Construction risk assessment method, device, system and equipment based on multi-source data and medium

Abstract

The application provides a construction risk assessment method, device, system, equipment and medium based on multi-source data. The method comprises the following steps: acquiring multi-source data of a construction mine field, and performing time sequence alignment and standardization processing on the multi-source data to obtain a multi-source time sequence matrix; determining residual data of the multi-source time sequence matrix, and extracting a sensor drift component from the multi-source data according to the residual data; performing compensation processing on the multi-source data by using the sensor drift component to obtain compensation data, and performing interpolation repair on distortion data of the compensation data to obtain optimized data; calculating a comprehensive risk index by using the optimized data, real-time dynamic weights corresponding to the optimized data and multi-scale gradients, and determining a construction risk grade according to the comprehensive risk index. The application eliminates data distortion deviation, greatly improves the accuracy of multi-parameter fusion risk assessment, and adapts to the real-time safety warning scene of the underground mining working face.

Description

Technical Field

[0001] This application relates to the field of risk assessment, and in particular to a method, apparatus, system, equipment and medium for construction risk assessment based on multi-source data. Background Technology

[0002] With the continuous promotion and application of mining production technologies and the continuous improvement of management levels, the comprehensive monitoring level of mine production and safety is also improving. Therefore, in addition to considering accelerating the speed and efficiency of certain links in the production process, it is also necessary to conduct on-site safety monitoring and alarms to achieve real-time and accurate supervision of high-risk production lines and key areas, and to ensure the safety of workers.

[0003] To quickly identify construction safety risks in various high-risk areas and ensure the safety of on-site personnel, one common method is to install multiple different sensors in the construction area to collect multiple different detection data (including temperature and humidity, vibration frequency, gas concentration, rock stress, and wind speed, etc.), and then input the multi-source detection data into a pre-trained risk assessment model for risk assessment to determine the real-time construction risks.

[0004] However, the above method has the following technical problems: the sensor is subject to environmental physical interference at the mining site, which can lead to distortion and zero drift, resulting in errors in the detection data. If biased data is directly used for fusion and risk assessment, the assessment will be biased, reduce the accuracy of the assessment, and lead to underestimation or overestimation of the risk. Summary of the Invention

[0005] In view of the aforementioned problems, this application is made to provide a construction risk assessment method, apparatus, system, equipment, and medium based on multi-source data to overcome or at least partially solve the aforementioned problems, including: A construction risk assessment method based on multi-source data, the method comprising: Multi-source data from the construction site is acquired, and the multi-source data is subjected to time-series alignment and standardization to obtain a multi-source time-series matrix. The multi-source data consists of raw time-series data from multiple different sensors. The residual data of the multi-source time series matrix is ​​determined, and the sensor drift component is extracted from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The multi-source data is compensated using the sensor drift component to obtain compensated data, and the distorted data of the compensated data is repaired by interpolation to obtain optimized data. The comprehensive risk index is calculated using the optimized data, the corresponding real-time dynamic weights, and multi-scale gradients, and the construction risk level is determined based on the comprehensive risk index.

[0006] Optionally, determining the residual data of the multi-source time series matrix and extracting the sensor drift component from the multi-source data based on the residual data includes: The multi-source time series matrix is ​​traversed to filter the raw time series data collected by two sensors from the same source at the same time, and the absolute value of the difference between the two raw time series data is taken to obtain the residual data. Several individual drift components are extracted from multiple residual data based on the minimum variance method, wherein the individual drift components are data components of individual sensors that have experienced zero-point drift or compound drift. Each individual drift component is filtered, corrected, and trend-verified to obtain the sensor drift component. The filtering correction is a correction of noise and component values, and the trend verification is a screening process that uses a non-parametric trend test algorithm to calculate the trend significance statistic of the individual drift component for selection.

[0007] Optionally, the extraction of individual drift components from the multiple residual data based on the minimum variance method includes: A residual matrix is ​​formed from several residual data from the same source, and the residual matrix is ​​divided into a global common component and an individual initial component based on the minimum variance method. Determine the global common variance corresponding to the global common component and the individual initial variance corresponding to the individual initial component, respectively. Based on the individual initial variance and the global common variance, the individual initial components that meet the drift conditions are selected as individual drift components.

[0008] Optionally, the step of filtering and correcting each individual drift component and performing trend verification to obtain the sensor drift component includes: The environmental noise components of the mine are determined, and each individual drift component is filtered and corrected based on the environmental noise components of the mine to obtain the individual corrected component; The trend significance statistic for each individual corrected component is calculated using a nonparametric trend test algorithm. Based on the trend significance statistic, individual corrected components that meet the preset statistic values ​​are selected as sensor drift components.

[0009] Optionally, the step of using the sensor drift component to compensate the multi-source data to obtain compensated data, and then interpolating and repairing the distorted data of the compensated data to obtain optimized data, includes: Based on the timestamp of the sensor drift component, the corresponding original monitoring value is determined from the multi-source data, and the original monitoring value is compensated using the sensor drift component to obtain compensated data. Determine the saturation distortion interval of the compensation data, wherein the saturation distortion interval is a data interval in the compensation data in which at least a number of consecutive values ​​reach the saturation threshold within a sliding window; The distorted data in the saturated distortion range is interpolated and replaced to obtain optimized data.

[0010] Optionally, the step of calculating a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and determining the construction risk level based on the comprehensive risk index, includes: The real-time dynamic weights and multi-scale gradients corresponding to the optimization data are determined respectively, wherein the multi-scale gradient is the gradient of the optimization data under different time window scales, and the real-time dynamic weights are weight values ​​assigned according to the gradient of the optimization data. The comprehensive risk index is obtained by nonlinear coupling calculation using the optimized data, the real-time dynamic weights, and the multi-scale gradients. The construction risk level is determined based on the comprehensive risk index and several preset risk interval thresholds.

[0011] A construction risk assessment device based on multi-source data, the device comprising: The acquisition module is used to acquire multi-source data from the construction mine and perform time-series alignment and standardization on the multi-source data to obtain a multi-source time-series matrix, wherein the multi-source data is the original time-series data of multiple different sensors; The determination module is used to determine the residual data of the multi-source time series matrix and extract the sensor drift component from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The optimization module is used to compensate the multi-source data using the sensor drift component to obtain compensated data, and to interpolate and repair the distorted data of the compensated data to obtain optimized data. The evaluation module is used to calculate a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and to determine the construction risk level based on the comprehensive risk index.

[0012] A construction risk assessment system based on multi-source data, the system comprising: a risk assessment platform and several mine monitoring terminals; The risk assessment platform is communicatively connected to several of the mine detection terminals, and the risk assessment platform is applicable to the construction risk assessment method based on multi-source data as described above.

[0013] An apparatus includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the construction risk assessment method based on multi-source data as described above.

[0014] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the construction risk assessment method based on multi-source data as described above.

[0015] This application has the following advantages: In the embodiments of this application, multi-source data from the construction site can be acquired, and the multi-source data can be time-series aligned and standardized to obtain a multi-source time-series matrix. The residual data of the multi-source time-series matrix is ​​determined, and sensor drift components are extracted from the multi-source data based on the residual data. The sensor drift components are used to compensate the multi-source data to obtain compensated data, and the distorted data in the compensated data is interpolated to obtain optimized data. The optimized data, along with its corresponding real-time dynamic weights and multi-scale gradients, are used to calculate a comprehensive risk index, and the construction risk level is determined based on the comprehensive risk index. After acquiring multi-source data, this application can determine the drift components based on the residual data of the multi-source data, and perform compensation and interpolation repair based on the drift components to eliminate data distortion and bias, thereby reducing data distortion and bias and improving the accuracy of risk assessment. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the description of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the steps of a construction risk assessment method based on multi-source data, provided in one embodiment of this application. Figure 2 This is a structural block diagram of a construction risk assessment device based on multi-source data provided in one embodiment of this application; Figure 3 This is a structural block diagram of a construction risk assessment system based on multi-source data provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, features, and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] With the continuous promotion and application of mining production technologies and the continuous improvement of management levels, the comprehensive monitoring level of mine production and safety is also improving. Therefore, in addition to considering accelerating the speed and efficiency of certain links in the production process, it is also necessary to conduct on-site safety monitoring and alarms to achieve real-time and accurate supervision of high-risk production lines and key areas, and to ensure the safety of workers.

[0020] To quickly identify construction safety risks in various high-risk areas and ensure the safety of on-site personnel, one common method is to install multiple different sensors in the construction area to collect multiple different detection data (including temperature and humidity, vibration frequency, gas concentration, rock stress, and wind speed, etc.), and then input the multi-source detection data into a pre-trained risk assessment model for risk assessment to determine the real-time construction risks.

[0021] However, the above method has the following technical problems: the sensor is subject to environmental physical interference at the mining site, which can lead to distortion and zero drift, resulting in errors in the detection data. If biased data is directly used for fusion and risk assessment, the assessment will be biased, reduce the accuracy of the assessment, and lead to underestimation or overestimation of the risk.

[0022] To address the aforementioned issues, the following specific embodiments will provide a detailed description and explanation of a construction risk assessment method, apparatus, system, equipment, and medium based on multi-source data provided in this application.

[0023] To address the technical problem of low accuracy and subsequent bias in evaluation caused by drift in existing multi-source data acquisition technologies, this paper refers to... Figure 1 The diagram shows a flowchart of a construction risk assessment method based on multi-source data provided in an embodiment of this application.

[0024] In one embodiment, the construction risk assessment method based on multi-source data is applicable to a risk assessment platform, which can be set up in the management system of the mine.

[0025] As an example, the construction risk assessment method based on multi-source data may include: S11. Acquire multi-source data from the construction site and perform time-series alignment and standardization on the multi-source data to obtain a multi-source time-series matrix, wherein the multi-source data is the original time-series data from multiple different sensors.

[0026] In one embodiment, multi-source data from the construction site can be acquired. Since multiple different sensors may cause subsequent fusion distortion due to asynchrony, timestamp drift, or inconsistent sampling rates, the sensors can be grouped according to "same source monitoring type and spatial proximity".

[0027] Specifically, read the whole mine sensor configuration table, then group them according to the monitored physical quantities (gas / CO / stress / wind speed), and then divide them into groups of the same source measuring points according to the three-dimensional coordinate distance (≤5m).

[0028] Next, the raw time-series data (including noise, drift, and jump values) of each group of sensors can be read to obtain multi-source data. In practice, the raw ADC digital values ​​can be retrieved according to their respective sampling rates, and the original timestamps can be retained to preserve the original distortion characteristics and avoid premature truncation.

[0029] After obtaining multi-source data, the local clock deviation of each sensor can be detected. Specifically, the arrival time difference of the same physical event at different sensors can be calculated to output clock offsets Δt1, Δt2…Δtn. Next, a unified timestamp recalibration can be performed. Specifically, linear phase correction can be applied to all time series based on the edge gateway's GPS / BeiDou timing.

[0030] For data with different sampling rates, a unified upsampling to the highest frequency can be performed. For low-frequency sensors, Lanczos interpolation upsampling is used to maintain the trend and eliminate frequency mismatch. After alignment, a standardized multi-source time series matrix D(t) = [d1(t), d2(t), ... dn(t)] can be constructed; each row represents one sensor, and each column represents the value at the same absolute time. This unified spatiotemporal matrix structure facilitates subsequent cross-validation.

[0031] In an optional embodiment, a preprocessing log can be output to mark abnormal sampling channels. Specifically, sensors that are always at 0, always at full scale, or disconnected can be marked, without being discarded, to maintain traceability.

[0032] During data acquisition, sensors may experience clock drift or timestamp misalignment. The above processing can achieve multi-source timing alignment and construct a standardized matrix that can be directly used for mutual verification, laying the foundation for subsequent compensation without destroying the original drift characteristics.

[0033] In another alternative embodiment, since traditional fixed benchmarks may be inaccurate, in order to facilitate subsequent determination of whether there is drift in multi-source data, a steady-state benchmark library and a sensor inherent drift characteristic model can be established based on the collected multi-source data.

[0034] In one operating mode, the historical dataset of the full-time sensor time series matrix after timestamp alignment and standardization can be retrieved. Simultaneously, the mine operating log library can be loaded, removing time periods including blasting operations, equipment start-up and shutdown, roadway ventilation adjustment, large-scale personnel operations, and gas extraction switching. Then, a fixed-length sliding window is used to segment the remaining time series, with the window length uniformly set to 30 minutes. For each window, three statistical indicators are calculated: variance of the monitoring data, mean fluctuation amplitude, and maximum difference between adjacent sampling points. A steady-state judgment threshold is preset; when the window variance, fluctuation amplitude, and extreme value difference are all below the corresponding threshold, the window is determined to be a undisturbed steady-state window. All selected steady-state windows are classified and archived according to sensor number and spatial location of the measuring points, removing window samples with disconnections, saturation, or abrupt changes. A grouped steady-state dataset is output, bound to the corresponding underground average temperature and average humidity environmental parameters for the corresponding time period, as the baseline fitting original sample.

[0035] Next, the data can be grouped by individual sensor to extract all steady-state window time-series data of the device. Instead of traditional least-squares linear fitting, the Theil-Sen median regression algorithm is used for baseline fitting. The algorithm calculates the slope of each steady-state sample point pair, taking the median of all slopes as the baseline change slope and the sample median as the basic baseline value. Outliers caused by minor electromagnetic interference and airflow fluctuations within the steady-state window are automatically identified and mitigated; outliers are not included in the baseline-dominant calculation. Combined with the temperature and humidity parameters bound to the window, the baseline is simultaneously fitted to reflect the basic trend of slow drift with temperature and humidity. A preliminary steady-state baseline function for each sensor is output, recording the baseline base value, baseline change slope, and temperature-humidity correlation coefficient. Simultaneously, the fitting residual sequence can be retained for subsequent generation of normal response interval control limits.

[0036] Theil-Sen robust median regression baseline fitting has a much stronger outlier suppression capability compared to least squares, and is suitable for downhole random electromagnetic and gas flow disturbance scenarios. The baseline fitting results are not affected by a small number of outliers.

[0037] Then, the inherent zero-point baseline Bi for each sensor can be extracted. Specifically, the median baseline value obtained from robust fitting and the temperature and humidity correlation correction term can be read; the formula for the inherent zero-point baseline of a single sensor can be constructed as shown in the following equation: Bi= Baselinemedian + a*T + b*H; Where Baselinemedian is the baseline median under no environmental disturbance, T is the ambient temperature, H is the ambient humidity, and a and b are temperature and humidity correction factors.

[0038] Then, the standard zero-point baselines corresponding to different environments are calculated by substituting the typical underground temperature and humidity range (temperature 10℃~40℃, humidity 60%~98%). The factory-designed theoretical zero-point baseline and the inherent baseline measured in the underground field are distinguished and stored for comparison of drift. The corresponding baseline formula and coefficient parameters are bound according to the unique sensor number to form a one-to-one baseline mapping relationship. Invalid baselines of long-term faulty and frequently saturated sensors are removed, and only the valid zero-point baselines of normal equipment are retained. Finally, all inherent zero-point baselines of sensors are temporarily stored in memory, waiting to be entered into the feature library.

[0039] The baseline incorporates a temperature and humidity coupling correction term, breaking away from the traditional fixed zero-point baseline mode. The baseline value is dynamically and adaptively adjusted according to the real-time temperature and humidity downhole.

[0040] After temporarily storing all inherent zero-point baselines of the sensors in memory, a sensor drift velocity feature library can be established. Specifically, the daily steady-state zero-point baseline Bi of each sensor over seven consecutive days can be retrieved and arranged in chronological order. Then, the daily baseline change is calculated using the difference method, and the daily drift velocity is equal to the daily baseline value. The baseline value of the previous day; and the average daily drift speed, maximum single-day drift speed, and drift speed fluctuation range over the past 7 days.

[0041] Next, drift speed levels can be categorized: low-speed slow drift, medium-speed stable drift, and high-speed abnormal aging drift. Simultaneously, the average drift rate, maximum drift amount, and drift level can be stored in a data table using the sensor number as the primary key. Furthermore, the 7-day rolling window baseline data can be periodically updated to dynamically refresh drift speed characteristics and adapt to the long-term aging process of the sensor. Subsequently, an abnormal drift speed judgment threshold can be established; when the real-time drift speed exceeds the historical normal range, the device aging risk is flagged. By quantifying the baseline change rate using a 7-day rolling window, a real-time comparable drift speed quantitative feature library can be constructed, enabling digital differentiation of sensor aging levels.

[0042] Next, a temperature-humidity-drift mapping model can be constructed. Specifically, three sets of data corresponding to all steady-state window samples can be collected: ambient temperature T, ambient humidity H, and baseline drift ΔB. Then, a bivariate polynomial fitting is used to establish the mapping relationship, as shown in the following formula: ΔB = f(T,H) = k1T2 + k2H2 + k3TH + k4T + k5H + k0; Then, the minimum residual constraint is used to solve all the coefficients of the polynomial (including k0, k1, k2...k5) to reduce the model fitting error; finally, the baseline drift is calculated by traversing the entire temperature and humidity range downhole to generate a two-dimensional comparison table of temperature and humidity-drift.

[0043] It can distinguish between two types of drift: environmental drift caused only by changes in temperature and humidity, and zero-point aging drift of sensor hardware; it can bind complete polynomial functions, coefficients, and two-dimensional lookup tables to the corresponding sensor files; and the model can support millisecond-level input of real-time temperature and humidity, and instantly output the corresponding theoretical drift amount.

[0044] The binary polynomial temperature and humidity coupled drift mapping model quantifies the baseline drift under the combined effect of temperature and humidity, while the single variable model cannot reproduce the coupled interference scenario of high humidity and high temperature downhole.

[0045] Additionally, the normal response range bands [UCL, LCL] of the sensor can be generated. Specifically, the entire steady-state window fitting residual sequence can be extracted, and the overall mean μ and standard deviation σ of the residuals can be calculated. Based on the 3σ statistical control limit rule, the dynamic upper and lower control limits can be calculated, as shown in the following formula: Upper bound UCL = Bi + 3σ, lower bound LCL = Bi 3σ; It can also combine a temperature and humidity drift mapping model to correct UCL and LCL in real time. The interval band expands and contracts synchronously with environmental changes, and is not fixed. The interval band represents the normal value fluctuation range of the sensor under conditions of no zero drift and no equipment failure. The real-time monitored value is compared with the [UCL, LCL] interval time sequence. If it exceeds the interval, it is determined that there is abnormal drift. It distinguishes between two types of scenarios: short-term small-amplitude exceedance of the interval (environmental fluctuation) and long-term continuous exceedance of the interval (zero drift). Finally, the dynamic interval calculation formula and the basic standard deviation are bound and stored in the sensor feature file. The 3σ dynamic confidence interval band, which is adjusted synchronously with the temperature and humidity baseline, is different from the traditional fixed upper and lower limit thresholds and adapts to the dynamic changes of the downhole environment.

[0046] In one embodiment, all models can be stored in a real-time callable feature library. Specifically, all model parameters output previously can be collected, including: sensor inherent zero-point baseline formula, temperature and humidity drift polynomial, drift velocity characteristics, dynamic 3σ interval [UCL, LCL], and Theil-Sen fitting coefficients. A high-speed feature library is constructed using a memory key-value pair database, where the key is the sensor's unique identifier and the value is the complete set of drift feature model parameters for the device. All model parameters are then stored using binary serialization compression to reduce memory usage and improve read speed.

[0047] Optionally, a millisecond-level API can be provided for subsequent steps such as residual decomposition and drift identification to retrieve the baseline, interval, and drift model. It can also be configured with a timed automatic update mechanism to re-select steady-state windows and iteratively update the entire model every 24 hours to adapt to sensor aging; or a faulty sensor isolation mechanism can be established to archive faulty device models separately and prevent them from participating in real-time calculations. It can also generate feature library update logs to record the time of each modeling update, the number of samples, and the average change in baseline, thus preserving a record of modeling traceability.

[0048] A high-speed, real-time in-memory key-value pair feature library with integrated binding and storage of all parameters and daily automatic iterative updates, balancing real-time call performance with long-term model validity. It provides a complete set of real-time decision parameters for subsequent steps such as drift separation and distortion identification.

[0049] S12. Determine the residual data of the multi-source time series matrix, and extract the sensor drift component from the multi-source data based on the residual data, wherein the residual data is the data difference between two sensors of the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift.

[0050] In one embodiment, the residual data of the multi-source time series matrix can be determined. As can be seen from the previous analysis, there are multiple sets of sensor data, and the residual data is the difference between the data of two sensors from the same source and from the same group at the same time.

[0051] Next, the sensor drift component of the multi-source data can be determined based on the residual data. This sensor drift component is the data component in which the sensor experiences zero-point drift.

[0052] In an optional embodiment, determining the residual data of the multi-source time series matrix and extracting the sensor drift component from the multi-source data based on the residual data may include the following sub-steps: S121. Traverse the multi-source time series matrix to filter the original time series data collected by two sensors from the same source at the same time, and take the absolute value of the difference between the two original time series data to obtain residual data.

[0053] S122. Extract several individual drift components from the multiple residual data based on the minimum variance method, wherein the individual drift components are data components of individual sensors that have experienced zero-point drift or compound drift.

[0054] S123. Filter and correct each individual drift component and perform trend verification to obtain sensor drift components. The filtering correction is a correction of noise and component values, and the trend verification is a screening process that uses a non-parametric trend test algorithm to calculate the trend significance statistic of the individual drift components.

[0055] In one embodiment, the pairwise residuals of the same source measurement points can be calculated to obtain residual data. Specifically, a multi-source time series matrix D(t) = [d1(t), d2(t)...dn(t)] can be extracted, with n rows (n sensors in the same source group) and T columns (T time series sampling points under a unified timestamp). Then, all sensors in the same source measurement point group can be paired and combined (i, j), where (i≠j), and (i, j) represent the different sensor numbers in the group.

[0056] For the same absolute time t, the original time series values ​​di(t) of sensor i and dj(t) of sensor j are retrieved respectively, and the difference operation di(t) - dj(t) is performed; then the absolute value of the difference is taken to generate the paired residual Rij(t) = |di(t) at that time. dj(t))|, to obtain the residual data.

[0057] The above difference and absolute value calculations can be repeated for each sampling point to generate a complete time-series residual sequence for each pair of sensors. The residual sequence corresponds one-to-one with the original monitoring time-series length and timestamp. At the same time, all pairwise paired residual sequences can be temporarily stored without window mean smoothing or truncating single-point residual data, thus fully preserving the instantaneous residual fluctuation characteristics. Paired sensor numbers, corresponding timestamps, and residual values ​​can also be associated and stored to establish a residual original index table for subsequent matrix construction.

[0058] This application employs a point-by-point instantaneous residual calculation mechanism, which differs from the traditional sliding window average residual algorithm. It fully preserves short-term sudden zero-point drift and instantaneous temperature drift fluctuations, without smoothing out short-term drift mutation signals, thus avoiding the omission of minute, gradual zero-point drifts in subsequent drift identification.

[0059] After calculating the residual data, the data components of several individual sensors that have experienced zero-point drift or compound drift can be identified and extracted from multiple residual data based on the minimum variance method, thereby obtaining several individual drift components.

[0060] Next, each individual drift component can be filtered and corrected, and a trend verification can be performed to obtain the sensor drift component. The filtering correction is a correction of noise and component values, and the trend verification is a screening process that uses a non-parametric trend test algorithm to calculate the trend significance statistic of the individual drift component for selection.

[0061] In one embodiment, extracting individual drift components from the multiple residual data based on the minimum variance method may include the following sub-steps: S1221. Form a residual matrix from several residual data from the same source, and divide the residual matrix into global common components and individual initial components based on the minimum variance method.

[0062] S1222. Determine the global common variance corresponding to the global common component and the individual initial variance corresponding to the individual initial component, respectively.

[0063] S1223. Based on the individual initial variance and the global common variance, select the individual initial components that meet the drift conditions as individual drift components.

[0064] In one embodiment, all pairwise paired residual time-series datasets can be read to determine the total number of sensors n in the same group; then, an n*n two-dimensional square matrix can be defined as the basic structure of the residual matrix, with the row index i and column index j corresponding to two paired sensors respectively.

[0065] Furthermore, at the diagonal position of the matrix (i=j): the residual of the same sensor is always 0 when comparing itself, and the value is uniformly filled with 0; at the off-diagonal position of the matrix (i≠j): the pairwise residual data Rij(t) of the corresponding time t is filled in, and the absolute value property is used to ensure the symmetry of the matrix, that is, Rij(t) = Rji(t).

[0066] Then, generate independent n*n symmetric residual matrices time-by-time according to the time sequence. Each time stamp corresponds to an independent residual matrix, forming a three-dimensional residual dataset {R(t1),R(t2),…,R(tT)}.

[0067] Based on this, timestamp labels and source point group numbers can be added to each residual matrix to establish a matrix time series index; at the same time, all residual matrices can be cached in the high-speed memory read and write area to avoid disk I / O latency affecting real-time computing efficiency.

[0068] Because scattered residual sequences are stored without structure, batch matrix operations are impossible; local pairing of residuals easily misses individual abnormal sensors, making it difficult to identify common distortion patterns globally. This application constructs a fully connected symmetric residual topology matrix that covers all sensor interaction deviation relationships within the same source group, ensuring no pairings are missed. It utilizes matrix symmetry to compress storage overhead and achieves parallel analysis of measurement point deviations across all dimensions.

[0069] After constructing the residual matrix, it is divided into global common components and individual initial components based on the minimum variance method. Specifically, the residual matrix R(t) at the same time can be extracted, and the matrix is ​​preprocessed to be centered. Specifically, the mean of all elements of the matrix can be calculated, and the mean is subtracted from all elements to eliminate the global constant drift. Then, principal component analysis (PCA) combined with the minimum variance constraint is used to perform eigenvalue decomposition on the centered residual matrix to solve for the eigenvalues ​​and corresponding eigenvectors. Next, the eigenvalues ​​can be sorted from largest to smallest. The first principal component corresponds to the variation pattern with the highest variance percentage, which is identified as the global environmental common drift pattern (synchronous shift of all sensors caused by fluctuations in downhole temperature and humidity and overall ventilation). Then, the global common component represented by the first principal component is removed, and the remaining residual components are all attributed to the individual sensor drift components. At the same time, the temporal variance of each sensor's individual component can be calculated, and sensors with variances significantly higher than the group average are marked as having individual distortion. In this way, two independent time-series datasets can be separated and output: one is the global common component of global environmental drift, and the other is the component of deviation of each sensor's individual components, obtaining the initial individual components.

[0070] After dividing the components into two different groups, the two groups of components can be bound to the corresponding timestamps, and the variance ratio of each component can be recorded synchronously for subsequent distortion type determination.

[0071] In underground mines, temperature and humidity drift and sensor zero-point drift are simultaneously superimposed, resulting in a coupling and mixing of these two types of distortion, which traditional algorithms cannot separate. They can only identify overall data drift, not pinpoint individual faulty sensors. This application utilizes the PCA minimum variance constraint bisection method to separate global environmental drift and individual sensor zero-point drift, achieving mathematical decoupling of these two types of coupled distortion and overcoming the technical bottleneck of traditional filtering methods that cannot distinguish between environmental interference and sensor hardware drift.

[0072] After dividing the system into two components, the global common variance corresponding to the global common component and the individual initial variance corresponding to the individual initial components of each sensor can be determined. Then, two-level decision thresholds can be obtained (specifically, these can be generated based on the 3σ dynamic confidence interval of the steady-state benchmark library obtained in the above steps).

[0073] If the global common variance exceeds the threshold and the initial variance of all individual sensors is below the threshold, the distortion type can be determined as global temperature and humidity environmental drift. If the initial variance of a single sensor or a few sensors exceeds the threshold, but the global common variance is within the normal range that meets the threshold, the distortion type can be determined as sensor zero-point drift. If the global common variance exceeds the threshold and the initial variance of a single sensor exceeds the threshold, it can be marked as a composite distortion of "environmental drift plus sensor zero-point drift".

[0074] Each time-series sampling point can be labeled with a distortion type tag: environmental drift, zero-point drift, composite drift, or no distortion; the distortion type tag is then bound and stored with the global component and individual component time-series data. Furthermore, the individual initial component of the sensor labeled as zero-point drift or composite drift is taken as the individual drift component.

[0075] This application employs an automatic classification mechanism linked by two variance thresholds to distinguish between three scenarios: single environmental interference, single sensor zero-point failure, and composite coupling distortion. This differentiates environmental interference from sensor hardware aging and drift, providing a classification basis for subsequent differentiated dynamic compensation. Furthermore, the dynamic thresholds are adaptively adjusted based on a historical steady-state database, rather than using fixed factory thresholds, which further improves the accuracy of the distinction.

[0076] In one embodiment, the step of filtering and correcting each individual drift component and verifying its trend to obtain the sensor drift component may include the following sub-steps: S1231. Determine the environmental noise components of the mine, and filter and correct each individual drift component based on the environmental noise components of the mine to obtain the individual corrected component.

[0077] S1232. Calculate the trend significance statistic for each individual modified component using a nonparametric trend test algorithm.

[0078] S1233. Based on the trend significance statistic value, select the individual correction components that meet the preset statistic value as sensor drift components.

[0079] In one embodiment, after obtaining the individual drift component, the temperature and humidity-drift mapping model of the above steps can be retrieved to determine the mine environmental noise component of the individual drift component, and then the mine environmental noise component in the individual drift component can be removed to eliminate the residual weak environmental noise terms.

[0080] Next, a sliding median filter can be used to remove random high-frequency downhole electromagnetic noise and instantaneous gas flow disturbance noise, retaining only the trend-driven slow drift signal.

[0081] After completing the above operations, difference correction can be performed. Specifically, the short-time random noise component can be subtracted from the denoised individual drift component to obtain the pure zero-point drift value, which is the individual correction component Sdrift(t). In addition, for composite distortion scenarios, the global environment drift component is superimposed and subtracted, retaining only the sensor's own zero-point drift, to obtain the individual correction component Sdrift(t).

[0082] Optionally, an independent point-by-point drift component curve can be generated for each abnormal sensor, with the timestamps fully aligned with the original monitoring data; the sensor number, the drift amount at the corresponding time, and the distortion type label can also be bound and stored in the memory compensation cache.

[0083] The noise layering and stripping algorithm can distinguish high-frequency random noise, global environmental drift, and sensor zero-point trend drift, accurately purifying noise-free zero-point drift components and avoiding data oscillation caused by noise participating in compensation.

[0084] Next, a trend test can be performed on the components. Specifically, after reading the individual modified component Sdrift(t), a continuous time series sample sequence can be constructed; then, the Mann-Kendall (MK) nonparametric trend test algorithm is used to calculate the trend significance statistic Z.

[0085] Then, set the significance level threshold α=0.05. If the absolute value of the trend significance statistic |Z|>1.96, it is determined that the drift component has a significant unidirectional gradual trend and belongs to the true zero-point drift. This individual correction component is the sensor drift component. Conversely, if the absolute value of the trend significance statistic |Z|≤1.96, it is determined that it is only random fluctuation and there is no effective drift.

[0086] Optionally, the least squares fitting drift time series slope can be used simultaneously to quantify the speed of zero-point drift (drift speed); for random fluctuation points without significant trends, they are marked as "invalid drift, no compensation is performed"; for points with significant trends, the drift components are retained for compensation; finally, a trend test result table can be generated, recording the trend significance statistic Z, fitting slope, and whether the drift is valid for each sampling point; the test mark is attached to the drift component time series and sent to the next compensation link simultaneously.

[0087] This application uses the MK nonparametric trend significance test combined with drift slope quantification to distinguish between long-term asymptotic zero-point drift and short-term random disturbances, thus avoiding spurious compensation operations from a statistical perspective.

[0088] In an optional embodiment, invalid drift data without a significant trend can be removed based on trend test markers, retaining the valid sensor drift component Sdrift(t). Then, a drift component output time series table is constructed using the sensor drift component Sdrift(t), with fields including: unified timestamp, sensor number, original monitoring value, pure zero-point drift amount, distortion type, trend test result, and drift velocity. The time axis of the original multi-source time series matrix D(t) is then aligned to ensure that the drift component curve corresponds one-to-one with the original monitoring data sampling points, with no time misalignment or missing segments. At the same time, the drift component curve data structure can be visualized and encapsulated, supporting millisecond-level single-point reading and interval batch reading for easy user viewing.

[0089] Simultaneously, it can generate operation logs for this step, recording the number of sensors involved in the calculation, the identified drift sensor numbers, the average drift amount, and the trend verification pass rate; the logs are bound to the drift component curves for storage, retaining complete traceable evidence of the computation chain.

[0090] Traditional segmented drift compensation suffers from numerical jumps at interval boundaries; the drift amount is not synchronized with the original data in time, resulting in phase shifts during compensation; and there is a lack of traceability records for the calculation process. This application outputs a point-by-point zero-point drift correction curve that is completely synchronized with the original monitoring data, without any breaks or jumps, seamlessly connecting to the real-time dynamic compensation process, while retaining a full-process calculation log to meet the requirements for mine safety traceability.

[0091] S13. The multi-source data is compensated using the sensor drift component to obtain compensated data, and the distorted data of the compensated data is interpolated and repaired to obtain optimized data.

[0092] In one embodiment, real-time dynamic compensation and saturation distortion repair can be performed based on sensor drift components to achieve dynamic correction of the sensor's original data and range saturation distortion repair.

[0093] In one operating mode, after reading the sensor drift components, point-by-point real-time compensation can be performed on multi-source data to obtain compensated data. Next, sensor saturation distortion points can be detected to determine the distorted data in the compensated data. Then, interpolation is performed on the distorted data to obtain optimized data. After compensation and interpolation, the data can be restored to its true physical value.

[0094] In one embodiment, the step of using the sensor drift component to compensate the multi-source data to obtain compensated data, and then interpolating and repairing the distorted data of the compensated data to obtain optimized data, may include the following sub-steps: S131. Based on the timestamp of the sensor drift component, determine the corresponding original monitoring value from the multi-source data, and use the sensor drift component to compensate the original monitoring value to obtain compensated data.

[0095] S132. Determine the saturation distortion interval of the compensation data, wherein the saturation distortion interval is a data interval in the compensation data in which at least a number of consecutive values ​​reach the saturation threshold within the sliding window.

[0096] S133. Interpolate and replace the distorted data in the saturated distortion range to obtain optimized data.

[0097] In one embodiment, the time-series dataset of sensor drift components can be read directly; Next, key fields of the dataset can be matched: global unified timestamp, unique sensor number, value of sensor drift component Sdrift(t) marked as valid by MK trend test, distortion type label, and drift rate; group index is made according to the same source measurement point group, and key-value mapping is established according to sensor number to realize millisecond-level directional retrieval of single sensor drift amount.

[0098] Simultaneously, the original multi-source time series matrix D(t) can be read synchronously, and the multi-source time series matrix D(t) and the sensor drift component Sdrift(t) can be pre-bound based on the timestamp; for measurement points without effective drift markers (random fluctuations), Sdrift(t) is set to 0 by default, and the compensation operation is skipped.

[0099] The verification process ensures that the drift component time series length and sampling interval are completely consistent with the original monitoring data. If any time series is missing, the missing points are marked and filled with empty identifiers without interrupting the overall operation. Simultaneously, the original time series and drift component datasets can be loaded into a high-speed memory processing buffer, cutting off the low-speed hard disk I / O link. A low-latency memory direct-read compensation link uses key-value pair indexes for targeted matching of sensor drift data, avoiding the millisecond-level latency caused by traditional database queries and disk reads, thus meeting the real-time requirements of downhole monitoring.

[0100] Next, we can iterate through each unified timestamp t and each sensor in the same source group in the memory buffer; retrieve the sensor's original monitoring value Draw(t) and the corresponding effective drift component Sdrift(t) at the current moment; Performing point-by-point correction can be done as follows: Dcorrect(t) = Draw(t) Sdrift(t); Optionally, for scenarios with composite distortion (environmental drift + zero-point drift), the formula automatically subtracts the global environmental drift component, retaining only the true physical monitoring value. Furthermore, it calculates independently for each sampling point, without using window averaging or segmented uniform correction; each time point uses its own drift-specific value for individual correction. After the calculation is completed, the corrected time series Dcorrect(t) can be temporarily stored to obtain the compensated data. The original value, drift correction amount, timestamp, and sensor number are bound synchronously; at the same time, two copies of the data before and after correction can be retained without overwriting the original monitoring data, ensuring data traceability.

[0101] This application adopts a full-time, point-by-point independent real-time subtraction compensation mechanism, which differs from the traditional fixed-period, segmented unified calibration. It can dynamically adjust the correction amplitude as the zero-point drift changes slowly, without segmented boundary jump errors.

[0102] When performing interpolation, the original monitoring value Draw(t) before compensation can be retrieved, and the full-scale threshold of each sensor can be read from the factory calibration. The saturation judgment threshold can be set: the value of continuous sampling points ≥ 95% × full scale, eliminating the interference of single-point instantaneous peak values.

[0103] Next, a sliding length determination rule can be adopted: set the sliding window length W=10 synchronous sampling points. If 8 or more consecutive values ​​in the window reach the saturation threshold, the time period is determined to be the saturation distortion interval. A saturation label can be added to each time point: normal, positive saturation, negative lower limit saturation.

[0104] At this point, two types of saturation scenarios can be distinguished: hardware range jamming saturation and short-term over-limit real danger signals. Only the saturation points in continuous intervals are marked, and instantaneous over-limits are not judged as distortion. Then, the start and end timestamps of all saturation distortion intervals, the corresponding sensor numbers, and the duration of saturation are recorded.

[0105] The saturation marker is bound to the corrected time series Dcorrect(t) (i.e., the compensation data), and the marker points await interpolation repair.

[0106] In the compensation data, time-series points marked as saturation distortion and their corresponding homogeneous measurement point groups are selected; faulty, disconnected, or drift-exceeding sensors within the same group are removed, and only the remaining normal homogeneous measurement points without saturation distortion within the group are retained; and weighting coefficients are set based on spatial distance: the closer to the saturation sensor, the higher the interpolation weight; the farther the distance, the linearly decreasing weight.

[0107] Next, a weighted interpolation formula can be constructed, which can use the values ​​of multiple normal, homogeneous sensors calibrated at the same time to fit the true physical values ​​of saturated points. Furthermore, during interpolation, only points in the continuous saturation interval can be interpolated, while unsaturated and momentarily over-limit points retain their original compensated values.

[0108] After interpolation is completed, the clean time series Dfix(t) after repair can be output, yielding optimized data. Simultaneously, the interpolation weights and the numbers of the corresponding sensors involved in the calculation are recorded for each saturation interval; and repair markers are stored to distinguish between the original compensation values ​​and the repaired values ​​from the interpolation, facilitating later traceability and differentiation between the actual acquired data and the fitted repaired data.

[0109] This application can fit data from sensors that are spatially adjacent and of the same monitoring type during interpolation, without introducing external data from irrelevant measurement points, thus eliminating distortion and interference from external data sources.

[0110] In an optional embodiment, a consistency check is performed after compensation. Specifically, the consistency check may include the following steps: Step 1: Divide the data into comparison samples: repaired timing sequence Dfix(t) and original timing sequence Draw(t) before compensation; The second step is to calculate the time series variance of the two sets of data according to the same source measurement point group: the original data variance Varraw and the variance after compensation and repair Varfix. Step 3: Verification and judgment rules: In normal compensation scenarios, Varfix is ​​less than Varraw, and the dispersion of values ​​of each measurement point within the group is significantly reduced; Step 4: Calculate the mean deviation of each pair of measuring points within the group, compare the decrease in the mean deviation before and after compensation, and quantify the compensation effect. Step 5: If the variance does not decrease significantly or even increases, mark the sensor compensation as abnormal and determine that there is interference in the extraction of drift components. Step 6: Generate a single sensor compensation effect score, which is calculated based on the variance reduction and measurement point consistency deviation. Step 7: Bind the consistency verification results to the repaired timing sequence as an auxiliary basis for subsequent fault diagnosis.

[0111] By verifying the effectiveness of drift compensation and saturation repair, abnormal compensation channels can be quickly identified, ensuring the data consistency of the input risk assessment model.

[0112] S14. Calculate the comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and determine the construction risk level based on the comprehensive risk index.

[0113] In one embodiment, the real-time dynamic weights and multi-scale gradients corresponding to the optimized data can be determined. Then, the comprehensive risk index is calculated using the optimized data, real-time dynamic weights, and multi-scale gradients. The construction risk level is determined based on the comparison between the comprehensive risk index and the risk threshold.

[0114] If the construction risk level does not meet the construction conditions, an alarm can be triggered to prompt construction personnel to evacuate and ensure their safety.

[0115] In one embodiment, the step of calculating a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and determining the construction risk level based on the comprehensive risk index, may include the following sub-steps: S141. Determine the real-time dynamic weights and multi-scale gradients corresponding to the optimized data, wherein the multi-scale gradient is the gradient of the optimized data under different time window scales, and the real-time dynamic weights are weight values ​​assigned according to the gradient of the optimized data.

[0116] S142. Using the optimized data, the real-time dynamic weights, and the multi-scale gradients, a nonlinear coupling calculation is performed to obtain the comprehensive risk index.

[0117] S143. Determine the construction risk level based on the comprehensive risk index and several preset risk interval thresholds.

[0118] In one embodiment, optimized data Dfix(t) can be read. Multi-dimensional synchronous data is extracted according to the monitored physical quantities: gas concentration, CO concentration, surrounding rock stress, tunnel wind speed, and temperature and humidity corrected values; then, based on the globally unified BeiDou / GPS timestamp, multi-parameter time sequence alignment verification is completed, and invalid sampling frames with time misalignment or missing data are removed; at the same time, each data item can be synchronously bound with attached tags: saturation repair mark, sensor drift warning level, and compensation consistency verification score.

[0119] Next, a hierarchical index can be constructed according to mine zones and groups of similar measuring points to enable batch synchronous retrieval of data from multiple types of sensors on the same monitoring section; distorted data from sensors marked with serious faults can be filtered out, and faulty measuring points can only retain records without participating in risk fusion calculations; all valid multi-parameter time series can be loaded into a dedicated calculation buffer for risk assessment, and disk I / O latency can be isolated to ensure real-time calculations.

[0120] Next, for each type of physical quantity in the buffer, the timing of the correction can be set with three sliding time windows: a short-term 5-second window, a medium-term 1-minute window, and a long-term 10-minute window.

[0121] After determining three different windows, first-order difference operations can be performed on the time series within each of the three windows to calculate the numerical change gradient at each scale, as shown in the following formula: Gradient = (End value of window) (Initial window value) / Window duration; Next, positive gradients (rapid increase in concentration / stress, increased risk), negative gradients (indicators falling back, reduced risk), and zero gradients (indicators remaining stable with no change) can be distinguished. For each sampling point, three sets of features are simultaneously stored: a 5-second instantaneous gradient, a 1-minute trend gradient, and a 10-minute long-term evolution gradient. Combined with mine safety thresholds, gradient amplitudes are categorized and labeled as three types: gentle fluctuations, rapid increases, and sharp increases. Simultaneously, false abrupt gradients generated by saturation repair interpolation points can be removed, retaining only the evolution characteristics of the true monitoring indicators. Finally, the multi-scale gradient features are bound and stored one-to-one with the original correction values ​​and timestamps.

[0122] This application uses a three-scale gradient joint feature extraction method to simultaneously capture the evolution patterns of three types of downhole disasters: instantaneous sudden risks, short-term continuous deterioration, and long-term slow accumulation. A single-scale gradient cannot cover the risk features of the entire scenario.

[0123] After determining the multi-scale gradient, the real-time dynamic weights can be determined. Specifically, a preset basic weight matrix [w1, w2, w3…] can be initialized first, and basic weighting coefficients for gas, stress, CO, and wind speed can be preset according to mine safety regulations.

[0124] Next, the three-scale gradients of each monitoring indicator are traversed. If the gradient of an indicator reaches a rapid or sharp increase level, the corresponding fusion weight of that indicator is automatically increased. For measurement points with drift warnings or saturation repair markers, their weights are appropriately reduced to reduce the interference of distorted repair data on the evaluation results. Then, all indicator weights are normalized in real time to ensure that the sum of all indicator weights is always 1 at the same time, avoiding risk index deviation. The basic weight baseline can also be dynamically switched according to different mining scenarios (differentiated configuration of weights for tunneling faces, mining faces, and roadways / chambers). Finally, a unique dynamic weight set for each sampling point can be output, which is bound to the corresponding gradient and correction value. Next, a comprehensive risk index can be calculated based on real-time dynamic weights, multi-scale gradients, and optimized data. In accordance with coal mine safety regulations, four risk index interval thresholds are set to achieve a four-level standardized classification: 1. Low risk: 0 ≤ R(t) < 30; 2. Moderate risk: 30 ≤ R(t) < 50; 3. Significant risk: 50 ≤ R(t) < 80; 4. Major risk: 80 ≤ R(t) ≤ 100; Optionally, gradient features can be used for level correction: if the current risk level is low but the gradient rises sharply, the warning level is automatically raised by one level. The risk level is determined based on a comparison between the comprehensive risk index and the risk index interval threshold.

[0125] If the risk level is low, there will be no alarm; if it is a general risk, a yellow warning light will be displayed locally, and a notification will be sent to the background log; if it is a significant risk, an intermittent audible and visual alarm will be triggered, and pop-up notifications will be sent simultaneously to the underground substation and the surface monitoring room; if it is a major risk, a continuous high-decibel audible and visual alarm will be triggered, and an alarm signal will be sent remotely via SMS / platform.

[0126] To further illustrate this point, let's take a practical example. Suppose that a coal mine has multiple types of calibrated sensors arranged in a coal mining source measurement group, including: 1. Gas sensors T1 and T2; 2. CO sensor C1; 3. Roadway wind speed sensor V1; 4. Surrounding rock stress sensor S1.

[0127] The current working condition underground is coal cutting operation at the working face, with gas slowly and continuously emerging, and the gas index showing a continuous upward trend. In the previous stage, time alignment, benchmark modeling, drift separation, zero-point dynamic compensation, and saturation repair were completed through steps S11-S13, outputting undistorted and pure optimized data, and entering the risk assessment and graded alarm stage.

[0128] The current unified BeiDou timestamp is 202x-0x-30 14:22:xx. All sensor data has been freed from temperature and humidity drift, zero-point aging drift, and range saturation distortion.

[0129] The edge gateway memory channel directly retrieves optimized data without needing to read the disk; Multi-parameter synchronous data (corrected values ​​at the same time 14:22:xx) are split according to monitoring type, including: T1 gas: 0.68%, marked as normal with no saturation repair, slight drift warning; T2 gas: 0.71%, no fault indication; C1 Carbon monoxide: 0.0012%, normal; V1 wind speed: 0.75m / s (minimum requirement in regulations is 0.6m / s); S1 Surrounding rock stress: 12.6MPa, slowly increasing.

[0130] All data were verified using the BeiDou unified timestamp to confirm that the timing of gas, CO, wind speed, and stress was completely aligned, with no time misalignment or missing sampling frames. Severely faulty sensors were filtered out: this group had no disconnected or drift-exceeding level 3 fault devices; only T1 had a slight drift marker, and its data was retained for calculation. Measurement point indexes were established according to the mining face partitions to achieve batch synchronous reading of all multi-parameter data for this section. Invalid measurement points with severely distorted markers or excessively long saturation repair intervals were also removed; in this example, all measurement points were valid. The entire set of multi-parameter data was loaded into a dedicated memory buffer for risk assessment, awaiting gradient calculation.

[0131] Next, the real-time gradient can be calculated. The specific processing flow is as follows: The first step is to set up three sliding windows for each type of monitoring indicator in the buffer zone: a 5-second short-term window, a 1-minute medium-term window, and a 10-minute long-term window.

[0132] The second step is to calculate the gradients of the three layers (taking gas T2 as an example): Instantaneous gradient over 5 seconds: (0.71-0.66) / 5 = 0.01% / s, a small but rapid increase; 1-minute gradient: (0.71-0.52) / 60 ≈ 0.00317 % / s, continuously increasing; 10-minute gradient: (0.71-0.38) / 600 ≈ 0.00055 % / s, which increases cumulatively over a long period.

[0133] In the third step, the gradient of stress S1 was calculated synchronously, and it also showed a positive upward trend; the gradients of wind speed and CO were close to 0, and the indicators were stable.

[0134] The fourth step is to mark the gradient levels: gas and stress are rapidly increasing gradients, while CO and wind speed are gently fluctuating gradients.

[0135] Fifth, this group has no saturated interpolation repair points, no spurious abrupt gradients, and all gradients are valid.

[0136] The sixth step is to bind the three sets of gradient values ​​and gradient labels to the monitoring data at the corresponding time 14:22:xx.

[0137] The seventh step is to retain all three gradient characteristics of the continuous rise in gas levels as the core characteristics of increased risk.

[0138] The three-scale gradient extraction method can capture both instantaneous gas surges and identify long-term, slow-accumulating hazards. In this case, the instantaneous value of 0.71% did not reach the alarm threshold of 1.0%, but the multi-scale gradient continued to rise positively, which can predict the risk escalation in advance and make up for the shortcomings of traditional methods that only look at the instantaneous concentration.

[0139] Then the real-time dynamic weights can be determined. The specific processing flow is as follows: The first step is to load the preset basic weights of the coal mine safety regulations: gas w=0.45, stress w=0.25, wind speed w=0.20, and CO w=0.10; The second step is to determine the gradient of the indicators: gas and stress are rapidly increasing gradients, and the weights are automatically adjusted: the weight of gas is increased to 0.55 and the weight of stress is increased to 0.30. The third step is to reduce the weight of the T1 gas measurement point by 20% and use the T2 gas measurement point as the main reference. The fourth step is to normalize the weights of all indicators in real time, with the sum fixed at 1. The real-time dynamic weights after correction are: gas 0.55, stress 0.30, wind speed 0.10, CO 0.05. Fifth step: The current scenario is normal coal cutting at the longwall face, and there is no need to switch the weight baseline of the tunneling face; Step 6: Bind and store the dynamic weight set specific to time 14:22:xx along with the gradient and monitoring values; The seventh step is to update the weights every second as the indicators change. If the gas level continues to rise in the next second, the gas weight will be increased further.

[0140] Finally, real-time dynamic weights, multi-scale gradients, and corrected optimized data can be retrieved, and the comprehensive risk index R(t) can be calculated using a nonlinear coupled scoring formula, as shown in the following equation: ; The closer the indicator is to the safety limit, the higher the score amplification factor. In one operating mode, the risk contribution of each parameter is calculated separately as follows: Gas content is 0.71%, continuing to rise, contributing 42.5 points. The surrounding rock stress increases slowly, contributing 16.8 points to the score; The wind speed just meets the minimum standard, contributing 3.2 points; The CO value is extremely low, contributing only 0.9 points.

[0141] In one embodiment, the weighted summation yields a local comprehensive risk index R(t) of 63.4, with a value range of 0 to 100. If the risk indices of all other working faces in the entire mine are below 30, the overall comprehensive risk index of the entire mine is 28.7; all sensors have no large-area failures, and the risk index credibility is marked as "high credibility"; the local cross-sectional risk index, the overall risk index, and the credibility label can be stored simultaneously.

[0142] Finally, the risk level can be determined based on the comprehensive risk index.

[0143] The four-level classification criteria are: low risk [0,30), moderate risk [30,50), relatively high risk [50,80], and major risk [80,100].

[0144] The current comprehensive risk index is 63.4, falling within the 50-80 range, and is basically judged as a relatively high risk. The current judgment of a relatively high risk matches the corresponding graded alarm logic. It can perform precise alarms in different zones: only the intermittent audible and visual alarm of the underground substation of the mining face is activated, and the pop-up window of the ground monitoring center is highlighted, without triggering alarms in other roadways or working faces, thus avoiding noise interference to workers throughout the area.

[0145] Optionally, an alarm anti-shake verification can be set: if the alarm level is maintained stably for 3 seconds without any instantaneous fluctuations, an alarm signal will be officially issued; at the same time, this alarm can be recorded: trigger time 14:22:xx, risk cause gas accumulation; the underground alarm equipment transmits its working status back to the edge gateway in real time to verify that the alarm signal is output normally. In addition, complete end-to-end data at this moment can be collected, including: The data includes the original sensor data, zero-point drift compensation, saturation repair marker, dynamic weight, risk index (63.4), major risk classification, zone alarm trigger record, and BeiDou timestamp (14:22:xx). Then, a lightweight edge blockchain node packages all fields to generate a unique hash digest. The hash digest and complete monitoring record are appended to the block, ensuring that on-chain data is only added, not deleted, and cannot be tampered with. Additionally, the local time-series database synchronously stores the complete plaintext data, while the blockchain retains the hash certificate, allowing for data verification during safety inspections and accident investigations. A block index is established based on the work surface and 202x-0x-30, supporting rapid retrieval by alarm time and sensor number. This block is automatically stamped with the edge gateway device's digital signature, distinguishing the computational entities in the data acquisition, compensation, evaluation, and alarm stages. The logs simultaneously meet the retrieval needs of daily safety supervision, gas accident tracing, and sensor fault verification, forming a complete closed-loop evidence chain.

[0146] In this embodiment, the present application provides a construction risk assessment method based on multi-source data. Its advantages are as follows: This application can acquire multi-source data from the construction site and perform time-series alignment and standardization on the multi-source data to obtain a multi-source time-series matrix; determine the residual data of the multi-source time-series matrix and extract sensor drift components from the multi-source data based on the residual data; use the sensor drift components to compensate the multi-source data to obtain compensated data, and interpolate to repair the distorted data in the compensated data to obtain optimized data; use the optimized data and the corresponding real-time dynamic weights and multi-scale gradients to calculate a comprehensive risk index, and determine the construction risk level based on the comprehensive risk index. After collecting multi-source data, this application can determine the drift components based on the residual data of the multi-source data, and perform compensation and interpolation repair based on the drift components to reduce data distortion and bias, thereby improving the accuracy of risk assessment.

[0147] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0148] This application also provides a construction risk assessment device based on multi-source data, see [link to relevant documentation]. Figure 2 The diagram shows a structural schematic of a construction risk assessment device based on multi-source data provided in an embodiment of this application.

[0149] As an example, the construction risk assessment device based on multi-source data may include: The acquisition module is used to acquire multi-source data from the construction mine and perform time-series alignment and standardization on the multi-source data to obtain a multi-source time-series matrix, wherein the multi-source data is the original time-series data of multiple different sensors; The determination module is used to determine the residual data of the multi-source time series matrix and extract the sensor drift component from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The optimization module is used to compensate the multi-source data using the sensor drift component to obtain compensated data, and to interpolate and repair the distorted data of the compensated data to obtain optimized data. The evaluation module is used to calculate a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and to determine the construction risk level based on the comprehensive risk index.

[0150] Optionally, determining the residual data of the multi-source time series matrix and extracting the sensor drift component from the multi-source data based on the residual data includes: The multi-source time series matrix is ​​traversed to filter the raw time series data collected by two sensors from the same source at the same time, and the absolute value of the difference between the two raw time series data is taken to obtain the residual data. Several individual drift components are extracted from multiple residual data based on the minimum variance method, wherein the individual drift components are data components of individual sensors that have experienced zero-point drift or compound drift. Each individual drift component is filtered, corrected, and trend-verified to obtain the sensor drift component. The filtering correction is a correction of noise and component values, and the trend verification is a screening process that uses a non-parametric trend test algorithm to calculate the trend significance statistic of the individual drift component for selection.

[0151] Optionally, the extraction of individual drift components from the multiple residual data based on the minimum variance method includes: A residual matrix is ​​formed from several residual data from the same source, and the residual matrix is ​​divided into a global common component and an individual initial component based on the minimum variance method. Determine the global common variance corresponding to the global common component and the individual initial variance corresponding to the individual initial component, respectively. Based on the individual initial variance and the global common variance, the individual initial components that meet the drift conditions are selected as individual drift components.

[0152] Optionally, the step of filtering and correcting each individual drift component and performing trend verification to obtain the sensor drift component includes: The environmental noise components of the mine are determined, and each individual drift component is filtered and corrected based on the environmental noise components of the mine to obtain the individual corrected component; The trend significance statistic for each individual corrected component is calculated using a nonparametric trend test algorithm. Based on the trend significance statistic, individual corrected components that meet the preset statistic values ​​are selected as sensor drift components.

[0153] Optionally, the step of using the sensor drift component to compensate the multi-source data to obtain compensated data, and then interpolating and repairing the distorted data of the compensated data to obtain optimized data, includes: Based on the timestamp of the sensor drift component, the corresponding original monitoring value is determined from the multi-source data, and the original monitoring value is compensated using the sensor drift component to obtain compensated data. Determine the saturation distortion interval of the compensation data, wherein the saturation distortion interval is a data interval in the compensation data in which at least a number of consecutive values ​​reach the saturation threshold within a sliding window; The distorted data in the saturated distortion range is interpolated and replaced to obtain optimized data.

[0154] Optionally, the step of calculating a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and determining the construction risk level based on the comprehensive risk index, includes: The real-time dynamic weights and multi-scale gradients corresponding to the optimization data are determined respectively, wherein the multi-scale gradient is the gradient of the optimization data under different time window scales, and the real-time dynamic weights are weight values ​​assigned according to the gradient of the optimization data. The comprehensive risk index is obtained by nonlinear coupling calculation using the optimized data, the real-time dynamic weights, and the multi-scale gradients. The construction risk level is determined based on the comprehensive risk index and several preset risk interval thresholds.

[0155] This application also provides a construction risk assessment system based on multi-source data, see [link to relevant documentation]. Figure 3 The diagram shows a structural schematic of a construction risk assessment system based on multi-source data provided in an embodiment of this application.

[0156] As an example, the construction risk assessment system based on multi-source data may include: a risk assessment platform and several mine detection terminals; The risk assessment platform is communicatively connected to several of the mine detection terminals, and the risk assessment platform is applicable to the construction risk assessment method based on multi-source data as described in the above embodiments.

[0157] Among them, the mine detection terminal can be different types of sensors used to collect different data for the risk assessment platform to conduct risk assessment.

[0158] Reference Figure 4 The present application illustrates a computer device for a construction risk assessment method based on multi-source data, which may specifically include the following: The computer device 12 described above is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0159] Bus 18 refers to one or more of several types of bus 18 architectures, including memory bus 18 or memory controller, peripheral bus 18, graphics acceleration port, processor, or local bus 18 using any of the various bus 18 architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) bus 18, Micro Channel Architecture (MAC) bus 18, Enhanced ISA bus 18, Audio / Video Electronics Standards Association (VESA) local bus 18, and Peripheral Component Interconnect (PCI) bus 18.

[0160] Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0161] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard disk drive"). Although Figure 4 As not shown, a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (such as a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. The memory may include at least one program product having a set (e.g., at least one) of program modules 42 configured to perform the functions of the embodiments of this application.

[0162] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory. Such program modules 42 include—but are not limited to—an operating system, one or more application programs, other program modules 42, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of this application.

[0163] Computer device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, camera, etc.), and with one or more devices that enable a user to interact with the computer device 12, and / or with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 22. Furthermore, computer device 12 can also communicate with one or more networks (e.g., local area network (LAN)), wide area network (WAN), and / or public networks (e.g., the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18. It should be understood that, although... Figure 4 Not shown, it can be combined with computer device 12 to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing unit 16, external disk drive array, RAID system, tape drive and data backup storage system 34, etc.

[0164] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the construction risk assessment method based on multi-source data provided in the embodiments of this application.

[0165] That is, when the processing unit 16 executes the above program, it achieves the following: Multi-source data from the construction site is acquired, and the multi-source data is subjected to time-series alignment and standardization to obtain a multi-source time-series matrix. The multi-source data consists of raw time-series data from multiple different sensors. The residual data of the multi-source time series matrix is ​​determined, and the sensor drift component is extracted from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The multi-source data is compensated using the sensor drift component to obtain compensated data, and the distorted data of the compensated data is repaired by interpolation to obtain optimized data. The comprehensive risk index is calculated using the optimized data, the corresponding real-time dynamic weights, and multi-scale gradients, and the construction risk level is determined based on the comprehensive risk index.

[0166] That is, to implement the following when the program is executed by the processor: Multi-source data from the construction site is acquired, and the multi-source data is subjected to time-series alignment and standardization to obtain a multi-source time-series matrix. The multi-source data consists of raw time-series data from multiple different sensors. The residual data of the multi-source time series matrix is ​​determined, and the sensor drift component is extracted from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The multi-source data is compensated using the sensor drift component to obtain compensated data, and the distorted data of the compensated data is repaired by interpolation to obtain optimized data. The comprehensive risk index is calculated using the optimized data, the corresponding real-time dynamic weights, and multi-scale gradients, and the construction risk level is determined based on the comprehensive risk index.

[0167] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0168] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0169] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider). The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably.

[0170] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0171] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0172] The foregoing has provided a detailed description of the construction risk assessment method, apparatus, system, equipment, and medium based on multi-source data provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A construction risk assessment method based on multi-source data, characterized in that, The method comprises: obtaining multi-source data of a construction mine and performing time sequence alignment and standardization processing on the multi-source data to obtain a multi-source time sequence matrix, wherein the multi-source data is original time sequence data of multiple groups of different sensors; determining residual data of the multi-source time sequence matrix and extracting a sensor drift component from the multi-source data according to the residual data, wherein the residual data is a difference value of data of two homologous sensors at the same time, and the sensor drift component is a data component of a sensor that has zero-point drift; compensating the multi-source data using the sensor drift component to obtain compensated data, and performing interpolation repair on distorted data of the compensated data to obtain optimized data; calculating a comprehensive risk index using the optimized data, real-time dynamic weights corresponding to the optimized data, and multi-scale gradients, and determining a construction risk level according to the comprehensive risk index. 2.The construction risk assessment method based on multi-source data according to claim 1, wherein, The method comprises: traversing the multi-source time sequence matrix to screen original time sequence data collected by two homologous sensors at the same time, and taking an absolute value of a difference value of the two original time sequence data to obtain residual data; extracting a plurality of individual drift components from the plurality of residual data based on a least variance method, wherein the individual drift component is a data component of an individual sensor that has zero-point drift or compound drift; filtering and correcting each individual drift component to obtain a sensor drift component, wherein the filtering and correction is a correction of noise and component values, and the trend verification is screening by calculating a trend significance statistic of the individual drift component using a non-parametric trend test algorithm. 3.The construction risk assessment method based on multi-source data according to claim 2, characterized in that, The method comprises: grouping a plurality of homologous residual data into a residual matrix, and dividing the residual matrix into a global common component and an individual initial component based on the least variance method; determining a global common variance corresponding to the global common component and an individual initial variance corresponding to the individual initial component, respectively; screening an individual initial component that meets a drift condition as an individual drift component according to the individual initial variance and the global common variance. 4.The construction risk assessment method based on multi-source data according to claim 2, characterized in that, The method comprises: determining a mine environment noise component, and filtering and correcting each individual drift component based on the mine environment noise component to obtain an individual corrected component; calculating a trend significance statistic value of each individual corrected component using a non-parametric trend test algorithm; screening an individual corrected component that meets a preset statistic value as a sensor drift component according to the trend significance statistic value. 5.The construction risk assessment method based on multi-source data according to claim 1, wherein, The method comprises: Based on the timestamp of the sensor drift component, the corresponding original monitoring value is determined from the multi-source data, and the original monitoring value is compensated using the sensor drift component to obtain compensated data. Determine the saturation distortion interval of the compensation data, wherein the saturation distortion interval is a data interval in the compensation data in which at least a number of consecutive values ​​reach the saturation threshold within a sliding window; The distorted data in the saturated distortion range is interpolated and replaced to obtain optimized data.

6. The construction risk assessment method based on multi-source data according to any one of claims 1-5, characterized in that, The process of calculating a comprehensive risk index using the optimized data, the corresponding real-time dynamic weights, and multi-scale gradients, and determining the construction risk level based on the comprehensive risk index, includes: The real-time dynamic weights and multi-scale gradients corresponding to the optimization data are determined respectively, wherein the multi-scale gradient is the gradient of the optimization data under different time window scales, and the real-time dynamic weights are weight values ​​assigned according to the gradient of the optimization data. The comprehensive risk index is obtained by nonlinear coupling calculation using the optimized data, the real-time dynamic weights, and the multi-scale gradients. The construction risk level is determined based on the comprehensive risk index and several preset risk interval thresholds. 7.A construction risk assessment device based on multi-source data, characterized by The device includes: The acquisition module is used to acquire multi-source data from the construction mine and perform time-series alignment and standardization on the multi-source data to obtain a multi-source time-series matrix, wherein the multi-source data is the original time-series data of multiple different sensors; The determination module is used to determine the residual data of the multi-source time series matrix and extract the sensor drift component from the multi-source data based on the residual data. The residual data is the data difference between two sensors from the same source at the same time, and the sensor drift component is the data component in which the sensor experiences zero-point drift. The optimization module is used to compensate the multi-source data using the sensor drift component to obtain compensated data, and to interpolate and repair the distorted data of the compensated data to obtain optimized data. The evaluation module is used to calculate a comprehensive risk index using the optimized data and the corresponding real-time dynamic weights and multi-scale gradients, and to determine the construction risk level based on the comprehensive risk index. 8.A construction risk assessment system based on multi-source data, characterized in that, The system includes: a risk assessment platform and several mine detection terminals; The risk assessment platform is communicatively connected to several of the mine detection terminals, and the risk assessment platform is applicable to the construction risk assessment method based on multi-source data as described in any one of claims 1-6.

9. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the construction risk assessment method based on multi-source data as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer-executable program, which is used to cause a computer to perform the construction risk assessment method based on multi-source data as described in any one of claims 1-6.

Images