Mine monitoring data analysis method and system based on pattern recognition

Abstract

The present invention relates to the field of mine monitoring, and in particular to a method and system for analyzing in-mine monitoring data based on pattern recognition. The present invention first obtains multidimensional data at different locations in the mine at each sampling moment, and composes all multidimensional data within a preset time period into a multidimensional data sequence; then, during clustering, obtains the initial distance between every two multidimensional data in the multidimensional data sequence, and then obtains the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor; uses the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance, and divides the multidimensional data into at least one cluster according to all the optimized distances; finally, according to the multidimensional data in each cluster, obtains the comprehensive abnormality score of each cluster, and screens the multidimensional data with abnormalities within the preset time period. The present invention can effectively cope with the complexity and variability in the mine environment and enhance the accuracy of identifying potential abnormal areas.

Description

Technical Field

[0001] The present invention relates to the field of mine monitoring, and in particular to a method and system for analyzing in-mine monitoring data based on pattern recognition. Background Art

[0002] Coal seam spontaneous combustion is a major hidden danger in coal mine safety production. Due to the complex mine environment, differences in geological structure and ventilation system lead to uneven diffusion of heat in space, making it impossible to detect coal seam spontaneous combustion in time. Therefore, early monitoring of coal seam spontaneous combustion is crucial to preventing mine fires and ensuring the safety of miners.

[0003] Traditional coal seam spontaneous combustion monitoring methods mainly rely on single-point temperature measurement combined with K-means clustering analysis anomaly detection systems, but this method has certain limitations in practical applications: first, the temperature rises extremely slowly in the early stages of coal seam spontaneous combustion, and may only rise by 0.1-0.2°C per day. This slight change can easily be mistaken by traditional methods as normal environmental fluctuations. Secondly, the diffusion of heat in space is non-uniform, and the traditional K-means clustering method is difficult to accurately reflect the overall temperature anomaly trend. As a result, traditional methods easily ignore potential spontaneous combustion risks, affecting the accuracy and reliability of early warning of coal seam spontaneous combustion.

[0004] Therefore, how to identify potential spontaneous combustion risks in complex and changing mining environments has become an urgent problem that needs to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method and system for analyzing in-mine monitoring data based on pattern recognition to solve the problem of how to identify potential spontaneous combustion risks in a complex and changeable mine environment.

[0006] In a first aspect, an embodiment of the present invention provides a method for analyzing in-mine monitoring data based on pattern recognition, the method comprising the following steps:

[0007] Acquire multidimensional data at different locations in the mine at each sampling time, and combine all multidimensional data within a preset time period into a multidimensional data sequence, wherein each multidimensional data includes temperature, sampling time and spatial position;

[0008] When clustering the multidimensional data in the multidimensional data sequence, respectively obtain the initial distance between each two multidimensional data in the multidimensional data sequence, for any initial distance, obtain the spatiotemporal correlation optimization factor of the initial distance according to the data difference between the two multidimensional data corresponding to the initial distance, and obtain the adaptive spatiotemporal weight factor of the initial distance according to the data features of the two multidimensional data corresponding to the initial distance;

[0009] Optimizing the initial distance using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain an optimized distance, obtaining an optimized distance corresponding to each of the initial distances, and dividing the multidimensional data in the multidimensional data sequence into at least one cluster according to all the optimized distances;

[0010] According to the multidimensional data in each of the clusters, a comprehensive anomaly score of each of the clusters is obtained, and according to the comprehensive anomaly score of each of the clusters, the multidimensional data with anomalies within a preset time period is screened.

[0011] Preferably, obtaining the spatiotemporal correlation optimization factor of the initial distance according to the data difference between the two multidimensional data corresponding to the initial distance includes:

[0012] Obtaining a sampling time difference between two multidimensional data corresponding to the initial distance, obtaining a ratio of a square value of the sampling time difference to a square value of a preset time scale parameter as a first ratio, performing inverse proportional normalization processing on the first ratio, and obtaining a result as a first inverse proportional normalized value;

[0013] Obtaining the Euclidean distance of the spatial position between the two multidimensional data corresponding to the initial distance, obtaining the ratio of the square value of the Euclidean distance to the square value of the preset spatial scale parameter as the second ratio, performing inverse proportional normalization processing on the second ratio, and obtaining the result as the second inverse proportional normalized value;

[0014] The product of the first inversely proportional normalized value and the second inversely proportional normalized value is obtained as a spatiotemporal correlation optimization factor of the initial distance.

[0015] Preferably, obtaining the adaptive spatiotemporal weight factor of the initial distance according to the data features of the two multidimensional data corresponding to the initial distance includes:

[0016] For any multidimensional data of the two multidimensional data corresponding to the initial distance, obtain the temperature difference between the multidimensional data and all multidimensional data in the multidimensional data sequence except the multidimensional data, form the multidimensional data whose temperature difference is less than a set temperature difference threshold into a neighbor set of the multidimensional data, and obtain the local clustering coefficient of the multidimensional data according to the neighbor set of the multidimensional data;

[0017] Obtaining a local time continuity coefficient of the multidimensional data according to the temperature of the multidimensional data at the same spatial position as the multidimensional data;

[0018] The local clustering coefficients and the local time continuity coefficients of the two multidimensional data corresponding to the initial distance are respectively obtained, and the adaptive spatiotemporal weight factor of the initial distance is obtained according to the local clustering coefficients and the local time continuity coefficients of the two multidimensional data corresponding to the initial distance.

[0019] Preferably, obtaining the local clustering coefficient of the multidimensional data according to the neighbor set of the multidimensional data includes:

[0020] Obtaining the number of multidimensional data in a neighbor set of the multidimensional data, calculating the difference between the number of multidimensional data in the neighbor set of the multidimensional data and a constant 1 to obtain a first difference, and obtaining the product of the first difference and the number of multidimensional data in the neighbor set of the multidimensional data as a first product;

[0021] Respectively obtain the temperature difference between every two multidimensional data in the neighbor set of the multidimensional data, and obtain the number of all temperature differences in the neighbor set of the multidimensional data that are less than a temperature difference threshold;

[0022] A ratio of the number of the temperature difference values ​​of a preset multiple to the first product is obtained, and the obtained ratio result is used as the local clustering coefficient of the multidimensional data.

[0023] Preferably, obtaining the local time continuity coefficient of the multidimensional data according to the temperature of the multidimensional data having the same spatial position as the multidimensional data comprises:

[0024] In the preset time period, the sampling moment of the multidimensional data and the preset short-term time before the sampling moment of the multidimensional data are obtained as the short-term preset time, the multidimensional data in the same spatial position as the multidimensional data in the short-term preset time are grouped into a short-term subsequence, and the temperature variance of the short-term subsequence is obtained according to the temperature of the multidimensional data in the short-term subsequence;

[0025] In the preset time period, a sampling moment of the multidimensional data and a preset long-term time before the sampling moment of the multidimensional data are obtained as the long-term preset time, and the multidimensional data having the same spatial position as the multidimensional data within the long-term preset time are combined into a long-term subsequence, and a temperature variance of the long-term subsequence is obtained according to the temperature of the multidimensional data in the long-term subsequence;

[0026] The ratio of the temperature variance of the short-term subsequence to the temperature variance of the long-term subsequence is obtained as a third ratio, and the difference between a constant 1 and the third ratio is obtained as a local time continuity coefficient of the multidimensional data.

[0027] Preferably, the step of obtaining the adaptive spatiotemporal weight factor of the initial distance according to the local clustering coefficient and the local time continuity coefficient of the two multidimensional data corresponding to the initial distance includes:

[0028] Obtaining a mean of the local clustering coefficients of the two multidimensional data corresponding to the initial distance as a first mean, and calculating an addition result of a constant 1 and the first mean to obtain a first addition result;

[0029] Obtaining a mean of local time continuity coefficients of two multidimensional data corresponding to the initial distance as a second mean, and calculating an addition result of a constant 1 and the second mean to obtain a second addition result;

[0030] Obtaining a temperature difference between two multidimensional data corresponding to the initial distance, obtaining a ratio of a square value of the temperature difference to a square value of a preset temperature scale parameter as a third ratio, performing inverse proportional normalization processing on the third ratio, obtaining a result as a third inverse proportional normalized value, calculating a difference between a constant 1 and the third inverse proportional normalized value to obtain a first difference, and obtaining a first difference of a preset ratio as a nonlinear difference;

[0031] Calculate the product of the first addition result, the second addition result and the nonlinear difference to obtain a second product, and obtain the product of a preset spatiotemporal weight factor parameter and the second product as the adaptive spatiotemporal weight factor of the initial distance.

[0032] Preferably, the optimizing the initial distance by using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance comprises:

[0033] Calculate the product of the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain a third product, calculate the difference between constant 1 and the third product to obtain a second difference, and obtain the product of the initial distance and the second difference as the optimized distance.

[0034] Preferably, obtaining a comprehensive anomaly score of each cluster according to the multidimensional data in each cluster includes:

[0035] For any cluster, the temperature standard score of the cluster is obtained according to the temperature of the multidimensional data in the cluster, the size standard score of the cluster is obtained according to the number of multidimensional data in the cluster, and the score difference between the temperature standard score of the cluster and the size standard score of the cluster is obtained as the comprehensive anomaly score of the cluster.

[0036] Preferably, screening the multidimensional data having abnormalities within a preset period of time according to the comprehensive abnormality score of each cluster includes:

[0037] The comprehensive anomaly score of each cluster is sorted in descending order, and a preset proportion of clusters are selected from the sorted clusters as abnormal clusters, and the multidimensional data in the abnormal clusters are used as the multidimensional data with anomalies within a preset time period.

[0038] In the second aspect, an embodiment of the present invention also provides an in-mine monitoring data analysis system based on pattern recognition, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that when the processor executes the computer program, the steps of the in-mine monitoring data analysis method based on pattern recognition as described in the first aspect are implemented.

[0039] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0040] The present invention acquires multidimensional data at different positions in a mine at each sampling moment, and forms a multidimensional data sequence with all multidimensional data within a preset time period; when clustering the multidimensional data in the multidimensional data sequence, respectively acquires the initial distance between every two multidimensional data in the multidimensional data sequence, and for any initial distance, acquires the spatiotemporal correlation optimization factor of the initial distance according to the data difference between the two multidimensional data corresponding to the initial distance, and acquires the adaptive spatiotemporal weight factor of the initial distance according to the data features of the two multidimensional data corresponding to the initial distance; optimizes the initial distance using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance, acquires the optimized distance corresponding to each of the initial distances, and divides the multidimensional data in the multidimensional data sequence into at least one clustering cluster according to all the optimized distances; acquires the comprehensive abnormality score of each clustering cluster according to the multidimensional data in each clustering cluster, and screens the multidimensional data with abnormalities within a preset time period. The present invention optimizes the initial distance by acquiring a spatiotemporal correlation optimization factor and an adaptive spatiotemporal weight factor, and captures the spatiotemporal relationship between multidimensional data through the spatiotemporal correlation optimization factor to identify those temperature anomalies that persist in time but change slowly; the adaptive spatiotemporal weight factor is used to adjust its sensitivity according to local dynamic characteristics, so as to more accurately capture complex temperature anomaly patterns; therefore, the present invention can effectively cope with the complexity and variability in the mine environment, whether in a well-ventilated main tunnel or in a poorly ventilated coal mining face, the method can maintain efficient anomaly detection capabilities, improve the detection sensitivity to tiny temperature changes, and enhance the accuracy of identifying potential abnormal areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 This is a method flow chart of a method for analyzing in-mine monitoring data based on pattern recognition provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0043] Embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0044] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0045] In order to illustrate the technical solution of the present invention, specific embodiments are provided below for illustration.

[0046] See also Figure 1 , is a method flow chart of a method for analyzing in-mine monitoring data based on pattern recognition provided by Embodiment 1 of the present invention, such as Figure 1 As shown, the method may include:

[0047] Step S101, acquiring multidimensional data at different locations in the mine at each sampling time, and composing all multidimensional data within a preset period into a multidimensional data sequence, wherein each multidimensional data includes temperature, sampling time and spatial position;

[0048] Temperature sensors are arranged at different locations in the mine, and the temperature data of the temperature sensors are collected every 5 minutes. At the same time, the position data of each sensor is recorded. The temperature, sampling time and spatial position of each sensor collected are recorded as a multi-dimensional data as follows: , obtain all multidimensional data within 30 days to form a multidimensional data sequence, which is not limited here and can be set according to the specific implementation scenario, where, is the i-th multidimensional data in the multidimensional data sequence, i is the sequence number of the multidimensional data in the multidimensional data sequence, is the temperature of the i-th multidimensional data in the multidimensional data sequence, is the spatial position of the i-th multidimensional data in the multidimensional data sequence, is the sampling time of the i-th multidimensional data in the multidimensional data sequence.

[0049] Step S102, when clustering the multidimensional data in the multidimensional data sequence, respectively obtain the initial distance between every two multidimensional data in the multidimensional data sequence, for any initial distance, obtain the spatiotemporal correlation optimization factor of the initial distance according to the data difference between the two multidimensional data corresponding to the initial distance, and obtain the adaptive spatiotemporal weight factor of the initial distance according to the data features of the two multidimensional data corresponding to the initial distance.

[0050] After obtaining the multidimensional data sequence, it is necessary to perform K-means clustering on the multidimensional data in the multidimensional data sequence. During the clustering process, the initial distance between each two multidimensional data in the multidimensional data sequence can be obtained. The traditional coal seam spontaneous combustion monitoring method performs cluster analysis based on the initial distance. According to the particularity of coal seam spontaneous combustion, the temperature rises extremely slowly in the initial stage of spontaneous combustion, only rising by 0.1-0.2°C per day. This slight change can easily be mistaken for normal environmental fluctuations. At the same time, the complex geological structure and ventilation system of the mine make the diffusion of heat in space non-uniform, resulting in a complex nonlinear distribution of temperature anomalies. For example, the temperature at a certain location in a coal seam with a depth of 500 meters begins to rise slowly. Due to the difference in thermal conductivity of the surrounding rock formations, the temperature distribution around this location will present an irregular shape. Some areas near this location will heat up faster, while other areas may have almost no change. This complex temperature distribution pattern makes it difficult for traditional coal seam spontaneous combustion monitoring methods that only rely on the temperature value of a single location to accurately capture the abnormal trend of coal seam spontaneous combustion.

[0051] In order to solve this problem, it is necessary to obtain the spatiotemporal correlation optimization factor and adaptive spatiotemporal weight factor of the initial distance. The spatiotemporal correlation optimization factor can be used to capture the spatiotemporal relationship between multidimensional data to identify those temperature anomalies that persist in time but change slowly. The adaptive spatiotemporal weight factor can be used to adjust its sensitivity according to local dynamic characteristics to more accurately capture complex temperature anomaly patterns.

[0052] In the early monitoring of coal seam spontaneous combustion, it is necessary to accurately identify temperature anomalies that persist in time but change slowly in a complex mine environment. This problem mainly stems from the slowness of temperature change, the non-uniformity of heat diffusion, and the correlation between time and space. Taking a coal mining face with a depth of 800 meters as an example, the temperature of an area only rises by 0.1°C every day. This small change can easily be mistaken for normal environmental fluctuations or measurement errors. Traditional monitoring methods generally take weeks or even months to confirm whether this change is statistically significant, and such a delay may lead to missing the best time for intervention. This is because the geological structure of the mine is complex and the rock composition and crack distribution in different areas are different, resulting in non-uniform diffusion of heat in space. For example, around a spontaneous combustion location, along the direction of the coal seam cracks, the heat may spread faster, and in the direction perpendicular to the cracks, the heat may spread slowly. This non-uniformity makes it difficult for traditional methods that rely only on the temperature value of a single location to accurately describe the overall temperature anomaly trend. In addition, coal seam spontaneous combustion is a continuous process in both time and space. Temperature changes in one area often affect the surrounding areas, and this impact takes a certain amount of time to appear. If a slight temperature rise is detected in a corner of a coal mining face, then over time, this temperature anomaly will spread to the surrounding areas.

[0053] In order to identify those temperature anomalies that persist in time but change slowly, it is necessary to obtain the spatiotemporal correlation optimization factor of the initial distance. The spatiotemporal correlation optimization factor not only considers the temperature value of a single multidimensional data, but also integrates the relationship between the multidimensional data and other multidimensional data in time and space. For example, if the temperature at a monitoring location has been 0.1°C higher than the historical average in the past week, then even if this difference is very small, it will be considered to be of concern. This method can effectively capture those slow but continuous temperature increase trends; similarly, even if a single monitoring location does not show obvious temperature anomalies, multiple monitoring locations around it have slight temperature increases, which will also be considered worthy of attention. Therefore, for any initial distance, the method for obtaining the spatiotemporal correlation optimization factor of the initial distance is as follows:

[0054] Obtaining a sampling time difference between two multidimensional data corresponding to the initial distance, obtaining a ratio of a square value of the sampling time difference to a square value of a preset time scale parameter as a first ratio, performing inverse proportional normalization processing on the first ratio, and obtaining a result as a first inverse proportional normalized value;

[0055] Obtaining the Euclidean distance of the spatial position between the two multidimensional data corresponding to the initial distance, obtaining the ratio of the square value of the Euclidean distance to the square value of the preset spatial scale parameter as the second ratio, performing inverse proportional normalization processing on the second ratio, and obtaining the result as the second inverse proportional normalized value;

[0056] The product of the first inversely proportional normalized value and the second inversely proportional normalized value is obtained as a spatiotemporal correlation optimization factor of the initial distance.

[0057] In one embodiment, multidimensional data and multidimensional data As an example, calculate the multidimensional data and multidimensional data Optimization factor of spatiotemporal correlation of initial distance:

[0058]

[0059] in, For multidimensional data and multidimensional data Optimization factor of spatiotemporal correlation of initial distance; is the i-th multidimensional data of the multidimensional data sequence; is the j-th multidimensional data in the multidimensional data sequence; i and j are the serial numbers of the multidimensional data in the multidimensional data sequence; , Multidimensional data and multidimensional data The sampling time of is the preset time scale parameter; , Multidimensional data and multidimensional data spatial location; is the preset spatial scale parameter; ‖ ‖ is the Euclidean norm; exp() is an exponential function with a natural constant as the base.

[0060] It should be noted that the preset time scale parameter , controlling time correlation The decay rate of The increase in time correlation Approaching 0, the spatiotemporal correlation optimization factor of the initial distance Reduce; preset spatial scale parameters , controlling spatial correlation The decay rate increases with the spatial distance The increase in spatial correlation Approaching 0, the spatiotemporal correlation optimization factor of the initial distance Reduce; preset time scale parameters Set to 1440min, preset spatial scale parameters It is set to 50m. There is no limit here and it can be implemented according to specific scenarios.

[0061] For example, in a coal mining face 800 meters deep, even if the temperature rises by only 0.1°C per day, by adjusting the time scale parameter The spatiotemporal correlation optimization factor can be kept at a high value over a longer time span, which means that even if two multidimensional data are separated by several days, their temporal correlation is still strong. Therefore, when the temperature continues to rise, even if the rise is small, the spatiotemporal correlation optimization factor will remain at a high value, making this small but continuous change stand out. For the problem of non-uniform heat diffusion, when heat spreads faster along the direction of coal seam fractures, by adjusting the spatial scale parameter , select the spatial scale parameter according to the sensor setting distance The parameter value makes the multidimensional data with farther spatial locations still have strong correlation, and the spatiotemporal correlation optimization factor will also maintain a high value, making it easier for the multidimensional data along the crack direction to be clustered together, thus reflecting the rapid propagation of heat in this direction; if two multidimensional data are close in time and space, their spatiotemporal correlation optimization factors will be very high, which means that in the collected multidimensional data sequence, if the temperature of an area starts to rise and over time, the correlation of the multidimensional data in the surrounding areas will be very large, reflecting the diffusion process of temperature anomalies; for short-term, local temperature fluctuations (such as temporary heating caused by equipment operation), although the spatial correlation may be high, the temporal correlation will be low. In this case, the value of the spatiotemporal correlation optimization factor will not be high, avoiding the problem of misjudging short-term fluctuations as danger signals.

[0062] There are also the following problems in the scenario of coal seam spontaneous combustion monitoring. Since the temperature distribution in the mine environment is affected by many factors and these factors vary in different regions and periods, for example, in a large mine, the temperature in the area close to the main ventilation tunnel is relatively stable, and the temperature in the coal mining face far away from the ventilation system is more likely to accumulate and fluctuate. Coal seams at different depths often show different temperature change patterns due to the influence of ground pressure and geothermal heat. In addition, production activities in the mine will also have a significant impact on the temperature distribution. In the intensive mining area, the operation of mechanical equipment and the release of coal seam stress cause the temperature to show more complex nonlinear changes. Therefore, if the same temperature anomaly judgment standard is used for all areas, a large number of false alarms will be generated in some areas, while important abnormal signals may be missed in other areas. For example, a temperature rise of 0.5°C in a well-ventilated main tunnel does not touch the abnormal warning line, but the same temperature rise in a closed goaf indicates the occurrence of safety hazards.

[0063] In order to deal with the above problems, it is necessary to obtain an adaptive spatiotemporal weight factor of the initial distance. The adaptive spatiotemporal weight factor adjusts its sensitivity according to the local dynamic characteristics, so as to more accurately capture the complex temperature anomaly pattern. There are two key factors to consider when obtaining the adaptive spatiotemporal weight factor: local clustering characteristics and temporal continuity. The local clustering characteristics reflect the degree of aggregation of temperature data in the local range. Around the potential spontaneous combustion location, it is easy to observe that the temperature data shows a certain degree of aggregation. For example, in an 800-meter-deep coal mining face, if multiple monitoring locations in a 10m×10m area show a slight but consistent temperature rise, then even if this increase is not significant in the entire mine, it is also a signal worthy of attention. Temporal continuity is obtained based on the persistence of temperature changes. Coal seam spontaneous combustion is usually a slow but continuous process. Special attention should be paid to those temperature anomalies that are small but persistent. That is to say, if the temperature at a monitoring location has been 0.2°C higher than its long-term average for the past two weeks, even if this difference seems small, it may indicate a potential problem.

[0064] By comprehensively considering the above factors, the adaptive spatiotemporal weight factor can automatically adjust its judgment criteria according to the characteristics of the current data. When detecting areas with high local concentration, it will increase sensitivity and more easily capture small but potentially dangerous temperature changes. In areas with more uniform temperature distribution, it will reduce sensitivity accordingly, reducing the possibility of false alarms. In actual scenarios, when a mine starts new mining activities or changes ventilation strategies, the temperature distribution pattern may change significantly. Traditional fixed parameter methods require manual adjustments to adapt to these changes, while the adaptive spatiotemporal weight factor can automatically adjust its parameters to maintain efficient anomaly detection capabilities, and can more accurately identify various types of temperature anomalies. Whether it is a slow but continuous temperature rise or a local sudden temperature rise, it can be appropriately classified, improving the ability to identify potential coal seam spontaneous combustion risks at an early stage.

[0065] Among them, the method for obtaining the adaptive spatiotemporal weight factor of the initial distance is as follows:

[0066] (1) For any multidimensional data among the two multidimensional data corresponding to the initial distance, obtain the temperature difference between the multidimensional data and all multidimensional data in the multidimensional data sequence except the multidimensional data, form the multidimensional data whose temperature difference is less than a set temperature difference threshold into a neighbor set of the multidimensional data, and obtain the local clustering coefficient of the multidimensional data according to the neighbor set of the multidimensional data.

[0067] Wherein, obtaining the local clustering coefficient of the multidimensional data includes:

[0068] Obtaining the number of multidimensional data in a neighbor set of the multidimensional data, calculating the difference between the number of multidimensional data in the neighbor set of the multidimensional data and a constant 1 to obtain a first difference, and obtaining the product of the first difference and the number of multidimensional data in the neighbor set of the multidimensional data as a first product;

[0069] Respectively obtain the temperature difference between every two multidimensional data in the neighbor set of the multidimensional data, and obtain the number of all temperature differences in the neighbor set of the multidimensional data that are less than a temperature difference threshold;

[0070] A ratio of the number of the temperature difference values ​​of a preset multiple to the first product is obtained, and the obtained ratio result is used as the local clustering coefficient of the multidimensional data.

[0071] In one embodiment, multidimensional data and multidimensional data For example, for multidimensional data , get multidimensional data Divide multidimensional data by multidimensional data sequence The temperature difference of all multidimensional data except Multidimensional data composed of The neighbor set of , where the temperature difference threshold Set to 0.5°C. There is no limit here. It can be set according to the specific implementation scenario. Get the neighbor set The temperature difference between each two multidimensional data in the neighbor set All the values ​​less than the temperature difference threshold The temperature difference values ​​of form a temperature difference sequence, the number of temperature differences in the temperature difference sequence is recorded as M, and the neighbor set is obtained. The number of multidimensional data in the calculation of multidimensional data The local clustering coefficient of is:

[0072]

[0073] in, For multidimensional data The local clustering coefficient of is the i-th multidimensional data in the multidimensional data sequence; i is the sequence number of the multidimensional data in the multidimensional data sequence; For multidimensional data The set of neighbors of ; For multidimensional data The neighbor set The number of multidimensional data in the temperature difference sequence; M is the number of temperature differences in the temperature difference sequence (i.e., the number of multidimensional data The neighbor set The number of all temperature difference values ​​that are less than the temperature difference threshold.

[0074] It should be noted that the local clustering coefficient reflects the degree of aggregation of multidimensional data in a local range. When the number of multidimensional data in the neighbor set is certain, the more temperature difference values ​​in the neighbor set that are less than the temperature difference threshold (that is, the more temperature difference values ​​in the temperature difference sequence), the larger the local clustering coefficient, and the higher the degree of aggregation of the multidimensional data within the range of the neighbor set; the fewer temperature difference values ​​in the neighbor set that are less than the temperature difference threshold (that is, the fewer temperature difference values ​​in the temperature difference sequence), the smaller the local clustering coefficient, and the lower the degree of aggregation of the multidimensional data within the range of the neighbor set.

[0075] (2) Obtaining a local time continuity coefficient of the multidimensional data according to the temperature of the multidimensional data having the same spatial position as the multidimensional data.

[0076] In the preset time period, the sampling moment of the multidimensional data and the preset short-term time before the sampling moment of the multidimensional data are obtained as the short-term preset time, the multidimensional data in the same spatial position as the multidimensional data in the short-term preset time are grouped into a short-term subsequence, and the temperature variance of the short-term subsequence is obtained according to the temperature of the multidimensional data in the short-term subsequence;

[0077] In the preset time period, a sampling moment of the multidimensional data and a preset long-term time before the sampling moment of the multidimensional data are obtained as the long-term preset time, and the multidimensional data having the same spatial position as the multidimensional data within the long-term preset time are combined into a long-term subsequence, and a temperature variance of the long-term subsequence is obtained according to the temperature of the multidimensional data in the long-term subsequence;

[0078] The ratio of the temperature variance of the short-term subsequence to the temperature variance of the long-term subsequence is obtained as a third ratio, and the difference between a constant 1 and the third ratio is obtained as a local time continuity coefficient of the multidimensional data.

[0079] In one embodiment, multidimensional data For example, multidimensional data The sampling time is , within the preset period, obtain and The previous 24 hours is used as the short-term preset time, and the short-term preset time is combined with the multidimensional data The multidimensional data with the same spatial position are combined into a short-term subsequence, and the temperature variance of the multidimensional data in the short-term subsequence is calculated according to the temperature of the multidimensional data in the short-term subsequence; and The previous 7 days are used as the long-term preset time, and the long-term preset time is combined with the multidimensional data The multidimensional data with the same spatial position form a long-term subsequence, and the temperature variance of the multidimensional data in the long-term subsequence is calculated according to the temperature of the multidimensional data in the long-term subsequence; among them, 24h is the preset short-term time, and 7 days is the preset long-term time. There is no restriction here and it can be set according to the specific implementation scenario.

[0080] Calculating multidimensional data The local time continuity coefficient of :

[0081]

[0082] in, For multidimensional data The local time continuity coefficient of ; is the i-th multidimensional data in the multidimensional data sequence; i is the sequence number of the multidimensional data in the multidimensional data sequence; is the temperature variance of the short-term subseries; is the temperature variance of the long-term subseries; For multidimensional data temperature.

[0083] It should be noted that multidimensional data The local time continuity coefficient It is the ratio of the difference between the long-term variance and the short-term variance to the long-term variance. In the early stage of autoignition, when the temperature rises slowly but continuously, the long-term variance will gradually increase, and the short-term variance will be relatively small. The value will approach 1, indicating that there is a continuous trend of change; when the temperature remains stable for a long time, the temperature variance of the short-term subsequence Approximately equal to the temperature variance of the long-term subsequence , The value will approach 0, indicating that there is no obvious continuous trend.

[0084] (3) respectively obtaining local clustering coefficients and local time continuity coefficients of two multidimensional data corresponding to the initial distance, and obtaining an adaptive spatiotemporal weight factor of the initial distance according to the local clustering coefficients and local time continuity coefficients of the two multidimensional data corresponding to the initial distance.

[0085] Specifically, according to the local clustering coefficient calculation formula and the local time continuity coefficient calculation formula, the local clustering coefficient and the local time continuity coefficient of the two multidimensional data corresponding to the initial distance are obtained respectively; the mean of the local clustering coefficients of the two multidimensional data corresponding to the initial distance is obtained as the first mean, and the addition result of the constant 1 and the first mean is calculated to obtain the first addition result; the mean of the local time continuity coefficients of the two multidimensional data corresponding to the initial distance is obtained as the second mean, and the addition result of the constant 1 and the second mean is calculated to obtain the second addition result; the two multidimensional data corresponding to the initial distance are obtained. The method comprises the following steps: obtaining a temperature difference between two adjacent portions of the initial distance, obtaining a ratio of the square value of the temperature difference to the square value of a preset temperature scale parameter as a third ratio, performing inverse normalization on the third ratio, and obtaining a result as a third inverse normalized value, calculating a difference between a constant 1 and the third inverse normalized value to obtain a first difference, and obtaining the first difference of a preset ratio as a nonlinear difference; calculating the product between the first addition result, the second addition result and the nonlinear difference to obtain a second product, and obtaining the product between a preset spatiotemporal weight factor parameter and the second product as the adaptive spatiotemporal weight factor of the initial distance.

[0086] In one embodiment, multidimensional data and multidimensional data As an example, calculate the multidimensional data and multidimensional data Adaptive spatiotemporal weight factor for the initial distance:

[0087]

[0088] in, For multidimensional data and multidimensional data Adaptive spatiotemporal weighting factor for the initial distance of ; is the i-th multidimensional data of the multidimensional data sequence; is the j-th multidimensional data in the multidimensional data sequence; i and j are the serial numbers of the multidimensional data in the multidimensional data sequence; is the preset spatiotemporal weight factor parameter; , Multidimensional data and multidimensional data The local clustering coefficient of , Multidimensional data and multidimensional data The local time continuity coefficient of ; , Multidimensional data and multidimensional data Temperature; is the preset temperature scale parameter, is the Euclidean norm; exp() is an exponential function with a natural constant as the base.

[0089] It should be noted that the nonlinear term Makes the distance calculation more sensitive to temperature differences. When the temperature difference is small, the value of the nonlinear term is close to 0. As the temperature difference increases, the value of the nonlinear term gradually approaches 1. Preset temperature scale parameters Set to 0.1°C, preset spatiotemporal weight factor parameters It is set to 0.5, and there is no restriction here. It can be set according to the specific implementation scenario. The local clustering coefficient reflects the degree of aggregation of multidimensional data in the local range. The greater the degree of aggregation of multidimensional data in the local range, the greater the adaptive spatiotemporal weight factor. The local time continuity coefficient reflects the continuity of temperature change. The greater the continuity of temperature change in the local range, the greater the adaptive spatiotemporal weight factor.

[0090] At this point, the spatiotemporal correlation optimization factor of the initial distance and the adaptive spatiotemporal weight factor can be obtained.

[0091] Step S103, optimizing the initial distance using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain an optimized distance, obtaining an optimized distance corresponding to each of the initial distances, and dividing the multidimensional data in the multidimensional data sequence into at least one cluster according to all optimized distances.

[0092] After obtaining the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor of the initial distance, the initial distance is optimized using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance.

[0093] Specifically, the product of the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor is calculated to obtain a third product, the difference between the constant 1 and the third product is calculated to obtain a second difference, and the product of the initial distance and the second difference is obtained as the optimized distance.

[0094] In one embodiment, multidimensional data and multidimensional data Take the initial distance as an example to calculate the optimized distance:

[0095]

[0096] in, For multidimensional data and multidimensional data The optimized distance of For multidimensional data and multidimensional data The initial distance; is the i-th multidimensional data of the multidimensional data sequence; is the j-th multidimensional data in the multidimensional data sequence; i and j are the serial numbers of the multidimensional data in the multidimensional data sequence; For multidimensional data and multidimensional data Adaptive spatiotemporal weighting factor of the initial distance; For multidimensional data and multidimensional data The spatiotemporal correlation optimization factor of the initial distance, 1 represents a constant.

[0097] It should be noted that the final optimized distance takes into account the temporal and spatial correlation of multidimensional data by introducing the spatiotemporal correlation optimization factor, and dynamically adjusts its sensitivity according to the characteristics of the current data by introducing the adaptive spatiotemporal weight factor. When areas with highly localized aggregation (high LC values) or areas with persistently abnormal temperatures (high TC values) are detected, the adaptive spatiotemporal weight factor will increase accordingly, making it easier to capture potential anomalies in these areas.

[0098] After obtaining the optimized distance after the initial distance is optimized, the optimized distance corresponding to each initial distance is obtained, and the multidimensional data sequence is continued to be clustered by K-means according to all the optimized distances, and the multidimensional data in the multidimensional data sequence is divided into at least one cluster cluster. Among them, the selection of the number of clusters K is determined according to the elbow rule. Since the traditional K-means clustering and the elbow rule are both existing methods, the detailed step description is not given here.

[0099] At this point, clusters divided according to the optimized distance can be obtained.

[0100] Step S104, obtaining a comprehensive anomaly score of each cluster according to the multidimensional data in each cluster, and screening the multidimensional data with anomalies within a preset time period according to the comprehensive anomaly score of each cluster.

[0101] K-means clustering by optimizing distance enhances the classification effect of data with potential risk of spontaneous combustion compared with traditional clustering. From the results, data clusters with abnormal characteristics will have a smaller scale and slightly higher temperature than ordinary clusters, and potential risk of spontaneous combustion is often a minority in actual mines. Therefore, it is necessary to calculate the characteristics of each cluster, including the average temperature and cluster size, and calculate the Z-score (standard score) of each cluster feature. The comprehensive abnormal score of each cluster is calculated through the Z-score (standard score) of each cluster feature, and the degree of deviation of the cluster from the overall distribution is evaluated. The abnormal detection of the cluster results is completed, and the multi-dimensional data with abnormalities is screened out, thereby completing the monitoring of potential risk of spontaneous combustion in the mine.

[0102] Among them, the calculation of the comprehensive anomaly score of the cluster includes:

[0103] For any cluster, the temperature standard score of the cluster is obtained according to the temperature of the multidimensional data in the cluster, the size standard score of the cluster is obtained according to the number of multidimensional data in the cluster, and the score difference between the temperature standard score of the cluster and the size standard score of the cluster is obtained as the comprehensive anomaly score of the cluster.

[0104] In one implementation, taking the i-th cluster as an example, the standard score of the cluster feature is calculated as follows:

[0105]

[0106] in , are the temperature standard score and size standard score of the i-th cluster respectively; i is the serial number of the multiple clusters divided; is the average temperature of the i-th cluster; and are the mean and standard deviation of the average temperature of all clusters respectively; is the number of multidimensional data in the i-th cluster; and are the mean and standard deviation of all cluster sizes, respectively.

[0107] It should be noted that the larger the average temperature of the cluster is, the larger the corresponding temperature standard score is; the more multidimensional data there are in the cluster, the larger the corresponding size standard score is.

[0108] Furthermore, after obtaining the standard score of the clustering feature, the comprehensive anomaly score of the cluster cluster is calculated according to the standard score of the clustering feature, including:

[0109]

[0110] in, is the comprehensive anomaly score of the i-th cluster; , are the temperature standard score and size standard score of the i-th cluster respectively; i is the sequence number of the multiple clusters divided.

[0111] It should be noted that the larger the average temperature of the cluster, the larger the corresponding temperature standard score and the higher the comprehensive anomaly score; the more multidimensional data in the cluster, the larger the corresponding size standard score and the lower the comprehensive anomaly score.

[0112] Obtain the comprehensive anomaly score of each cluster, sort the comprehensive anomaly scores of each cluster in descending order, select a preset proportion of clusters from the sorted clusters as abnormal clusters, and use the multidimensional data in the abnormal clusters as the multidimensional data that has abnormalities within a preset period of time. The preset proportion is set to 5%, which is not limited here and can be set according to the specific implementation scenario.

[0113] At this point, the multidimensional data of anomalies occurring within the preset time period is obtained by clustering the comprehensive anomaly scores of the clusters.

[0114] In summary, the present embodiment acquires multidimensional data at different locations in the mine at each sampling moment, and groups all multidimensional data within a preset time period into a multidimensional data sequence; when clustering the multidimensional data in the multidimensional data sequence, respectively acquires the initial distance between each two multidimensional data in the multidimensional data sequence, and for any initial distance, acquires the spatiotemporal correlation optimization factor of the initial distance according to the data difference between the two multidimensional data corresponding to the initial distance, and acquires the adaptive spatiotemporal weight factor of the initial distance according to the data features of the two multidimensional data corresponding to the initial distance; optimizes the initial distance using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance, acquires the optimized distance corresponding to each of the initial distances, and divides the multidimensional data in the multidimensional data sequence into at least one clustering cluster according to all the optimized distances; acquires the comprehensive anomaly score of each clustering cluster according to the multidimensional data in each clustering cluster, and filters the multidimensional data with anomalies within a preset time period. This embodiment optimizes the initial distance by obtaining a spatiotemporal correlation optimization factor and an adaptive spatiotemporal weight factor, and captures the spatiotemporal relationship between multidimensional data through the spatiotemporal correlation optimization factor to identify those temperature anomalies that persist in time but change slowly; the adaptive spatiotemporal weight factor is used to adjust its sensitivity according to local dynamic characteristics to more accurately capture complex temperature anomaly patterns; therefore, this embodiment can effectively cope with the complexity and variability in the mine environment, whether in a well-ventilated main tunnel or in a poorly ventilated coal mining face, the method can maintain efficient anomaly detection capabilities, improve the detection sensitivity of small temperature changes, and enhance the accuracy of identifying potential abnormal areas.

[0115] Based on the same inventive concept as the above method, an embodiment of the present invention also provides an in-mine monitoring data analysis system based on pattern recognition, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that the processor implements the steps of the above-mentioned in-mine monitoring data analysis method based on pattern recognition when executing the computer program.

[0116] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for analyzing in-mine monitoring data based on pattern recognition, characterized in that: The method comprises: Acquire multidimensional data at different locations in the mine at each sampling moment, and combine all multidimensional data within a preset time period into a multidimensional data sequence, wherein each multidimensional data includes temperature, sampling moment, and spatial position; When clustering the multidimensional data in the multidimensional data sequence, respectively obtaining an initial distance between each two multidimensional data in the multidimensional data sequence, obtaining a spatiotemporal correlation optimization factor of any initial distance based on a data difference between the two multidimensional data corresponding to the any initial distance, and obtaining an adaptive spatiotemporal weight factor of the any initial distance based on data features of the two multidimensional data corresponding to the any initial distance; Optimizing any of the initial distances using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain an optimized distance, obtaining an optimized distance corresponding to each of the initial distances, and dividing the multidimensional data in the multidimensional data sequence into at least one cluster based on all the optimized distances; Obtaining a comprehensive anomaly score for each cluster based on the multidimensional data in each cluster, and screening the multidimensional data that exhibits anomalies within a preset time period based on the comprehensive anomaly score for each cluster; The obtaining, based on the data difference between the two multidimensional data corresponding to the any initial distance, a spatiotemporal correlation optimization factor of the any initial distance includes: Obtaining a sampling time difference between two multidimensional data corresponding to any initial distance, obtaining a ratio of a square value of the sampling time difference to a square value of a preset time scale parameter as a first ratio, performing inverse proportional normalization processing on the first ratio, and obtaining a result as a first inverse proportional normalized value; Obtaining a Euclidean distance of a spatial position between two multidimensional data corresponding to any of the initial distances, obtaining a ratio of a square value of the Euclidean distance to a square value of a preset spatial scale parameter as a second ratio, performing inverse proportional normalization processing on the second ratio, and obtaining a result as a second inverse proportional normalized value; Obtaining a product of the first inversely proportional normalized value and the second inversely proportional normalized value as a spatiotemporal correlation optimization factor of any initial distance; The step of obtaining the adaptive spatiotemporal weight factor of any initial distance according to the data features of the two multidimensional data corresponding to any initial distance includes: For any multidimensional data of the two multidimensional data corresponding to any initial distance, obtaining a temperature difference between the any multidimensional data and all multidimensional data in a multidimensional data sequence except the any multidimensional data, forming a neighbor set of the any multidimensional data with the temperature difference less than a set temperature difference threshold, and obtaining a local clustering coefficient of the any multidimensional data based on the neighbor set of the any multidimensional data; Obtaining a local time continuity coefficient of any multidimensional data according to a temperature of multidimensional data having the same spatial position as the any multidimensional data; Respectively obtaining local clustering coefficients and local time continuity coefficients of two multidimensional data corresponding to any initial distance, and obtaining an adaptive spatiotemporal weight factor of any initial distance according to the local clustering coefficients and local time continuity coefficients of the two multidimensional data corresponding to any initial distance; The optimizing any one of the initial distances by using the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain the optimized distance includes: Calculate the product of the spatiotemporal correlation optimization factor and the adaptive spatiotemporal weight factor to obtain a third product, calculate the difference between constant 1 and the third product to obtain a second difference, and obtain the product of any initial distance and the second difference as the optimized distance.

2. The method for analyzing in-mine monitoring data based on pattern recognition according to claim 1, characterized in that: The obtaining of the local clustering coefficient of any multidimensional data according to the neighbor set of any multidimensional data includes: Obtaining the number of multidimensional data in a neighbor set of any multidimensional data, calculating the difference between the number of multidimensional data in the neighbor set of any multidimensional data and a constant 1 to obtain a first difference, and obtaining the product of the first difference and the number of multidimensional data in the neighbor set of any multidimensional data as a first product; respectively obtaining the temperature difference between each two multidimensional data in the neighbor set of any multidimensional data, and obtaining the number of all temperature differences in the neighbor set of any multidimensional data that are less than a temperature difference threshold; A ratio of the number of the temperature difference values ​​of a preset multiple to the first product is obtained, and the obtained ratio result is used as the local clustering coefficient of the any multidimensional data.

3. The method for analyzing in-mine monitoring data based on pattern recognition according to claim 1, characterized in that: The obtaining of the local time continuity coefficient of any multidimensional data according to the temperature of the multidimensional data having the same spatial position as the any multidimensional data comprises: Within the preset time period, obtaining a sampling moment of any multidimensional data and a preset short-term time before the sampling moment of any multidimensional data as a short-term preset time, forming a short-term subsequence of multidimensional data having the same spatial position as the multidimensional data within the short-term preset time, and obtaining a temperature variance of the short-term subsequence based on the temperature of the multidimensional data in the short-term subsequence; Within the preset time period, obtaining a sampling moment of any multidimensional data and a preset long-term time before the sampling moment of any multidimensional data as the long-term preset time, forming a long-term subsequence of multidimensional data having the same spatial position as the any multidimensional data within the long-term preset time, and obtaining a temperature variance of the long-term subsequence based on the temperature of the multidimensional data in the long-term subsequence; A ratio of the temperature variance of the short-term subsequence to the temperature variance of the long-term subsequence is obtained as a third ratio, and a difference between a constant 1 and the third ratio is obtained as a local time continuity coefficient of any multidimensional data.

4. The method for analyzing in-mine monitoring data based on pattern recognition according to claim 1, characterized in that: The step of obtaining the adaptive spatiotemporal weight factor of any initial distance according to the local clustering coefficient and the local time continuity coefficient of the two multidimensional data corresponding to any initial distance includes: Obtaining a mean of the local clustering coefficients of the two multidimensional data corresponding to any of the initial distances as a first mean, and calculating an addition result of a constant 1 and the first mean to obtain a first addition result; Obtaining a mean of local time continuity coefficients of the two multidimensional data corresponding to any of the initial distances as a second mean, and calculating an addition result of a constant 1 and the second mean to obtain a second addition result; Obtaining a temperature difference between two multidimensional data corresponding to any initial distance, obtaining a ratio of a square value of the temperature difference to a square value of a preset temperature scale parameter as a third ratio, performing inverse proportional normalization processing on the third ratio, obtaining a result as a third inverse proportional normalized value, calculating a difference between a constant 1 and the third inverse proportional normalized value to obtain a first difference, and obtaining the first difference of a preset ratio as a nonlinear difference; Calculate the product between the first addition result, the second addition result and the nonlinear difference to obtain a second product, and obtain the product between the preset spatiotemporal weight factor parameter and the second product as the adaptive spatiotemporal weight factor of any initial distance.

5. The method for analyzing in-mine monitoring data based on pattern recognition according to claim 1, characterized in that: Obtaining a comprehensive anomaly score for each cluster based on the multidimensional data in each cluster includes: For any cluster, the temperature standard score of the cluster is obtained according to the temperature of the multidimensional data in the cluster, the size standard score of the cluster is obtained according to the amount of multidimensional data in the cluster, and the score difference between the temperature standard score of the cluster and the size standard score of the cluster is obtained as the comprehensive anomaly score of the cluster.

6. The method for analyzing in-mine monitoring data based on pattern recognition according to claim 1, characterized in that: The step of screening the multidimensional data for abnormalities within a preset time period based on the comprehensive abnormality score of each cluster includes: The comprehensive anomaly score of each cluster is sorted in descending order, and a preset proportion of clusters are selected from the sorted clusters as abnormal clusters, and the multidimensional data in the abnormal clusters are used as the multidimensional data with anomalies within a preset time period.

7. A mine monitoring data analysis system based on pattern recognition, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the mine monitoring data analysis method based on pattern recognition as described in any one of claims 1 to 6 are implemented.

Images