Coal mine heterogeneous data visualization processing method combined with artificial intelligence

By constructing a real-time heterogeneous data matrix and dividing feature categories, extracting abnormal data clusters, and calculating abnormal feature factors, the dynamic correlation and visualization problems of multi-source heterogeneous data in coal mines are solved, and accurate monitoring and risk assessment of coal mine safety production is achieved, and data processing efficiency and visualization effect are improved.

CN120371911AInactive Publication Date: 2025-07-25SHAANXI YANCHANG PETROLEUM MINING CO LTD

Patent Information

Application Number
CN202510873918.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-27
Publication Date
2025-07-25
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the safety production of coal mines, the dynamic correlation of multi-source heterogeneous data is complex, and traditional methods are difficult to effectively capture the coupling effect between data, and the abnormal data recognition ability is insufficient, resulting in missed detection or false alarms, and the inability to locate potential risk areas in time. The existing visualization technology lacks in-depth information expression and cannot meet the precise positioning needs of key information areas.

Method used

By constructing a real-time heterogeneous data matrix, divide feature categories based on the dynamic correlation of data points, generate optimization matrix, extract abnormal data clusters, calculate abnormal feature factors, combine stable feature index and Manhattan distance to locate key information areas in the three-dimensional space of coal mines.

Benefits of technology

It realizes efficient processing and accurate abnormality detection of heterogeneous data of coal mines, improves the depth and breadth of data fusion, breaks through the limitations of traditional visualization, can intuitively display the abnormal risk level, supports real-time monitoring and early warning response, and reduces the possibility of accidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120371911A_ABST
    Figure CN120371911A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of coal mine safety monitoring, and discloses a coal mine heterogeneous data visualization processing method combined with artificial intelligence, and the method comprises the steps: collecting multi-source heterogeneous data above and under a coal mine to construct a real-time heterogeneous data matrix; feature categories are divided based on data dynamic relevance, and stable data areas are screened through a dynamic analysis window to generate an optimization matrix; performing pattern recognition on the unstable region to extract an abnormal data cluster, and calculating an abnormal feature factor of a data point; analyzing the feature distribution difference to generate a stable feature index, and calculating an abnormal risk value in combination with the distance, the abnormal feature factor and the stable feature index; and mapping the center coordinate of the stable region to a three-dimensional space of the coal mine, and positioning a key information region in combination with the abnormal risk value. An apparatus includes a memory, a processor, and a computer program. The coal mine heterogeneous data processing efficiency and the safety monitoring accuracy are improved, intelligent positioning of the key information area is achieved, and powerful support is provided for coal mine safety production.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety monitoring, and specifically to a visualization processing method for heterogeneous data in coal mines combined with artificial intelligence. Background Art

[0002] In the field of coal mine safety production, the environment above and below ground is complex and the types of equipment are diverse, resulting in a wide range of data sources and significant structural differences, forming multi-source heterogeneous data including environmental parameters (such as gas concentration, temperature, humidity, dust concentration, etc.) and equipment operating status (such as motor speed, voltage, current, equipment start-stop status, etc.). These data are characterized by strong real-time nature, complex dynamic correlation, and many noise interferences. How to efficiently process and extract key information from them to achieve accurate monitoring and early warning of the coal mine safety status is one of the core challenges faced by the current coal mine intelligent construction.

[0003] Traditional data processing and analysis methods have obvious limitations when dealing with coal mine heterogeneous data. On the one hand, the ability to fuse and process multi-source data is insufficient, and it is difficult to effectively capture the dynamic correlation between different types of data, resulting in insufficient exploration of data value. For example, abnormal changes in environmental parameters may have a potential connection with fluctuations in equipment operating status, but traditional methods often analyze the two types of data independently and cannot discover the coupling effect. On the other hand, in the identification of abnormal data, traditional methods mostly rely on fixed thresholds or simple statistical analysis, lacking the ability to intelligently identify data patterns and being difficult to adapt to the complex and changeable abnormal situations in the coal mine environment. When the data shows non-periodic fluctuations or multi-factor superposition anomalies, traditional methods are prone to missed detections or false alarms and cannot accurately locate potential risk areas in a timely manner.

[0004] In addition, the requirements for data visualization in coal mine safety production are increasing day by day. It is necessary to convert abstract data into intuitive and interactive visual information so that staff can quickly grasp the overall situation and make decisions. However, most existing visualization technologies stay at the level of simple data chart display, lacking the visualization expression of in-depth information such as data feature distribution and abnormal risk level, and cannot meet the need for accurate positioning of key information areas in coal mine safety monitoring. How to combine artificial intelligence technology with data visualization technology to build a set of intelligent processing and visualization identification systems suitable for coal mine heterogeneous data has become a technical problem to be solved urgently.

[0005] With the rapid development of artificial intelligence technology, it has demonstrated powerful capabilities in the fields of data mining, pattern recognition, intelligent analysis, etc., providing new ideas for the processing of heterogeneous data in coal mines. By introducing artificial intelligence algorithms, it is possible to deeply analyze the dynamic relevance of multi-source heterogeneous data, realize the intelligent classification of data feature categories and the accurate extraction of abnormal data clusters. At the same time, combined with visualization technology, mapping the data processing results to the three-dimensional space of the coal mine can intuitively display the abnormal risk levels in different regions, providing strong technical support for coal mine safety production. Therefore, studying a visualization processing method for heterogeneous data in coal mines combined with artificial intelligence has important theoretical significance and practical application value. Summary of the Invention

[0006] The purpose of the present invention is to provide a visualization processing method for heterogeneous data in coal mines combined with artificial intelligence to solve the problems proposed in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: A visualization processing method for heterogeneous data in coal mines combined with artificial intelligence, the method includes: Collect multi-source heterogeneous data above and below the coal mine, including environmental parameters and equipment operating status, and construct a real-time heterogeneous data matrix; Based on the dynamic relevance of each data point in the real-time heterogeneous data matrix, divide the data feature categories to generate a feature category matrix; preset a dynamic analysis window, and screen the stable data area according to the dispersion degree of the feature categories within the window to generate an optimized matrix; Perform pattern recognition on the unstable data areas in the optimized matrix to extract abnormal data clusters; combine the deviation degree of each data point in the abnormal data cluster from the global feature mean value and the feature correlation entropy value within the adjacent window to calculate the abnormal feature factor of each data point. Analyze the feature distribution differences between the stable data area and the abnormal data cluster, fuse the feature entropy change characteristics of the stable data area, and generate the stable feature index of each stable data area; calculate the minimum Manhattan distance between each stable data area and the core point of the abnormal data cluster, and combine the abnormal feature factor and the stable feature index to generate the abnormal risk value of each stable data area. Map the coordinates of the center of each stable data area in the real-time heterogeneous data matrix to the three-dimensional space of the coal mine, and locate the key information area in combination with the abnormal risk value.

[0008] Preferably, the specific process of dividing the data feature categories is as follows: Take the logarithmic mapping result of the variances of all data points in the real-time heterogeneous data matrix as the category division step size; take the minimum feature point of the real-time heterogeneous data matrix as the category starting benchmark, divide continuous category intervals, and assign feature categories to each data point. The feature category matrix is composed of the feature categories of all data points.

[0009] Preferably, the process of screening the stable data region is specifically as follows: Calculate the dispersion of the data within the window when the dynamic analysis window moves in the feature category matrix, and screen the continuous region with a dispersion of 0 as the stable data region.

[0010] Preferably, the optimization matrix is a real-time heterogeneous data matrix with the data points at the corresponding positions in the stable data region set to zero.

[0011] Preferably, the specific process of extracting the abnormal data cluster is as follows: Set the upper limit of the number of modes, and calculate the feature mean of the data points within each mode cluster; select the mode cluster with the smallest feature mean as the abnormal data cluster.

[0012] Preferably, the calculation process of the abnormal feature factor includes: Calculate the feature mean of all the data points within the abnormal data cluster; generate the local deviation degree according to the difference between the feature value of each data point and the feature mean; Use the Markov chain algorithm to calculate the correlation entropy value of each data point in the neighboring window; Fuse the local deviation degree and the correlation entropy value to obtain the abnormal feature factor; wherein, the abnormal feature factor is positively correlated with the local deviation degree and negatively correlated with the correlation entropy value.

[0013] Preferably, the process of generating the stable feature index of each stable data region is specifically as follows: Calculate the relative difference between the feature mean within each stable data region and the feature mean of the abnormal data cluster; extract the minimum value of the relative difference and zero; Fuse the minimum value and the feature entropy within the stable data region to obtain the stable feature index.

[0014] Preferably, the calculation formula of the abnormal risk value is: ; Wherein: represents the abnormal risk value of the j-th stable data region; represents the Manhattan distance between the j-th stable data region and the m-th core point in the abnormal data cluster; represents the abnormal feature factor of the m-th core point; represents the stable feature index of the j-th stable data region; M represents the total number of core points in the abnormal data cluster; is the distance attenuation coefficient; is the activation function.

[0015] Preferably, the specific process of locating the key information region is as follows: Map the real-time heterogeneous data matrix to the 3D coordinate system of the coal mine to generate a feature distribution space; calculate the geometric center coordinates of the data points in each stable data area in the feature distribution space; Convert the geometric center coordinates into proportional coordinates of the actual location of the coal mine, where the abscissa is the ratio of the abscissa of the geometric center to the width of the coal mine, and the ordinate is the ratio of the ordinate of the geometric center to the depth of the coal mine; When the proportional coordinates fall within the boundary of the preset key area, mark the stable data area as the area to be analyzed; if the abnormal risk value of the area to be analyzed exceeds the preset threshold, determine the corresponding geometric center as the key information location.

[0016] Preferably, the present invention further includes a coal mine heterogeneous data visualization identification device combined with artificial intelligence. The device includes a memory, a processor, and a computer program stored on the memory. When the processor executes the computer program, the steps of the coal mine heterogeneous data visualization processing method combined with artificial intelligence as described above are implemented.

[0017] Compared with the prior art, the beneficial effects of the present invention are: At the data processing level, by collecting multi-source heterogeneous data above and below the coal mine and constructing a real-time heterogeneous data matrix, a comprehensive data basis is provided for subsequent analysis. Based on the dynamic correlation of data points, feature categories are divided and a feature category matrix is generated, which can effectively capture the internal relationship between data. Compared with the traditional method of independently analyzing different types of data, it can better reveal the complex relationship behind the data and improve the depth and breadth of data fusion. By presetting a dynamic analysis window and screening stable data areas according to the dispersion, the stable and non-abnormal data parts can be quickly filtered out, reducing the computational complexity of subsequent processing. At the same time, focusing on the unstable data areas improves the efficiency of anomaly detection. Setting the data points at the corresponding positions in the stable data area to zero to generate an optimized matrix further highlights the unstable data, making the characteristics of abnormal data more obvious and facilitating subsequent pattern recognition and analysis.

[0018] In terms of anomaly detection and risk assessment, pattern recognition is carried out on the unstable data regions in the optimization matrix, and abnormal data clusters are extracted, which changes the traditional simple detection method relying on fixed thresholds. By using artificial intelligence algorithms to intelligently identify data patterns, complex and changeable abnormal situations can be captured more accurately. The abnormal feature factor is calculated by combining the deviation degree of the data points in the abnormal data cluster from the global feature mean value and the correlation entropy value of the adjacent window, comprehensively considering the local deviation situation and dynamic correlation characteristics of the data, making the description of abnormal features more comprehensive and accurate. When generating the stable feature index of the stable data region, the difference of feature means and the characteristics of feature entropy change are fused, which can objectively reflect the feature stability of the stable region and provide a reliable basis for risk assessment. The calculation of the abnormal risk value combines the Manhattan distance, the abnormal feature factor and the stable feature index, fully considering the spatial position relationship, the degree of abnormality of the stable region and the characteristics of the stable region itself, making the risk assessment result more scientific and reasonable, and being able to provide more targeted reference for coal mine safety management.

[0019] In terms of visualization and key information location, the central coordinates of the stable data region are mapped to the three-dimensional space of the coal mine and combined with the abnormal risk value to locate the key information area, breaking through the limitation of traditional visualization technology that only displays the surface information of data. By converting abstract data features into position information in the actual space of the coal mine and marking the key areas based on the abnormal risk value, the accurate matching of data information and physical space is realized, enabling the staff to intuitively and quickly locate the key positions with potential safety hazards. This visualization identification method not only improves the readability and understandability of data, but also provides strong technical support for the real-time monitoring, early warning response and emergency disposal of coal mines, helps to discover potential risks in advance, take preventive measures in time, reduce the possibility of coal mine accidents, and ensure the safe production of coal mines and the safety of personnel. In addition, this method has strong versatility and scalability, can adapt to the heterogeneous data processing requirements under different scenarios of coal mines, provides important technical support for the intelligent and informatization construction of coal mines, and helps to promote the development of the coal mine industry towards a safer, more efficient and intelligent direction. Brief Description of the Drawings

[0020] Figure 1 It is the working principle diagram of the heterogeneous data visualization processing method for coal mines combined with artificial intelligence described in the present invention; Figure 2 It is the flow chart of abnormal feature factor calculation; Figure 3 It is the flow chart of stable feature index generation; Figure 4 It is the flow chart of abnormal risk value calculation. Detailed Embodiment

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Please refer to Figures 1 - 4 , the coal mine heterogeneous data visualization processing method combined with artificial intelligence involved in the present invention is specifically implemented as follows: Through various sensors deployed above and below the coal mine (such as environmental parameter sensors, equipment operation status sensors, etc.), multi-source heterogeneous data including environmental parameters (such as gas concentration, temperature, humidity, dust concentration, etc.) and equipment operation status (such as equipment voltage, current, rotation speed, working duration, etc.) are collected in real time. The collected data is structured according to time series and parameter types to construct a real-time heterogeneous data matrix. This matrix has time as the row dimension and different types of parameters as the column dimension, and each data point corresponds to a specific parameter value at a specific time point.

[0023] Taking the logarithmic mapping result of the variances of all data points in the real-time heterogeneous data matrix as the class division step size, and taking the minimum feature point of the real-time heterogeneous data matrix as the class starting benchmark, continuous class intervals are divided, feature classes are assigned to each data point, and a feature class matrix is formed by the feature classes of all data points.

[0024] A preset dynamic analysis window with a fixed time length is used to move sequentially in the feature class matrix in chronological order. Calculate the dispersion of the data within the window when the dynamic analysis window moves in the feature class matrix. The dispersion can be determined by calculating the ratio of the number of different feature classes within the window to the total number of data points within the window. Screen the continuous regions with a dispersion of 0 as stable data regions, that is, all data points within this region belong to the same feature class. Set the data points at the corresponding positions in the stable data region of the real-time heterogeneous data matrix to zero to generate an optimized matrix.

[0025] Perform pattern recognition on the unstable data region in the optimization matrix, set the upper limit of the number of patterns, and use a clustering algorithm (such as the K-means algorithm, etc.) to cluster the data points in the unstable data region. Calculate the characteristic mean values of the data points within each pattern cluster, and select the pattern cluster with the smallest characteristic mean value as the abnormal data cluster. Calculate the characteristic mean value of all data points within the abnormal data cluster, and generate a local deviation degree based on the difference between the characteristic value of each data point and this characteristic mean value. The larger the difference, the larger the local deviation degree. Use the Markov chain algorithm to calculate the correlation entropy value of each data point in the neighboring windows (the previous and the next windows adjacent to the current window in terms of time). This correlation entropy value reflects the correlation and uncertainty of the data point in the time series. Integrate the local deviation degree and the correlation entropy value to obtain an abnormal characteristic factor. Among them, the abnormal characteristic factor is positively correlated with the local deviation degree, that is, the larger the local deviation degree, the larger the abnormal characteristic factor; and it is negatively correlated with the correlation entropy value, that is, the larger the correlation entropy value, the smaller the abnormal characteristic factor.

[0026] Analyze the difference in the characteristic distribution between the stable data region and the abnormal data cluster, calculate the relative difference between the characteristic mean value in each stable data region and the characteristic mean value of the abnormal data cluster, and extract the minimum value of this relative difference and zero, that is, if the relative difference is negative, take this negative number, and if it is positive, take zero. Integrate this minimum value and the characteristic entropy change within the stable data region (the characteristic entropy change reflects the degree of chaos and the change trend of the characteristic distribution within the stable data region) to obtain a stable characteristic index. Calculate the minimum Manhattan distance between the core points of each stable data region and the abnormal data cluster, and combine the abnormal characteristic factor and the stable characteristic index, according to the calculation formula of the abnormal risk value Generate the abnormal risk value of each stable data region. Among them, represents the abnormal risk value of the j-th stable data region; represents the Manhattan distance between the j-th stable data region and the m-th core point in the abnormal data cluster; represents the abnormal characteristic factor of the m-th core point; represents the stable characteristic index of the j-th stable data region; M represents the total number of core points in the abnormal data cluster; is the distance attenuation coefficient, which is used to measure the influence degree of the distance on the abnormal risk value; is the activation function, which is used to perform non-linear transformation on the distance factor.

[0027] The real-time heterogeneous data matrix is mapped to the three-dimensional coordinate system of the coal mine, which uses the length, width and depth of the coal mine as three coordinate axes to generate a feature distribution space. The geometric center coordinates of the data points in each stable data area in the feature distribution space are calculated. The geometric center coordinates are obtained by calculating the average values of all data points in the stable data area on each coordinate axis in the three-dimensional coordinate system. The geometric center coordinates are converted into proportional coordinates of the actual position of the coal mine, where the horizontal coordinate is the ratio of the horizontal coordinate of the geometric center to the width of the coal mine, the vertical coordinate is the ratio of the vertical coordinate of the geometric center to the depth of the coal mine, and the vertical coordinate is the ratio of the vertical coordinate of the geometric center to the length of the coal mine. The critical area boundary is preset. When the proportional coordinate falls into the preset critical area boundary, the stable data area is marked as the area to be analyzed; if the abnormal risk value of the area to be analyzed exceeds the preset threshold, the corresponding geometric center is determined to be the key information position.

[0028] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1:

[0029] In the implementation method of classifying data feature categories, it is first necessary to complete the category assignment and matrix construction of data points based on the real-time heterogeneous data matrix. The real-time heterogeneous data matrix is composed of multi-source heterogeneous data from coal mines, which include environmental parameters (such as gas concentration, temperature, humidity, dust concentration, etc.) and equipment operating status parameters (such as equipment voltage, current, speed, working hours, etc.). Various types of data are structured and arranged according to time series and parameter types, forming a two-dimensional matrix with time as rows and parameter types as columns. Each data point corresponds to a specific parameter measurement value at a specific time point.

[0030] The core of classification is to determine the classification step and starting benchmark. In the specific operation, the variance of all data points in the real-time heterogeneous data matrix is first calculated. Variance is a statistic that measures the degree of data dispersion. The variance of each data point reflects its fluctuation in the parameter dimension to which it belongs. Since the dimensions and numerical ranges of different parameters may differ, in order to unify the measurement standards, the variance of each data point needs to be logarithmically mapped, that is, the variance value is substituted into the logarithmic function (such as the natural logarithm function) to obtain the logarithmic mapping result. This result is used to eliminate the influence of the dimension and compress the numerical range, and is ultimately used as the step size for classification. For example, if the variance of a data point is 100, its logarithmic mapping result (taking the natural logarithm as an example) is approximately 4.605, which is the step size for classification of the parameter dimension where the data point is located.

[0031] The determination of the category starting benchmark is based on the minimum feature points of the real-time heterogeneous data matrix. The minimum feature points refer to the minimum value points of all data points in the matrix in the feature space. By comparing the values of each data point in all parameter dimensions, a point with the minimum value in each dimension or the comprehensive minimum value can be found as the starting benchmark. For example, if there is a data point in the matrix whose values of all parameters such as gas concentration, temperature, and equipment voltage are the minimum values in the currently collected data, then this point is determined as the category starting benchmark.

[0032] Based on the determined step size and starting benchmark, the continuous category intervals are started to be divided. The division of the category intervals starts from the starting benchmark and increases sequentially according to the step size, forming a sequence of continuous and non-overlapping intervals. For example, if the starting benchmark is the eigenvalue a and the step size is b, then the first category interval is [a, a + b), the second is [a + b, a + 2b), and so on, until the eigenvalue range of all data points in the real-time heterogeneous data matrix is covered. Each category interval corresponds to a unique feature category number, such as 0, 1, 2, etc., which is used to identify the category to which the data points within the interval belong.

[0033] When assigning feature categories to data points, each data point in the real-time heterogeneous data matrix is traversed one by one, and it is judged which category interval it falls into according to its eigenvalue. For example, if the eigenvalue of a certain data point is c, and if it satisfies a + nb ≤ c < a + (n + 1)b (n is a non-negative integer), then this data point is assigned to the nth feature category. For data points at the interval boundaries, they can be processed according to the preset rules (such as uniformly classifying them into the left interval or the right interval) to ensure that each data point belongs to only one category interval.

[0034] After all data points have completed the assignment of feature categories, a feature category matrix is constructed. The row and column dimensions of this matrix are exactly the same as those of the real-time heterogeneous data matrix, and the element at each position is the feature category number of the corresponding data point. For example, if the data point in the i-th row and j-th column of the real-time heterogeneous data matrix is assigned to the feature category k, then the element value in the i-th row and j-th column of the feature category matrix is k. In this way, the original multi-source heterogeneous data is converted into a category identification matrix, which is convenient for subsequent analysis of the distribution law and dynamic correlation of data features.

[0035] In actual operation, the following details need to be noted: First, the selection of the logarithmic mapping function should be determined according to the characteristics of the data distribution. Commonly used ones include the natural logarithm, the logarithm with base 10, etc. The purpose is to make the step size reasonably reflect the differences in data dispersion. Second, the determination of the minimum feature point should consider all parameter dimensions comprehensively. If there are inconsistent minimum values in multiple dimensions, methods such as weighted average or principal component analysis can be used to determine the comprehensive minimum value point. Third, the division of the category intervals should ensure that the eigenvalue range of all data points is covered, avoiding the situation where data points cannot be classified into any category. Fourth, the rules for assigning data points to categories should be clearly defined in the algorithm to ensure the consistency and repeatability of the classification process.

[0036] The construction of the feature category matrix provides a basis for subsequent data processing. By converting continuous numerical data into discrete category identifiers, the data structure can be effectively simplified, facilitating the analysis of the dynamic correlation between data points. For example, when screening stable data regions subsequently, the category dispersion (i.e., the number of different categories within the window) within the dynamic analysis window can be calculated to judge the stability of the data region. A region with a dispersion of 0 indicates that all data points belong to the same category, that is, the data characteristics remain stable within this region and can be classified as a stable data region. While the regions with non-zero dispersion in the feature category matrix are used as unstable data regions and further pattern recognition is required to extract abnormal data clusters.

[0037] In addition, the logarithmic mapping step size and the minimum feature point benchmark adopted in the process of feature category division can adapt to different types of coal mine data distributions. For example, for data with relatively small fluctuations such as voltage and current in equipment operation state parameters, the logarithmic mapping step size of their variance is small, and the category interval division is more refined, which can accurately capture the subtle changes in data characteristics. While for data with relatively large fluctuations such as dust concentration in environmental parameters, the step size is large, and the category interval division is wider, which can avoid frequent category changes caused by small data fluctuations and improve the stability of classification.

[0038] Example 2:

[0039] In the implementation method of screening stable data regions, the core lies in traversing the feature category matrix through a dynamic analysis window, judging the stability of the data region based on the dispersion of the data within the window, and generating an optimized matrix through zeroing operations. The specific implementation process is as follows: Set the parameters of the preset dynamic analysis window. The dynamic analysis window is a sliding window with a fixed time length, and its time length can be set according to the acquisition frequency and analysis requirements of coal mine data. For example, if the data acquisition frequency is once per minute, the window time length can be set to 10 minutes, that is, the window contains data of 10 consecutive time points. The moving step of the window is usually consistent with the time interval of data acquisition. For example, it moves 1 time point each time, ensuring that the window can cover all time regions in the feature category matrix successively and analyze the dynamic changes of data without omission.

[0040] The feature category matrix is a two-dimensional matrix composed of the feature categories of data points. Its rows correspond to the time series, and its columns correspond to different data parameter dimensions. When screening the stable data region, the window moves for analysis with the row as the main dimension, that is, the window slides successively in the row direction (time axis) of the feature category matrix. Each time it slides, the data in the same column (the same parameter dimension) is processed, or multiple columns of data can be comprehensively analyzed at the same time (such as calculating the average or maximum value of the dispersion degrees of each column).

[0041] For each window position, it is necessary to calculate the dispersion degree of the data within the window. The dispersion degree is an index to measure the uniformity of the data category distribution within the window. The specific calculation method is: count the number of different feature categories within the window, divide it by the total number of data points within the window to obtain the dispersion degree value. For example, if there are 10 data points within the window and 3 different feature categories among them, the dispersion degree is 3 / 10 = 0.3. When the dispersion degree is 0, it indicates that all data points within the window belong to the same feature category, that is, the data characteristics have not changed during this time period, and the data is in a stable state. The corresponding region can be determined as a stable data region; if the dispersion degree is greater than 0, it means that there are different feature categories within the window, and the data characteristics fluctuate, and this region is an unstable data region.

[0042] In actual operation, the feature category matrix needs to be processed row by row and column by column or comprehensively row by row. Taking a single parameter dimension (a single column) as an example, the window starts from the first row of the matrix and covers the 1st to the nth rows (n is the number of rows corresponding to the window time length), and calculates the dispersion degree of the feature categories of the n data points within this window. If the dispersion degree is 0, record the time interval corresponding to this window as the starting position of the stable data region, continue to move the window (move 1 row each time), and check whether the dispersion degree of the next window is still 0. If the dispersion degrees of multiple consecutive windows are all 0, the time intervals corresponding to these windows are merged into a continuous stable data region. For example, when the window moves from the 1st row to the 5th row, the dispersion degree is 0, then the stable data region is the time interval corresponding to the 1st to the 5th rows. When the dispersion degree is greater than 0 during the window movement, the current stable data region terminates, and the next possible stable data region needs to start detection again from the window position where the dispersion degree becomes 0 again.

[0043] For the comprehensive analysis of multi-parameter dimensions, various methods can be used to calculate the dispersion. For example, calculate the dispersion for each parameter dimension separately. If the dispersion of all dimensions is 0, then the area corresponding to this window is determined to be a stable data area; or calculate the average value of the dispersions of each dimension, and when the average value is 0, it is determined to be a stable area. This multi-dimensional comprehensive analysis can more comprehensively reflect the stability of the data and avoid misjudgment caused by the accidental stability of a single dimension.

[0044] After determining the stable data area, it is necessary to optimize the real-time heterogeneous data matrix to generate an optimized matrix. The specific operation is to set the data points at the positions corresponding to the stable data area in the real-time heterogeneous data matrix to zero. Here, "setting to zero" is a data marking method used to distinguish stable data from unstable data, facilitating subsequent focused analysis of the unstable data area. For example, if the stable data area corresponds to the 1st to 5th rows and the 3rd column of the real-time heterogeneous data matrix, then set the values of all data points in this area to 0, while the data points in other unstable areas remain unchanged with their original values. In this way, the optimized matrix can highlight the unstable data that needs further analysis and reduce the interference of stable data on anomaly detection.

[0045] During the implementation process, the following key details need to be noted: First, the setting of the time length of the dynamic analysis window needs to balance the analysis accuracy and calculation efficiency. If the window is too short, the calculation of the dispersion may be unstable due to insufficient data volume; if the window is too long, short-term data fluctuations may be masked, resulting in misjudgment of the stable area. In practical applications, a reasonable window length can be determined through pre-experiments or industry experience. For example, in the scenario of coal mine gas monitoring, combined with the periodic characteristics of gas concentration changes, the window length can be set to 30 minutes. Second, the calculation method of the dispersion can be adjusted according to the data characteristics. For example, for a feature category matrix with a small number of categories, boolean values (0 or 1) can be used to represent whether it is a stable area. When different categories exist within the window, it is directly marked as an unstable area, simplifying the calculation process. Third, the continuity requirement of the stable data area needs to be clear, that is, it must be a continuous time interval to avoid the stable area being split due to the abnormality of individual data points. For example, if the dispersion of a certain window is 0, but the dispersion of the adjacent window is greater than 0, then the time point corresponding to this window cannot be used as a stable area alone and needs to be attributed to the continuous stable interval before and after (if any) or be treated as an unstable area.

[0046] In addition, the moving direction of the dynamic analysis window can be adjusted according to requirements. In addition to moving forward in chronological order, it can also move backward or move in both directions to ensure a comprehensive analysis of the data. In the scenario of real-time data update, the dynamic analysis window needs to have the ability to slide in real time, be able to make stability judgments on newly collected data points in a timely manner, update the stable data area and the optimized matrix, and ensure the timeliness of the analysis results.

[0047] The implementation method for screening stable data regions realizes the efficient identification of stable regions in heterogeneous coal mine data through dynamic analysis windows and dispersion calculations. This process can effectively filter out the relatively stable parts of the data without abnormal fluctuations, focusing the analysis on the unstable data regions, and laying a foundation for subsequent extraction of abnormal data clusters, calculation of abnormal characteristic factors, and positioning of key information regions.

[0048] Example 3:

[0049] In the implementation method for extracting abnormal data clusters, the core is to perform pattern recognition on the unstable data regions in the optimized matrix. By setting an upper limit on the number of patterns and calculating the characteristic means, the pattern cluster with the smallest characteristic mean is selected as the abnormal data cluster. The implementation process is described in detail below with specific examples: I. Definition of the optimized matrix and unstable data regions Suppose the real-time heterogeneous data matrix contains three types of parameters, namely gas concentration, carbon monoxide concentration, and fan speed, within 24 hours in a coal mine. Data is collected once per minute, forming a 24×3 matrix (rows represent time points, and columns represent parameter types). After screening the stable data regions through Example 2, it is found that the dispersion of the gas concentration data (the first column) from 2:00 to 4:00 in the morning (corresponding to rows 121 - 240 of the matrix) is 0, and it is determined as a stable data region. The corresponding positions in the optimized matrix are set to zero, and the remaining non-zero regions (such as the gas concentration data from 0:00 to 2:00 and from 4:00 to 24:00, as well as the carbon monoxide concentration and fan speed data for the entire period) are all unstable data regions and require extraction of abnormal data clusters.

[0050] II. Application of pattern recognition and clustering algorithms For the unstable data regions in the optimized matrix (taking the gas concentration parameter as an example), first set an upper limit K on the number of patterns. Suppose, based on the historical data patterns of coal mine safety monitoring, K = 3 is set, meaning that the unstable data is allowed to be divided into at most 3 pattern clusters. A clustering algorithm (such as the DBSCAN algorithm or the K-means algorithm) is used to cluster the data points in this region. Taking the K-means algorithm as an example, the specific steps are as follows: Initialize the clustering centers: Randomly select 3 unstable data points as the initial clustering centers (such as the gas concentration values at the 10th minute, 60th minute, and 180th minute).

[0051] Assign data points: Calculate the distances from each unstable data point to the 3 clustering centers and assign it to the cluster to which the nearest clustering center belongs.

[0052] Update the clustering centers: Recalculate the mean of the data points within each cluster as the new clustering center.

[0053] Iterative optimization: Repeat steps 2 - 3 until the cluster centers no longer change significantly or the preset number of iterations is reached, and finally obtain 3 pattern clusters 、 、 。

[0054] III. Feature Mean Calculation and Abnormal Data Cluster Screening Calculate the feature mean (i.e., the average value of gas concentration) of the data points within each pattern cluster: Cluster : Contains most of the data points from 0:00 - 1:00 (the 1st - 60th minutes), and the feature mean is ; Cluster : Contains some data points from 4:00 - 12:00 (the 241st - 720th minutes), and the feature mean is ; Cluster : Contains scattered data points from 12:00 - 24:00 (the 721st - 1440th minutes), and the feature mean is 。

[0055] Compare the feature means of the three clusters. The feature mean of cluster is the smallest ( ), so it is selected as the abnormal data cluster. The "abnormal" judgment logic here is based on professional knowledge in the field of coal mine safety: A significantly lower gas concentration than the normal fluctuation range may imply potential problems such as sensor failure, abnormal data transmission, or abnormal local ventilation, which need to be focused on.

[0056] IV. Example of Comprehensive Analysis of Multiple Parameter Dimensions If considering the non - stable data regions of carbon monoxide concentration (the 2nd column) and fan speed (the 3rd column) simultaneously, it is necessary to fuse the multi - dimensional data into feature vectors for clustering. For example, each data point is represented as a three - dimensional vector [gas concentration, carbon monoxide concentration, fan speed], and a clustering algorithm that supports multi - dimensional input (such as the Gaussian mixture model) is used for pattern recognition. Assuming the upper limit of the number of patterns is set to K = 4, after clustering, four pattern clusters are obtained, and the feature mean vectors of each cluster are respectively: (normal operating state); (low - load operating state); (significantly low feature mean); (high - concentration abnormal state).

[0057] At this time, it is necessary to define the "abnormal" standard according to the business logic. If the coal mine safety regulations stipulate that a fan speed lower than 1200 rpm belongs to an abnormally low speed, then cluster The average fan speed (1000 rpm) is significantly lower than the threshold, and the concentrations of gas and carbon monoxide are also low, which may reflect that the ventilation system failure leads to excessive gas dilution. Therefore, is determined as an abnormal data cluster. If there are multiple clusters with low feature means, factors such as time series continuity and spatial distribution (such as sensor deployment locations) can be further combined for comprehensive screening to ensure the accuracy of abnormal data clusters.

[0058] V. Key Details in the Implementation Process Basis for setting the upper limit of the number of modes: The value of K needs to be combined with data complexity and business requirements. For example, for short-term data (a few hours) of a single parameter (such as gas concentration), K can be set to 2 - 3; for multi-parameter data with a long time span, K can be appropriately increased to 5 - 8, but it is necessary to avoid fragmentation of the clustering results due to being too large, which affects the effective identification of abnormal clusters.

[0059] Selection of clustering algorithms: K-means algorithm: Suitable for scenarios where the data distribution is spherical and the cluster sizes are similar, with high computational efficiency, but sensitive to the initial clustering centers; DBSCAN algorithm: Suitable for discovering clusters of any shape, capable of effectively identifying noise points, but density thresholds (such as neighborhood radius and minimum number of points) need to be set; Gaussian mixture model: Suitable for probability density estimation of multi-dimensional data, capable of outputting the probability that a data point belongs to each cluster, suitable for scenarios with high uncertainty.

[0060] In practical applications, the optimal algorithm can be selected through cross-validation, or an ensemble learning method (such as combining the results of multiple algorithms) can be adopted to improve robustness.

[0061] Integration of business logic for anomaly determination: Whether the cluster with the minimum feature mean is an abnormal cluster needs to be combined with knowledge in the field of coal mine safety. For example, among the operating state parameters of equipment, if the feature means of current and voltage are lower than the rated values, it may indicate insufficient equipment load or a fault; while among the environmental parameters, if the feature means of temperature and humidity are abnormally low, it may be related to abnormalities in the ventilation and drainage systems. Therefore, the definition of abnormal data clusters requires the collaboration of domain experts and algorithm engineers to establish determination rules that conform to the actual scenario.

[0062] VI. Roles and Extensions of Abnormal Data Cluster Extraction The abnormal data clusters extracted through the above steps can be used as the target areas for subsequent calculation of abnormal feature factors. For example, the gas concentration abnormal cluster extracted in the example , its internal data points will be further analyzed for local deviation and adjacent window correlation entropy (see Example 4) to quantify the degree of abnormality of each data point. At the same time, the time distribution (such as concentrated in the afternoon 3:00-5:00) and spatial distribution (such as sensor data in a mining area) of abnormal data clusters can assist in locating the abnormal source at the physical level and provide accurate investigation direction for coal mine safety managers.

[0063] It should be noted that the extraction of abnormal data clusters is not completed in one go, but is dynamically adjusted as real-time data is updated. For example, when newly collected data points are classified into unstable areas, the clustering algorithm needs to be rerun to update the abnormal data clusters to ensure that the system can respond to data changes in real time and avoid missed detection or delayed detection.

[0064] Embodiment 4:

[0065] In the calculation implementation of the abnormal characteristic factor, it is necessary to target the data points in the abnormal data cluster, comprehensively consider the degree of deviation between their characteristic values and the mean value within the cluster, and the correlation in the time series, and obtain the quantitative abnormal characteristic factor through multi-step processing. The following is the specific implementation process: After determining the abnormal data cluster, it is necessary to calculate the characteristic mean of all data points in the cluster. The characteristic mean is the arithmetic mean of all data points in the abnormal data cluster in the corresponding parameter dimension, reflecting the concentration trend of the cluster data. For example, if the abnormal data cluster contains 10 data points of gas concentration in a coal mine: 0.15%, 0.18%, 0.12%, 0.14%, 0.16%, 0.13%, 0.17%, 0.11%, 0.19%, 0.10%, then the characteristic mean is the sum of these values divided by the number of data points, that is, .

[0066] The local deviation is generated based on the difference between the eigenvalue of each data point and the eigenmean. For each data point in the abnormal data cluster, its eigenvalue is recorded as (i=1,2,...,n, n is the total number of data points in the cluster), and the feature mean The difference is , the local deviation takes the absolute value of the difference, that is The absolute value operation is used to eliminate the influence of positive and negative signs and only retain the degree of deviation. For example, the characteristic value of a data point is 0.19%, and the difference between it and the mean value of 0.145% is +0.045%, and the local deviation is 0.045%; the characteristic value of another data point is 0.10%, the difference is -0.045%, and the local deviation is also 0.045%. The larger the local deviation, the more significant the difference between the data point and the average characteristic value in the cluster, and the higher the possibility of potential abnormality.

[0067] The Markov chain algorithm is used to calculate the correlation entropy values of each data point in the neighboring window. The Markov chain algorithm is based on the Markov assumption of data points in the time series, that is, the current state only depends on the previous state and is independent of earlier states. The neighboring window usually refers to the previous window and the next window adjacent to the current window in time. For example, if the current window is the window at time t, the neighboring windows are the windows at time t - 1 and t + 1 (assuming the window length is fixed). For each data point, it is necessary to analyze the state transition probability within the neighboring window and construct a state transition matrix. The state can be defined as the feature category to which the data point belongs (such as the categories divided in Embodiment 1), or directly use the discretized interval of the feature value as the state.

[0068] Taking the feature category as an example, assume that the data point belongs to the category at time t - 1, belongs to the category at time t, and belongs to the category at time t + 1. Then the state transition path is . By counting all possible state transition times, the probability of transitioning from state to state can be calculated to form a state transition matrix . The correlation entropy value is calculated based on the Shannon entropy formula, and the formula is: ; where reflects the uncertainty of the state transition of the data point within the neighboring window. The larger the entropy value, the higher the randomness of the state transition and the weaker the correlation of the data point in the time series; the smaller the entropy value, the more stable the state transition and the stronger the correlation.

[0069] When obtaining the abnormal feature factor by fusing the local deviation degree and the correlation entropy value, it is necessary to consider the physical meanings of the two: the local deviation degree reflects the spatial abnormality of the data point (the difference from the within-cluster mean), and the correlation entropy value reflects the stability of the time series (the strength of the correlation). The calculation formula of the abnormal feature factor needs to satisfy the logical relationship of being positively correlated with the local deviation degree and negatively correlated with the correlation entropy value. Assuming a linear combination method, the formula can be expressed as: , where represents the abnormal feature factor of the i-th data point; and are weight coefficients, and , , which are used to adjust the relative importance of the two indicators; is the local deviation degree, and the unit is the same as the feature value (such as ); It is the associated entropy value, with the unit of bit (bit).

[0070] Weight coefficient and The value of needs to be determined according to data characteristics and business requirements. For example, in the monitoring of the operating status of coal mine equipment, if more attention is paid to the instantaneous anomalies of data (such as sudden current changes), the weight of can be increased. If more attention is paid to the stability of data trends (such as continuous fluctuations in rotational speed), the weight of can be increased. In practical applications, reasonable values can be determined through domain expert experience or historical data training. For example, setting = 2, = 1 can make the influence of the local deviation degree on the abnormal feature factor more significant.

[0071] Taking specific data as an example, assume that the local deviation degree of a data point A in an abnormal data cluster is , and the associated entropy value is 1.2 bit; the local deviation degree of data point B is , and the associated entropy value is 0.8 bit. Substituting into the formula, we get: Abnormal feature factor of data point A: ; Abnormal feature factor of data point B: .

[0072] Since the abnormal feature factor is positively correlated with the local deviation degree and negatively correlated with the associated entropy value, the local deviation degree of A is higher (although the associated entropy value is also higher), and the absolute value of its abnormal feature factor is larger, indicating that its abnormal degree is relatively more significant.

[0073] In the implementation process, the following key details need to be noted: Calculation range of the feature mean: Only for the data points within the abnormal data cluster, excluding the stable data region or the data of other clusters, to ensure that the mean can truly reflect the feature level of the abnormal cluster.

[0074] Definition of the neighboring window: The window length needs to be consistent with the dynamic analysis window (Example 2), for example, both are 10 minutes, to ensure the consistency of the time scale. The number of neighboring windows can be expanded according to needs, such as including the first two and the last two windows, to capture the correlation within a longer time range.

[0075] State discretization method: If the feature value is continuous data (such as temperature, voltage), it needs to be discretized first (such as equal-width binning, equal-frequency binning), and converted into a finite number of states to construct the state transition matrix of the Markov chain. The number of discretization intervals needs to be reasonable to avoid information loss caused by too few intervals or a sharp increase in computational complexity caused by too many intervals.

[0076] Normalization of Abnormal Feature Factors: If it is necessary to make a horizontal comparison of abnormal feature factors in different parameter dimensions (such as the degree of abnormality of the comprehensive gas concentration and the fan speed), the can be normalized and mapped to the interval [0, 1]. The formula is: , where and are the minimum and maximum values of the abnormal feature factors of all data points in the abnormal data cluster, respectively. The normalized values are convenient for subsequent fusion calculations with other indicators (such as the stable feature index and the Manhattan distance).

[0077] The calculation of the abnormal feature factor realizes the quantitative evaluation of the degree of abnormality of data points by fusing the spatial deviation degree and the time correlation. This process not only considers the characteristic differences of data points themselves but also introduces the dynamic characteristics of the time series, making the abnormal detection more comprehensive and timely.

[0078] Example 5: In the implementation method of generating the stable feature index and the abnormal risk value, it is necessary to realize the quantitative evaluation of the potential abnormal risk in the stable data area through multi-dimensional data processing. The specific process is as follows: Calculate the stable feature index of the stable data area. For each stable data area, it is necessary to analyze the difference in the characteristic distribution between it and the abnormal data cluster. Taking the gas concentration parameter as an example, assume that the characteristic mean in a certain stable data area is (unit: ), and the characteristic mean of the abnormal data cluster is (unit: . Calculate the relative difference between the two: , This relative difference reflects the degree of deviation of the characteristic mean of the stable data area from that of the abnormal data cluster. If , the relative difference is negative, indicating that the characteristic value of the stable area is closer to the abnormal cluster level; if it is positive, it indicates that the characteristic value is higher than the abnormal cluster. Subsequently, extract the minimum value of this relative difference and zero, that is: , The logic of this step is that only when the characteristic mean of the stable area is lower than the abnormal cluster is it considered that there may be a potential abnormal tendency in this area and needs to be included in the evaluation; if it is higher than the abnormal cluster, its abnormal risk is not considered for the time being.

[0079] At the same time, it is necessary to calculate the characteristic entropy change in the stable data area. The characteristic entropy change measures the change in the degree of chaos of the characteristic distribution through the Shannon entropy formula. The formula is: , where To stabilize the probability of the occurrence of feature category k in data region j, where K is the total number of categories. The larger the entropy value, the more dispersed the feature distribution and the lower the stability. When fusing the minimum value and the feature entropy change, a linear combination method is adopted: , where, is the stable feature index, and are weight coefficients ( , ), used to adjust the influence degree of the two indicators. For example, if , , then the stable feature index increases with the decrease (more negative) of the relative difference and increases with the increase of the entropy change, indicating that the more the feature mean is close to the abnormal cluster and the more chaotic the distribution, the worse the stability.

[0080] Next, calculate the abnormal risk value. Taking a certain stable data region j as an example, it is necessary to first calculate the Manhattan distance between it and each core point m in the abnormal data cluster. The Manhattan distance is the sum of the absolute values of the coordinate differences in each dimension. Assuming that the coordinates of the data point in the two-dimensional feature space are , and the coordinates of the core point are , then: , Take the minimum Manhattan distance among all core points as the spatial distance index between this region and the abnormal cluster.

[0081] The calculation formula of the abnormal risk value is: ; where: represents the abnormal risk value of the j-th stable data region; M is the total number of core points in the abnormal data cluster; is the activation function (such as the Sigmoid function), used to convert the distance into an influence factor in the 0-1 interval. The closer the distance, the closer the output value is to 1, amplifying the risk contribution; is the distance attenuation coefficient (unit: ), adjusting the influence rate of the distance on the risk. The larger , the faster the influence of the distant core point decays; is the abnormal feature factor of the m-th core point (the calculation result from Example 4), reflecting the abnormal degree of the core point itself;

[0082] Taking specific data as an example: Assume that the Manhattan distances between a certain stable data region j and 3 core points in the abnormal data cluster are 5, 3, and 4 respectively, and the corresponding abnormal feature factors Are 0.8, 1.2, 1.0 respectively, the stable feature index = 0.6, the distance decay coefficient = 0.2, the activation function uses the Sigmoid function .

[0083] Then ; The product term is , substituting into the formula gives: ; The larger this value is, the higher the abnormal risk faced by the stable area indicates.

[0084] The key details in the implementation process are as follows: Cross-cluster comparison of feature means: The feature means of the stable data area and the abnormal data cluster need to be calculated on the same parameter dimension, for example, both are gas concentration or both are equipment current, to avoid logical deviation caused by cross-dimensional comparison.

[0085] Business logic of weight coefficients: and The values of need to be combined with coal mine safety standards. For example, in gas monitoring, the risk weight of the feature mean close to the abnormal cluster can be higher ( The absolute value is larger), while in equipment status monitoring, the weight of feature entropy change can be higher ( Larger).

[0086] Selection of core points: The core points of the abnormal data cluster can be determined by the density clustering algorithm, and the points with higher density within the cluster are selected as core points to ensure that they can represent the typical characteristics of the cluster.

[0087] Selection of activation functions: In addition to the Sigmoid function, the ReLU function or piecewise function can be used to directly truncate the long-distance influence according to the distance threshold to improve the calculation efficiency.

[0088] Multi-parameter fusion: If multi-dimensional data is involved (such as gas concentration, temperature, fan speed), the stable feature index of each dimension needs to be calculated separately, and the comprehensive index is synthesized through principal component analysis or weighted average, and then substituted into the abnormal risk value formula.

[0089] The calculation of the stable feature index and the abnormal risk value realizes the comprehensive evaluation of the potential risks of the stable data area by integrating the feature distribution difference, spatial distance and time correlation.

[0090] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0091] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A visualization processing method for heterogeneous data in coal mines combined with artificial intelligence, characterized in that, The method includes the following steps: Collect multi-source heterogeneous data above and below the coal mine, including environmental parameters and equipment operating status, and construct a real-time heterogeneous data matrix; Based on the dynamic relevance of each data point in the real-time heterogeneous data matrix, divide the data feature categories to generate a feature category matrix; preset a dynamic analysis window, and screen the stable data area according to the dispersion degree of the feature categories within the window to generate an optimized matrix; Perform pattern recognition on the unstable data area in the optimized matrix to extract abnormal data clusters; combine the deviation degree of each data point in the abnormal data cluster from the global feature mean value and the feature correlation entropy value within the adjacent window to calculate the abnormal feature factor of each data point; Analyze the feature distribution differences between the stable data area and the abnormal data cluster, and fuse the feature entropy change characteristics of the stable data area to generate the stable feature index of each stable data area; calculate the minimum Manhattan distance between the center of each stable data area and the core point of the abnormal data cluster, and combine the abnormal feature factor and the stable feature index to generate the abnormal risk value of each stable data area; Map the coordinates of the center of each stable data area in the real-time heterogeneous data matrix to the three-dimensional space of the coal mine, and locate the key information area in combination with the abnormal risk value.

2. The method for visualizing and processing heterogeneous data in coal mines combined with artificial intelligence according to claim 1, wherein The specific process of dividing the data feature categories is as follows: Use the logarithmic mapping result of the variances of all data points in the real-time heterogeneous data matrix as the category division step size; use the minimum feature point of the real-time heterogeneous data matrix as the category starting benchmark, divide continuous category intervals, and assign feature categories to each data point; The feature category matrix is composed of the feature categories of all data points.

3. The method for visualizing and processing heterogeneous data in coal mines combined with artificial intelligence according to claim 1, characterized in that, The process of screening the stable data area is specifically as follows: Calculate the dispersion degree of the data within the window when the dynamic analysis window moves in the feature category matrix, and screen the continuous area with a dispersion degree of 0 as the stable data area.

4. The method for visual processing of heterogeneous data in coal mines combined with artificial intelligence according to claim 1, characterized in that The optimized matrix is the real-time heterogeneous data matrix with the data points at the corresponding positions of the stable data area set to zero.

5. The method for visual processing of heterogeneous data in coal mines combined with artificial intelligence according to claim 1, characterized in that The specific process of extracting the abnormal data cluster is as follows: Set the upper limit of the number of patterns, and calculate the feature mean value of the data points within each pattern cluster; Select the pattern cluster with the smallest feature mean value as the abnormal data cluster.

6. The method for visual processing of heterogeneous data in coal mines combined with artificial intelligence according to claim 1, wherein, The calculation process of the abnormal feature factor includes: Calculate the feature mean value of all data points within the abnormal data cluster; generate the local deviation degree according to the difference between the feature value of each data point and the feature mean value; Use the Markov chain algorithm to calculate the correlation entropy value of each data point in the adjacent window; Fuse the local deviation degree and the correlation entropy value to obtain the abnormal feature factor; among them, the abnormal feature factor is positively correlated with the local deviation degree and negatively correlated with the correlation entropy value.

7. The method for visual processing of heterogeneous data in coal mines combined with artificial intelligence according to claim 1, wherein The process of generating the stable feature index of each stable data area is specifically as follows: Calculate the relative difference between the feature mean value within each stable data area and the feature mean value of the abnormal data cluster; extract the minimum value of the relative difference and zero; Fuse the minimum value and the feature entropy change within the stable data area to obtain the stable feature index.

8. The method for visual processing of heterogeneous data in coal mines combined with artificial intelligence according to claim 1, wherein The calculation formula of the abnormal risk value is: ; Wherein: represents the abnormal risk value of the j-th stable data region; represents the Manhattan distance between the j-th stable data region and the m-th core point in the abnormal data cluster; represents the abnormal characteristic factor of the m-th core point; represents the stability characteristic index of the j-th stable data region; M represents the total number of core points in the abnormal data cluster; is the distance attenuation coefficient; is the activation function.

9. The method for visualizing and processing heterogeneous data in coal mines combined with artificial intelligence according to claim 1, wherein, The specific process of locating the key information area is as follows: Map the real-time heterogeneous data matrix to the three-dimensional coordinate system of the coal mine to generate a feature distribution space; calculate the geometric center coordinates of the data points in each stable data area in the feature distribution space; Convert the geometric center coordinates into proportional coordinates of the actual position of the coal mine, where the abscissa is the ratio of the abscissa of the geometric center to the width of the coal mine, and the ordinate is the ratio of the ordinate of the geometric center to the depth of the coal mine; When the proportional coordinates fall on the boundary of the preset key area, mark the stable data area as the area to be analyzed; if the abnormal risk value of the area to be analyzed exceeds the preset threshold, determine the corresponding geometric center as the position of key information.

10. A coal mine heterogeneous data visualization identification device combined with artificial intelligence, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-9.

Citation Information

Patent Citations

  • Three-dimensional visualization method for multi-source heterogeneous data of power equipment

    CN114022638A

  • Unmanned aerial vehicle-based river hydrological sampling inspection method and system

    CN119151387A

  • Multi-dimensional data fusion analysis method for financial transaction anomaly detection

    CN119377902A

  • Coal mine fire and gas disaster risk prediction method and system

    CN119417243A

  • Local mine ecological risk evaluation comprehensive model construction method

    CN119918960A

Cited By

  • Abnormal behavior detection method and system for large-scale intelligent agent

    CN120892759A

  • A method and system for detecting abnormal behavior in large-scale intelligent agents

    CN120892759B

  • Mine ventilation safety intelligent monitoring method

    CN121273390A

  • Low-voltage transformer area line loss anomaly detection method, system and equipment based on artificial intelligence

    CN121347954A

  • Underground coal mine early warning system and method

    CN121354330A