermore, it introduces the disturbance response amplitude Ω as a unified dimensional standard to achieve standard conversion of response characteristics between different measurement points, providing a fundamental guarantee for multi-source data fusion.

[0232] By using residual difference analysis under a sliding window and a high-order difference sensitivity judgment mechanism, sudden abnormal events in the disturbance process can be quickly identified; and the integral processing of the trend curve Rtr further forms the Ψres value, which improves the disturbance stability from trend image representation to integral numerical quantification, and has good judgment criteria attributes and hierarchical adaptation capabilities.

[0233] This method starts with multi-source data acquisition, goes through disturbance modeling, state identification, index calculation, and layer generation, and finally completes the classification of water-resistant capacity and tunneling strategy suggestions. It constructs a closed-loop processing flow from data to decision-making, and truly has the ability to be deployed in engineering and respond dynamically, meeting the practical needs of systematic and real-time safety assessment of coal mine floor.

[0234] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A comprehensive evaluation system for the water-tightness performance of coal seam floor driven by multi-source data, characterized in that: The method comprises a multi-source pressure data acquisition and processing module, a disturbance residual stress field layer reconstruction module, a disturbance field time series difference extraction module, a disturbance trend modeling and fitting module, a disturbance stability index calculation module, and a water-resisting performance grading and risk layer output module. The multi-source pressure data acquisition and processing module acquires original pressure data yPa through sensors installed in the coal seam floor area and performs preprocessing to obtain time series pressure values Pa. The disturbance residual stress field layer reconstruction module projects the time series data of the time series pressure values Pa to a two-dimensional space grid and reconstructs a disturbance stress field function Rf(x, y, tk) at each specified time point tk to form a continuous disturbance layer. The disturbance field time series difference extraction module compares the continuous disturbance layer frame by frame to extract the residual amplitude of the disturbance field change and obtain a disturbance change curve ΔRres. The disturbance trend modeling and fitting module performs smoothing and trend modeling on the obtained disturbance change curve ΔRres to form a continuous disturbance trend curve Rtr. The disturbance stability index calculation module performs sliding integration on the obtained disturbance trend curve Rtr to calculate a disturbance residual time series stability index Ψres. The water-resisting performance grading and risk layer output module binds the obtained disturbance residual time series stability index Ψres with the spatial region to form a dynamic water-resisting performance grading layer Πres(x, y) and performs decision-making.

2. The multi-source data driven coal floor waterproof performance comprehensive evaluation system according to claim 1, characterized in that: The multi-source pressure data acquisition and processing module comprises an original pressure data structure construction and abnormality removal unit and a disturbance response feature normalization standard conversion unit. The original pressure data structure construction and abnormality removal unit acquires original pressure data yPa through pressure sensors installed at each monitoring point in the coal seam floor area. The original pressure data yPa of each monitoring point is processed using a non-central offset slope scanning method to obtain a local slope abnormality rate Ys, which is compared with a preset abnormality rate threshold Tys to process the data of the monitoring point. The local slope abnormality rate Ys is obtained by the following formula: ； When the local slope abnormality rate Ys is greater than the abnormality rate threshold Tys, the monitoring point is determined to be an abnormal point and is removed. When the local slope abnormality rate Ys is less than or equal to the abnormality rate threshold Tys, the monitoring point is normal and is retained. The disturbance response feature normalization standard conversion unit performs dimensionless processing on the original pressure values of different monitoring points and uniformly converts the original pressure data yPa of each monitoring point to time series pressure values Pa with consistent response through a disturbance sensitive-equalization normalization mechanism.

3. The multi-source data-driven coal floor waterproof performance comprehensive evaluation system according to claim 2, characterized in that: The disturbance response amplitude Ω is obtained by the following formula: ​ ​ ； In the formula, Ωi represents the disturbance response amplitude of the i th monitoring point, yPai(t) represents the original pressure data of the i th monitoring point at time t, pyi represents the average pressure value of the i th monitoring point, and max(t) represents the maximum value at time t. The original pressure data yPa is normalized according to the obtained disturbance response amplitude Ω to obtain a time series pressure value Pa. The time series pressure value Pa is obtained by the following formula: ； In the formula, Pai(t) represents the time series pressure value of the i th monitoring point at time t, and es represents a non-zero constant.

4. The coal seam floor water-resisting performance comprehensive evaluation system driven by multiple source data according to claim 3, characterized in that: The disturbance residual stress field layer reconstruction module includes a spatial pressure projection and grid interpolation unit and a disturbance diffusion fitting and dynamic layer reconstruction unit. The spatial pressure projection and grid interpolation unit maps the time series pressure value Pa to a two-dimensional geographic grid according to the spatial position to construct an original grid pressure field at each time. First, the pressure values Pa of all monitoring points (xi, yi) are projected to a unified spatial grid (x, y); Next, based on the spatial distance and the disturbance attenuation characteristics, a local diffusion function with the monitoring point as the core is introduced to model the target grid point by weighting, and the disturbance stress field function Rf is reconstructed at each specified time point tk. The disturbance stress field function Rf is obtained by the following formula: ； In the formula, Rf(x, y, tk) represents the disturbance stress field function of the spatial grid (x, y) at the time point tk, N represents the total number of monitoring points, Pa(i, tk) represents the time series pressure value of the i th monitoring point at the time point tk, the coordinates of the monitoring point i are (xi, yi), exp represents the exponential function, Di(x, y) represents the Euclidean distance from the i th monitoring point to the target spatial grid (x, y), and Ld represents the disturbance decay scale constant. The disturbance diffusion fitting and dynamic layer reconstruction unit constructs a static disturbance field layer sequence based on multiple time points, and uses a time evolution modeling strategy to construct a continuous disturbance layer. The disturbance change rate function Vrf of each spatial grid (x, y) is defined to capture the dynamic evolution of the disturbance field. The disturbance change rate function Vrf is obtained by the ratio of the difference between the disturbance stress field function Rf at the time point tk and the disturbance stress field function Rf at the time point tk-1 and the time interval; The dynamic evolution is matched and obtained in the following way: When the disturbance change rate function Vrf is greater than the preset disturbance threshold Tvr, it indicates that the time point is abnormal and is not part of the continuous disturbance layer, and is excluded; When the disturbance change rate function Vrf is less than or equal to the preset disturbance threshold Tvr, it indicates that the time point is normal and is retained to form a continuous disturbance layer {Rf(x, y, t1), Rf(x, y, t2), …}.

5. The coal floor waterproof performance comprehensive evaluation system driven by multiple source data according to claim 4, characterized in that: The disturbance field time series difference amount extraction module includes a layer difference profile generation unit and a disturbance residual time series construction unit. The layer difference profile generation unit compares the obtained continuous disturbance layer frame by frame, and based on the disturbance change rate function Vrf, extracts the disturbance change curve ΔRres of the disturbance difference in the whole area A. The disturbance change curve ΔRres is obtained by the following formula: ； In the formula, ΔRres(tk) represents the disturbance change curve at time point tk, the disturbance change rate function Vrf(x, y, tk) represents the disturbance change rate function at the spatial grid (x, y) at time point tk, A represents the spatial area of the monitoring area, and d represents the derivative symbol; The disturbance residual time series construction unit constructs the disturbance change curves ΔRres in the continuous time period into a complete disturbance evolution curve sequence DR by using a sliding window mechanism; The formula of the disturbance evolution curve sequence DR is as follows: {ΔRres(t1), ΔRres(t2),..., ΔRres(tk)} The disturbance change curves ΔRres of adjacent times are analyzed to calculate the high-order difference sensitivity index Sdyn; The high-order difference sensitivity index Sdyn is obtained by taking the absolute value of the difference between the disturbance change curve ΔRres at time point tk and the disturbance change curve ΔRres at time point tk-1; The obtained high-order difference sensitivity index Sdyn is compared with the preset sensitivity threshold Tdyn to determine the interference state; When the high-order difference sensitivity index Sdyn is greater than or equal to the sensitivity threshold Tdyn, it indicates that the disturbance change curve suddenly increases, an abnormality occurs, and the interference state is determined; When the high-order difference sensitivity index Sdyn is less than the sensitivity threshold Tdyn, it indicates that the disturbance change curve is stable and there is no abnormal state.

6. The coal seam floor water-resisting performance comprehensive evaluation system driven by multiple source data according to claim 5, characterized in that: The disturbance trend modeling and fitting module includes a disturbance residual smoothing and denoising analysis unit and a multi-parameter nonlinear trend fitting unit; The disturbance residual smoothing and denoising analysis unit performs smoothing processing on each time point of the disturbance change curve ΔRres by using a window weighted smoothing algorithm to obtain a smoothed disturbance residual value nRre; The smoothed disturbance residual value nRre is obtained by the following formula: ； In the formula, nRre(tk) represents the smoothed disturbance residual value at time point tk, w represents the sliding window half-width, j represents the index variable, ΔRres(tj) represents the disturbance change curve at time point tj, and ηj represents the weight factor; The weight factor ηj is obtained by the following formula: ηj=exp(-λ×|ΔRres(tj)-ΔRres(tk)|); In the formula, exp represents the exponential function, and λ represents the disturbance amplitude sensitivity coefficient.

7. The multi-source data-driven coal floor waterproof performance comprehensive evaluation system according to claim 6, characterized in that: The multi-parameter nonlinear trend fitting unit performs nonlinear trend modeling on the smoothed disturbance residual value nRre to generate a differentiable trend function Rtr; The acquisition formula of the trend function Rtr is as follows: Rtr(t)=a1×exp(-a2×t)+a3; In the formula, Rtr(t) represents the trend function at time t, a1 represents the initial disturbance amplitude, a2 represents the disturbance decay rate, a3 represents the long-term disturbance residual steady-state value, and exp represents the exponential function; The obtained trend function Rtr is differentiated and analyzed. When the value after differentiation is negative, it indicates that the disturbance trend gradually decreases over time, and the absolute value represents the disturbance decay speed per unit time; The obtained trend function Rtr is combined with the smoothed disturbance residual value nRre to calculate a fitting error evaluation value Efit; The fitting error evaluation value Efit is obtained by the following formula: ； In the formula, K represents the total number of time points, and Rtr(tk) represents the trend function at time point tk; The obtained fitting error evaluation value Efit is analyzed to obtain the disturbance trend of the current data; When the fitting error evaluation value Efit≤0, it indicates that the current data is stable; When the fitting error evaluation value Efit>0, it indicates that the current data has irregular disturbance trend.

8. The multi-source data-driven coal floor waterproof performance comprehensive evaluation system according to claim 7, characterized in that: The disturbance stability index calculation module performs sliding integration on the obtained disturbance trend curve Rtr within a fixed period to calculate the disturbance residual time series stability index Ψres; The disturbance residual time series stability index Ψres is obtained by the following formula: ； In the formula, Ψres(t) represents the disturbance residual time series stability index at time t, Tw represents the integral window length, and d represents the integral sign; The obtained disturbance residual time series stability index Ψres is analyzed to determine the state of the disturbance; The disturbance state is matched and obtained by the following method: When the disturbance residual time series stability index Ψres≤0.015, it indicates that the disturbance state is unchanged, and the structure tends to be stable; When 0.015<disturbance residual time series stability index Ψres≤0.05, it indicates that the disturbance state is in a sub-stable state, and there is a micro-disturbance activity; When 0.05<disturbance residual time series stability index Ψres, it indicates that the disturbance state is active, and the structure is unstable.

9. The multi-source data-driven coal floor waterproof performance comprehensive evaluation system according to claim 8, characterized in that: The water resistance performance classification and risk layer output module binds the disturbance residual time series stability index Ψres with the spatial region, and obtains the stability index time series from each monitoring point; The formula is as follows: Ψres,i(tk), i=1, 2, …, N; In the formula, Ψres,i(tk) represents the disturbance residual time series stability index of the i-th monitoring point at time point tk, and N represents the total number of monitoring points; Based on the spatial decay and distance weighting idea, the diffusion projection method is used to construct a continuous stability field function based on the stability index time series, and a dynamic water resistance performance classification layer Πres(x, y) is obtained; The dynamic water resistance performance classification layer Πres(x, y) is obtained by the following formula: ； In the formula, Di(x, y) represents the Euclidean distance from the i-th monitoring point to the target spatial grid (x, y), and Ls represents the diffusion scale parameter; The obtained dynamic water resistance performance classification layer Πres(x, y) is mapped and compared with the preset classification threshold Tres, and each region is divided into different water resistance levels; The division method is as follows: When the dynamic water resistance performance classification layer Πres(x, y)≤classification threshold Tres, it indicates the first water resistance level, high stability state; the decision is: normal excavation, and the goaf is allowed to be mined; When the classification threshold Tres<the dynamic water resistance performance classification layer Πres(x, y)≤the classification threshold Tres×1.5, it indicates the second water resistance level, medium stability state; the decision is: reduce the speed and increase the support; 24h retest; When the classification threshold Tres×1.5<the dynamic water resistance performance classification layer Πres(x, y), it indicates the first water resistance level, low stability state; the decision is: stop operation, set a water resistance wall or grouting reinforcement.

10. A multi-source data-driven coal seam floor water-resisting performance comprehensive evaluation method applied to the multi-source data-driven coal seam floor water-resisting performance comprehensive evaluation system of any one of claims 1-9, characterized in that: The method comprises the following steps: Step one, the multi-source pressure data acquisition and processing module collects the original pressure data yPa through the sensor installed in the coal seam floor area and pre-processes to obtain the time series pressure value Pa; Step two, the disturbance residual stress field layer reconstruction module projects the time series data of the time series pressure value Pa to the two-dimensional space grid and reconstructs the disturbance stress field function Rf(x, y, tk) at each specified time point tk to form a continuous disturbance layer; Step three, the disturbance field time series difference amount extraction module compares the continuous disturbance layer frame by frame to extract the residual amplitude of the disturbance field change and obtain the disturbance change curve ARres; Step four, the disturbance trend modeling and fitting module performs smoothing processing and trend modeling on the obtained disturbance change curve ARres to form a continuous disturbance trend curve Rtr; Step five, the disturbance stability index calculation module performs sliding integration on the obtained disturbance trend curve Rtr to calculate the disturbance residual time series stability index Ψres; Step six, the water-resisting performance grading and risk layer output module binds the obtained disturbance residual time series stability index Ψres with the spatial region to form a dynamic water-resisting performance grading layer Πres(x, y) and makes a decision.

Images