A coal vein trend reasoning system

Abstract

The application provides a coal vein trend reasoning system, which is based on coal mine literature and coal mining data to reason the coal vein trend, and comprises a mining data unit, a knowledge graph unit and a reasoning unit. The mining data unit classifies the mined raw coal of each mining coal seam and generates coal mining data, which comprises the coal ratio, quality information, mining location, mining time period and mining direction of each mining coal seam. The coal mine field knowledge graph is generated based on the coal mine literature and the coal mining data. The reasoning unit reasons the coal vein trend based on the coal mining data and the coal mine field knowledge graph. The coal vein trend reasoning system provided by the application comprehensively utilizes the existing coal mine literature to form a knowledge graph, and combines the actual coal mining data to reason the coal vein trend, so that the effective planning and guidance of coal mining can be realized, and the mining efficiency and mining quality are significantly improved.

Description

Technical Field

[0001] The present application relates to the field of mineral assessment and detection technology, and in particular, to a coal mine vein trend reasoning system. Background Art

[0002] Coal will remain one of the most important non-renewable resources in my country and globally for the foreseeable future. Because mined raw coal contains various impurities, it must be washed using methods such as manual washing, heavy media washing, and jigging before use. The separated waste, including gangue, must then be returned to the coal mining area through methods such as gangue backfilling. Obviously, the coal-rock ratio and coal enrichment level in the mined raw coal significantly impact the resources and costs required for sorting and backfilling the raw coal. Accurately predicting the trajectory of coal veins during the mining process and selecting coal-rich seams for mining based on these trends can effectively reduce mining costs and minimize environmental damage.

[0003] Existing technologies for detecting mineral resources, such as coal mines, typically require a series of phases, including surveys, detailed investigations, and exploration. These phases are long, labor-intensive, and expensive, leading to the common practice of not changing mining plans once exploration is complete and the mining plan is established. However, because actual coal vein trends often differ from pre-exploration results, failure to adjust these pre-exploration results based on actual mining conditions often leads to misjudgments in the pre-determined mining direction and unnecessary increases in mining costs. Summary of the Invention

[0004] The purpose of this application is to solve the problems existing in the above-mentioned prior art and to provide a coal mine vein trend inference system that can not only infer the vein trend based on past data, but also flexibly modify the inference results based on actual mining data.

[0005] The embodiments of the present application can be implemented through the following technical solutions:

[0006] A coal mine vein trend inference system is provided, which infers the coal mine vein trend based on coal mine literature and coal mine mining data, and includes a mining data unit, a knowledge graph unit and an inference unit. The mining data unit classifies the coal and stone of the mined raw coal of each mining coal seam and generates coal mine mining data. The coal mine mining data includes the coal and stone ratio, quality distribution information, mining location, mining time period and mining direction of each mining coal seam; the knowledge graph unit generates a coal mine field knowledge graph based on coal mine literature and coal mine mining data; and the inference unit infers the coal mine vein trend based on the coal mine mining data and the coal mine field knowledge graph.

[0007] Furthermore, the mining data unit includes: an identification module, which generates a transmission image of the mined raw coal in each mined coal seam based on the X-ray transmission data of the mined raw coal in the coal seam and identifies multiple coal stone targets therefrom; a classification module, which determines the coal stone category and quality coefficient of each coal stone target based on the X-ray transmission data; a statistical module, which counts the coal stone categories of each coal stone target, obtains the coal stone ratio and quality distribution information of each mined coal seam and generates the coal mine mining data; and a mining data management module, which stores, edits and deletes the coal mine mining data.

[0008] Preferably, the X-ray transmission data is generated based on high-energy transmission intensity and low-energy transmission intensity of X-rays.

[0009] Furthermore, the classification module determines the coal rock category of each coal rock target through the following steps:

[0010] A1. Determine the energy detection vector of the coal rock target based on the X-ray transmission data Where m is the number of sampling points in the coal rock target, is the low-energy transmission intensity of X-rays at the i-th sampling point, R i is the classification feature of the i-th sampling point determined by the following formula:

[0011]

[0012] in, are the initial values ​​of the high-energy transmission intensity and low-energy transmission intensity of X-rays, is the high-energy transmission intensity of X-rays at the i-th sampling point;

[0013] A2. Calculate the coal classification coefficient r of the coal target based on the energy detection vector c Classification coefficient r g ,

[0014]

[0015]

[0016] Among them, c is coal, g is gangue, for The mean value of I i,c To R i Substitute the coal fitting curve The estimated value obtained, For I i,c The mean value of I i,g To R i Substitute the gangue fitting curve The estimated value obtained, For Ii,g The mean of

[0017] A3. Based on the r c 、r g Determine the coal rock type of the coal rock target.

[0018] Furthermore, the statistical module determines the coal-rock ratio of each mined coal seam based on the following formula: c ,

[0019]

[0020] Where M is the number of coal rock targets classified as coal in the mined raw coal of the mining coal seam, L j 、W j are the length and width of the jth coal rock target classified as coal, S is the number of coal rock targets classified as gangue, L′ k 、W k ′ are the length and width of the kth coal rock target classified as gangue.

[0021] Furthermore, the quality distribution information of each mined coal seam includes the distribution of rich coal, normal coal and lean coal in the mined coal seam, and the quality distribution information of each mined coal seam is determined by clustering each coal stone target of the mined raw coal in the mined coal seam.

[0022] Preferably, the identification module further optimizes the identified coal rock targets through the following steps:

[0023] B1. Divide each coal target into multiple sampling rings with equal spacing from the center to the edge;

[0024] B2. Delete the outermost sampling ring of the coal target;

[0025] B3. Select the sampling ring to be retained based on the area of ​​the remaining portion of the coal rock target;

[0026] B4. Traverse each retained sampling ring;

[0027] B5. Generate optimized coal rock targets based on the traversal results.

[0028] Furthermore, the knowledge graph unit includes: a preprocessing module, which is used to preprocess the coal mine literature and the coal mine mining data to construct a corpus; a knowledge extraction module, which performs entity / attribute joint extraction and relationship extraction from the corpus to form coal mine knowledge triples; a knowledge fusion module, which performs knowledge fusion on the coal mine knowledge triples; a knowledge management module, which performs structured storage on the coal mine knowledge triples to form the coal mine field knowledge graph, and performs editing and deletion operations on the coal mine knowledge triples.

[0029] Preferably, the entity / attribute joint extraction is performed through a deep learning model based on BiLSTM-CRF; and the relationship extraction is performed through a deep learning model based on R-BERT.

[0030] Preferably, the knowledge extraction module further performs entity / attribute joint extraction and relationship extraction from the coal mining data to form coal mining knowledge triples.

[0031] Furthermore, the inference unit infers the trend of the coal vein through the following steps:

[0032] C1. Construct the initial coal seam node;

[0033] C2. Construct an inference node and associate it with the initial coal seam node, the inference node including the inference vein trend of the associated coal seam node and the inference coal-rock ratio and inference quality coefficient along the inference vein trend of the coal seam;

[0034] C3. Set the initial coal seam node to the current coal seam node, read the initial mining coal seam mining data and set it to the current mining data;

[0035] C4 based on the current mining data to update the current coal seam node;

[0036] C5. Execute preset inference rules and update inference nodes according to the current coal seam node and the coal mine field knowledge graph;

[0037] C5. Generate a new coal seam node and associate the inference node with it;

[0038] C6. Set the newly generated coal seam node as the current coal seam node, read the coal mining data of the next mining coal seam and set it as the current mining data;

[0039] C7. Return to step C4 until mining is completed.

[0040] Furthermore, the preset inference rules include:

[0041] Rule 1: If the coal-rock ratio of the current coal seam node is greater than the inferred coal-rock ratio in the inference node, then update the inferred vein trend in the inference node based on the current mining direction of the coal seam node, and update the inferred coal-rock ratio and inferred quality distribution information based on the coal mining data of the current coal seam node. Otherwise, execute Rule 2.

[0042] Rule 2: Based on the coal mining data of the current coal seam node and the inference quality distribution information, the most likely vein trend is inferred from the coal mine field knowledge graph through the path sorting algorithm, and the inference vein trend in the inference node is updated based on the inference result.

[0043] Preferably, the coal mine vein trend reasoning system also includes: a retrieval unit, used to query and retrieve the coal mine mining data, the coal mine field knowledge graph and the reasoning results of the coal mine vein trend; a display unit, used to display the coal mine mining data, the coal mine field knowledge graph and the reasoning results of the coal mine vein trend on a display device; a user management unit, used to set and manage user information; a database, used to store the coal mine mining data, the coal mine field knowledge graph and the reasoning results of the coal mine vein trend.

[0044] The coal mine vein trend inference system provided by the embodiments of the present application has at least the following beneficial effects:

[0045] (1) The coal mine vein trend inference system provided in the embodiment of the present application can comprehensively utilize actual mining data and coal mine knowledge graphs to perform vein trend inference. It can use the a priori coal seam location, geological structure, quality attribute information and their mutual relationships of each mining area that has completed exploration or mining to predict and infer the vein trend of the area to be mined, thereby effectively reducing the investment and cost of large-scale surveys of the mining area. It can also timely correct the inference situation based on the real data obtained as the actual mining progresses, effectively avoiding the impact of early survey errors on subsequent mining;

[0046] (2) The coal mine vein trend inference system provided in the embodiment of the present application can determine the coal stone ratio and quality coefficient of the mined raw coal in real time and accurately by performing X-ray transmission on the mined raw coal to identify the various coal stone targets contained therein, and constructing the energy detection vector of the coal stone target based on the transmission intensity data to perform coal stone classification. The above-mentioned coal stone ratio and quality coefficient are used to correct the inference results, and the inference results can be adjusted in a timely manner based on the measured data on the basis of the inferred vein trend, which greatly enhances the guiding significance for the actual mining plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the framework of the coal mine vein trend reasoning system according to an embodiment of the present application;

[0048] Figure 2 A diagram of the equipment layout for X-ray irradiation of mined raw coal to generate X-ray transmission data;

[0049] Figure 3A schematic diagram of a transmission image generated by the recognition module according to an embodiment of the present application and a coal and rock target determined therefrom;

[0050] Figure 4 A schematic diagram of dividing a sampling ring for a coal and rock target according to a specific embodiment of the present application;

[0051] Figure 5 A plurality of actually measured energy detection vector distribution results of coal and gangue;

[0052] Figure 6 A specific process of constructing a knowledge graph in the field of coal mines according to an embodiment of the present application;

[0053] Figure 7 A structure of a BiLSTM-CRF model used in an embodiment of the present application;

[0054] Figure 8 A case of labeling text according to an embodiment of the present application;

[0055] Figure 9 A structure of an R-BERT model used in an embodiment of the present application;

[0056] Figure 10 A specific process of importing coal mine knowledge triples and constructing a knowledge graph in the field of coal mines by a knowledge management module according to an embodiment of the present application;

[0057] Figure 11a 、 11b A partial schematic diagram of a knowledge graph in the field of coal mines displayed visually according to some embodiments of the present application;

[0058] Figure 12 A schematic diagram of reasoning about the trend of a coal mine vein according to an embodiment of the present application. DETAILED DESCRIPTION

[0059] Hereinafter, the present application will be further described based on preferred embodiments and with reference to the accompanying drawings.

[0060] In addition, various components on the drawings are enlarged or reduced for the convenience of understanding, but such a practice is not intended to limit the scope of protection of the present application.

[0061] The singular form of a word also includes a plural meaning, and vice versa.

[0062] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "inner", "outer" and the like indicate an orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or are the orientation or positional relationship in which the products of the embodiments of the present application are usually placed when in use, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, in order to distinguish different units, words such as first and second are used in this specification, but these are not limited by the order of manufacture, nor can they be understood as indicating or implying relative importance. Their names may be different in the detailed description and claims of the present application.

[0063] The vocabulary in this specification is used to illustrate the embodiments of the present application, but is not intended to limit the present application. It should also be noted that, unless otherwise clearly specified and limited, the terms "disposed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, an indirect connection through an intermediate medium, or a communication between the two components. For those skilled in the art, the specific meanings of the above terms in this application can be specifically understood.

[0064] An embodiment of the present application provides a coal mine vein trend inference system, which infers the coal mine vein trend based on coal mine literature and coal mine mining data. Figure 1 FIG. 4 shows a schematic diagram of a coal mine vein trend inference system according to an embodiment of the present application. Figure 1 As shown, the system provided by this application includes a data mining unit, a knowledge graph unit, and a reasoning unit.

[0065] The mining data unit classifies the mined raw coal of each mining coal seam into coal rock and generates coal mining data, wherein the coal mining data includes the coal rock ratio, quality distribution information, mining location, mining period and mining direction of each mining coal seam;

[0066] The knowledge graph unit generates a coal mine field knowledge graph based on coal mine literature and coal mine mining data; the reasoning unit infers the trend of coal mine veins based on the coal mine mining data and the coal mine field knowledge graph.

[0067] The coal mine vein trend inference system provided in the embodiments of the present application performs semantic analysis on various existing coal mine literature materials and extracts and integrates knowledge from actual coal mine mining data to form a coal mine field knowledge map. On this basis, it infers the trend of coal mine veins in mining in real time and continuously corrects the inference of the vein trend by comparing the inference results with the actual mining data. Using the system provided by this application, it is possible to synchronously predict the trend of coal mine veins according to the actual mining progress of the coal mine, greatly improving the prediction accuracy of coal mine vein trends, achieving precise guidance of coal mining, and effectively improving mining efficiency and mining quality.

[0068] like Figure 1 As shown, in a preferred embodiment of the present application, the system further includes a retrieval unit, a display unit, a user management unit, and a database. The retrieval unit is used to query and retrieve the coal mining data, the coal mining domain knowledge graph, and the inference results of the coal mine vein trend; the display unit is used to display the coal mining data, the coal mining domain knowledge graph, and the inference results of the coal mine vein trend on a display device; the user management unit is used to set and manage user information; and the database is used to store the coal mining data, the coal mining domain knowledge graph, and the inference results of the coal mine vein trend.

[0069] The following describes in detail the specific implementation methods of the mining data unit, the knowledge graph unit and the reasoning unit, as well as the process of reasoning about the trend of coal mine veins, in conjunction with the accompanying drawings and examples.

[0070] Specifically, in the embodiment of the present application, the mining data unit performs target recognition on the components of the mined raw coal of each mined coal seam (obviously, different mined coal seams correspond to different mining periods) to extract multiple coal stone targets, and then classifies each coal stone target, and determines the coal stone ratio and quality distribution information of the mined raw coal through statistical analysis. Among them, the coal stone ratio (in the embodiment of the present application, Ratio cThe ore fraction (representing the coal-rock ratio) represents the proportion of coal-rock targets identified as coal in the mined raw coal of each mined coal seam. A higher ore fraction means that the mined coal seam contains more coal components overall. The quality distribution information represents the distribution of raw coal with different enrichment levels in each mined coal seam. Due to the uneven distribution of coal resources, even within the same mined coal seam, the enrichment of coal components in raw coal mined during different mining cycles (such as a mining cycle of one day or one week) varies, and exhibits regional clustering characteristics. Based on the enrichment level of coal in each clustering area, it can be divided into rich coal, normal coal, and lean coal. The distribution of these rich coal, normal coal, and lean coal in the same mined coal seam constitutes the quality distribution information of the mined coal seam. The ore fraction and quality distribution information of each mined coal seam are statistically analyzed, and combined with its corresponding mining location, mining period (furthermore, a mining period can include multiple subdivided mining periods), and mining direction, the coal mining data of each mined coal seam is formed.

[0071] In an embodiment of the present application, the coal mining data of each of the above-mentioned mined coal seams can be collected regularly according to the actual mining progress. For example, a quarter can be used as a mining period to determine the corresponding mined coal seam and obtain the coal mining data corresponding to the mined coal seam (as mentioned above, a mining period can be further divided into multiple mining cycles with one month or one week as a mining cycle). For another example, a month can be used as a mining period to determine the corresponding mined coal seam and obtain the coal mining data corresponding to the mined coal seam. Obviously, setting different mining period lengths will result in different coal mining data.

[0072] Specifically, if Figure 1 As shown, the mining data unit includes: an identification module, a classification module, a statistics module, and a mining data management module. The identification module generates a transmission image of the mined raw coal in each mining coal seam based on the X-ray transmission data of the mined raw coal in the coal seam and identifies multiple coal rock targets from the image. The classification module determines the coal rock category of each coal rock target based on the X-ray transmission data. The statistics module collects statistics on the coal rock categories of each coal rock target, obtains coal rock ratio and quality distribution information for each mining coal seam, and generates the coal mining data. The mining data management module stores, edits, and deletes the coal mining data.

[0073] Figure 2 FIG. 1 shows a diagram of an apparatus layout for generating X-ray transmission data by performing X-ray irradiation on mined raw coal in some specific embodiments. Figure 2As shown, a belt conveyor, driven by a motor, transports mined raw coal. An X-ray source and an X-ray detection device (including an X-ray detector and a data acquisition card) are positioned opposite each other on the conveyor's conveying surface. The X-ray source emits X-rays toward the mined raw coal. After the X-rays penetrate the mined raw coal, the X-ray detection device collects their transmission intensity to generate X-ray transmission data. In some embodiments, this X-ray transmission data can be transmitted in real time to a recognition module to simultaneously generate a transmission image of the mined raw coal and identify coal targets. In other embodiments, the X-ray transmission data can be stored in a database and then read and processed by the recognition module at an appropriate time.

[0074] Table 1 below gives the parameters of the equipment used to collect X-ray transmission data in a specific embodiment:

[0075] Table 1 Equipment parameters for collecting X-ray transmission data

[0076]

[0077] The X-ray detector includes 20 detector cards, each capable of collecting transmission intensities at 64 sampling points, each with a width of 1.5 mm. Each of the 20 detector cards can collect transmission intensities at 1280 x 1 sampling point to generate Z-ray transmission data. In the above embodiment, the transmission intensities can be mapped from analog signals to 8-bit digital signals (0-255) according to their intensity range.

[0078] After the recognition module reads the X-ray transmission data corresponding to each sampling time, it performs data splicing based on the aforementioned equipment parameters to generate a transmission image of the mined raw coal. For example, in the above embodiment, X-ray transmission data for 320 sampling times can be extracted at a time to form a 1280×320 pixel transmission image, where the grayscale value of each pixel reflects the transmission intensity at the corresponding sampling point. The above-mentioned techniques for performing analog-to-digital conversion based on the signal intensity at each sampling point and generating a grayscale image are well known to those skilled in the art.

[0079] After obtaining the above-mentioned transmission image, the recognition module can use the target recognition algorithm to identify multiple coal stone targets from it. The method of identifying targets from grayscale images is already known to those skilled in the art. For example, the recognition module can first reduce the noise of the transmission image through spatial smoothing processing, and then generate a binary edge image based on the edge detection algorithm of multi-operator fusion, and finally determine the coal stone target from the binary edge image through the clustering algorithm. Figure 3 A schematic diagram showing a transmission image generated by an identification module and coal rock targets determined therefrom in some specific embodiments is shown.

[0080] In some embodiments of the present application, the above-mentioned X-ray transmission data is generated based on the high-energy transmission intensity and low-energy transmission intensity of the X-ray. In the actual process of X-ray transmission and coal rock target identification, due to the complex composition and the wide distribution range of its mass, density, and thickness, the transmission image constructed by simply using high-energy transmission intensity or low-energy transmission intensity may not accurately reflect the attenuation characteristics during the X-ray transmission process. For example, when the thickness of some raw coal materials is large, the high-energy part of the X-ray may not be able to penetrate both the coal and the gangue; when the thickness of some raw coal materials is small, after the low-energy part of the X-ray penetrates the coal and the gangue, there may be little difference in intensity attenuation, resulting in a problem of low discrimination of the intensity data.

[0081] Therefore, in some preferred embodiments of the present application, the above X-ray transmission data also includes the equivalent transmission intensity of the X-ray. Specifically, in some preferred embodiments, the X-ray detector can be a dual-energy detector that can simultaneously detect and obtain the high-energy transmission intensity I of the X-ray. H and low energy transmission intensity I L , for the high energy transmission intensity I at each sampling point H and low energy transmission intensity I L After processing, the "equivalent" transmission intensity I corresponding to the sampling point can be obtained. e The transmission intensity equivalent to the high-energy transmission intensity of the above-mentioned X-rays is used as the X-ray transmission data, which can effectively characterize the comprehensive attenuation characteristics of raw coal materials of different masses, densities, and thicknesses for the transmission of high-energy and low-energy parts of X-rays, thereby greatly increasing the discrimination of the generated transmission image for different materials, which is beneficial to edge detection and clustering processing in the subsequent target extraction process.

[0082] In the actual process of identifying coal stone targets, due to the large size distribution span of different coal stone targets in the raw coal material, especially when X-rays are irradiating larger materials, the problem of transmission intensity data distortion may occur. By statistically analyzing the shapes of coal stone materials of various sizes, it is found that the material is wedge-shaped from the center to the edge, that is, the center is thick and the edge is thin, and the center thickness increases with the increase of size. The above characteristics make the transmission intensity data of materials of different sizes and thicknesses present different usability. For example, for some large-sized materials with higher center thickness, the area with excessive thickness may not be able to penetrate, so that the transmission intensity data of this part of the area cannot truly reflect the attenuation characteristics of X-rays; the X-ray transmittance of its edge area is higher, but correspondingly, its transmission intensity is also more affected by noise; in addition, larger materials contain too many sampling points in the image, which makes the classification calculation time too long, seriously affecting the real-time performance of the classification. Therefore, in some preferred embodiments of the present application, the recognition module also optimizes the identified coal stone targets through the following steps:

[0083] B1. Divide each coal target into multiple sampling rings with equal spacing from the center to the edge;

[0084] B2. Delete the outermost sampling ring of the coal target;

[0085] B3. Select the sampling ring to be retained based on the area of ​​the remaining portion of the coal rock target;

[0086] B4. Traverse each retained sampling ring;

[0087] B5. Generate optimized coal rock targets based on the traversal results.

[0088] Figure 4 A schematic diagram of a specific sampling ring for coal rock target is shown, as shown in Figure 4 As shown, the number of sampling rings in the above steps can be set according to the size distribution range of the coal rock target and the resolution of the coal rock target image (for example, for the case where most material sizes are distributed in [10mm, 300mm], the number of divided sampling rings can be set to 10, and a sampling ring can be inserted between every two rings during upsampling). After the coal rock target is divided into multiple sampling rings, the edge part is first removed, and then the sampling ring for subsequent coal rock classification is selected according to the area of ​​the remaining part, and then the sampling points that are not penetrated by the X-ray (i.e., the sampling points with a transmission intensity of 0) are removed, and finally an optimized coal rock target for subsequent coal rock classification is obtained.

[0089] After the recognition module identifies each coal stone target through the above steps, the coal stone category of each coal stone target can be determined by the classification module. In the embodiment of the present application, the classification of coal stone targets is based on the different attenuation characteristics of coal and gangue for X-ray transmission. Statistics show that the true density of coal is about 1.3-1.5, and the true density of gangue is about 1.7-1.9. The density of coal and gangue is very different. The above difference makes coal and gangue show different characteristics for X-ray penetration attenuation, which determines that X-rays can be used to classify coal and gangue.

[0090] Specifically, in the embodiment of the present application, the classification module determines the coal rock category of each coal rock target through the following steps:

[0091] A1. Determine the energy detection vector of the coal rock target based on the X-ray transmission data Where m is the number of sampling points in the coal rock target, is the low-energy transmission intensity of X-rays at the i-th sampling point, R i is the classification feature of the i-th sampling point determined by the following formula:

[0092]

[0093] in, are the initial values ​​of the high-energy transmission intensity and low-energy transmission intensity of X-rays, is the high-energy transmission intensity of X-rays at the i-th sampling point;

[0094] A2. Calculate the coal classification coefficient r of the coal target based on the energy detection vector c Classification coefficient r g ,

[0095]

[0096]

[0097] Among them, c is coal, g is gangue, for The mean value of I i,c To R i Substitute the coal fitting curve The estimated value obtained, For I i,c The mean value of I i,g To R i Substitute the gangue fitting curve The estimated value obtained, For I i,g The mean of

[0098] A3. Based on the r c 、r g Determine the coal rock type of the coal rock target;

[0099] Specifically, first, in step A1, the energy detection vector of each sampling point of each coal rock target is constructed, where: The high-energy transmission intensity and low-energy transmission intensity of X-rays can be detected and obtained in the absence of coal stone materials.

[0100] For coal and gangue with different transmission attenuation characteristics, their energy detection vectors show different R~I L Corresponding relationship, Figure 5 The energy detection vector distribution results of coal and gangue obtained from multiple field measurements are shown. Figure 5 It can be seen that the energy detection vector of coal and gangue is in its R~I L The feature space shows obvious clustering characteristics, and the separation degree of the clustered areas is good. Therefore, the energy detection vector constructed in this application can be effectively applied to the classification of coal and gangue.

[0101] In order to facilitate the classification of coal stone, y = a·e bxThe function relationship is used to perform curve fitting on the multiple energy detection vectors of the measured coal and gangue, to obtain a fitting curve of the coal and the fitting curve of the gangue Then, in step A2, the classification correlation coefficient r c of the coal classification coefficient r g and the gangue classification coefficient of each coal and gangue target is obtained respectively c .

[0102] After obtaining the r g and r c of each coal and gangue target, the classification of each coal and gangue target can be determined in step A3, and in some specific embodiments, the classification of the coal and gangue target with the obtained r g and r c values can be performed by using a preset classification strategy, for example, one of the following various classification strategies can be used to classify the coal and gangue target:

[0103] Simple classification strategy, if r g , the coal and gangue target is coal, otherwise, the coal and gangue target is gangue;

[0104] Reducing-gangue strategy, if r c > a first threshold value, the coal and gangue target is coal, otherwise, the coal and gangue target is gangue;

[0105] Reducing-coal strategy, if r g > a second threshold value, the coal and gangue target is gangue, otherwise, the coal and gangue target is coal.

[0106] After the classification of each coal and gangue target is completed by the classification module, the coal and gangue classification of each coal and gangue target can be counted by the statistical module, and through the counting, the coal and gangue proportion and quality distribution information of each mining coal seam can be obtained, and finally, the coal mining data of the mining coal seam can be generated.

[0107] Specifically, in the embodiments of the present application, the statistical module determines the coal and gangue proportion Ratio c based on the following formula

[0108]

[0109] wherein, M is the number of the coal and gangue targets classified as coal, L j and W j are the length and width of the jth coal and gangue target classified as coal respectively, S is the number of the coal and gangue targets classified as gangue, L' k and W' k are the length and width of the kth coal and gangue target classified as gangue respectively.

[0110] As described above, the coal-rock ratio generally represents the proportion of coal components in a mined coal seam to the total mined raw coal. However, even within the same coal seam, the enrichment of coal components in raw coal mined during different mining cycles can vary significantly. These different enrichment levels correspond to different mineral qualities at each mining location. Combined with the coal-rock ratio, it can comprehensively reflect the overall trend and local detailed characteristics of the mineral vein. Therefore, in the embodiment of the present application, the statistical module also collects statistical information on the quality distribution of each mined coal seam.

[0111] Specifically, the quality distribution information includes the distribution of rich coal, normal coal, and lean coal in the mined coal seam. An optional implementation method for determining the quality distribution information is to use a clustering algorithm to cluster the various coal rock targets of the mined raw coal in the mined coal seam. For example, as the various coal rock targets in the mined raw coal are continuously identified by the above-mentioned recognition module and classified by the classification module, their transmission intensity and r c 、r g The statistical module continuously acquires information such as these. Using a clustering algorithm, the statistical module continuously clusters coal targets with the same or similar properties, ultimately obtaining multiple coal target clusters. Each coal target within each coal target cluster has similar properties. The module further calculates the coal target ratio within each coal target cluster, and based on this ratio, the cluster can be classified as rich coal, normal coal, or lean coal. The above-mentioned method of clustering coal target components in mined raw coal to determine raw coal quality is well known to those skilled in the art and will not be elaborated on here.

[0112] After the statistical module has compiled statistics on the coal stone ratio and quality distribution information of each mined coal seam, it can combine its corresponding mining location, mining period (further, a mining period can include multiple subdivided mining cycles) and mining direction to finally form the coal mining data of each mined coal seam.

[0113] The above is a detailed introduction to the specific implementation of the mining data unit. The following is an introduction to the specific implementation of the knowledge graph unit with reference to the accompanying drawings.

[0114] Specifically, in the embodiments of the present application, Figure 1 As shown in the figure, the knowledge graph unit includes a preprocessing module, a knowledge extraction module, a knowledge fusion module, and a knowledge management module. The preprocessing module is used to preprocess coal mine literature and coal mining data to build a corpus; the knowledge extraction module performs entity / attribute joint extraction and relationship extraction on the corpus to form coal mine knowledge triples; the knowledge fusion module performs knowledge fusion on coal mine knowledge triples and forms a coal mine domain knowledge graph; and the knowledge management module performs storage, editing, and deletion operations on the coal mine domain knowledge graph.

[0115] With the rapid development of the internet, a vast amount of data is generated online. To quickly extract the required, valid data from this vast amount of data, it is necessary to clean and denoise the data, and integrate the data based on the existing relationships between them. The concept of knowledge graph emerged in this context. Its essence is a data structure in the form of a directed graph. Its basic components are a data structure composed of entities, relationships, and attributes. Entities with attributes are connected through relationships, forming the basic structure of the knowledge graph.

[0116] By extracting relevant coal mining literature related to coal seam characteristics and coal seam occurrence characteristics from data obtained through the internet and other channels, and combining this with actual mining data, a knowledge graph construction process can be used to generate a coal mining domain knowledge graph specifically tailored to coal mines. This coal mining domain knowledge graph encompasses the attributes of different coal seams at different mines and the relationships between them. This priori attribute and relationship information provides a basis for inferring coal vein trends. Figure 6 The specific process of constructing a knowledge graph in the coal mining field in some embodiments is shown.

[0117] In some specific embodiments of the present application, the preprocessing module analyzes the coal mine-related data retrieved and downloaded from various channels, extracts coal mine literature related to coal seam characteristics, coal seam occurrence characteristics, etc., and simultaneously obtains the coal mine mining data obtained by the mining data unit, and preprocesses the above-mentioned coal mine literature and coal mine mining data, such as using regular expressions and other related syntax to construct special symbols that need to be filtered, and finally forms a self-built corpus.

[0118] In some specific embodiments of the present application, the knowledge extraction module performs entity / attribute joint extraction and relationship extraction on the corpus through a deep learning model based on BiLSTM-CRF to form coal mine knowledge triples. The BiLSTM-CRF model combines a BiLSTM model with a CRF model. BiLSTM (i.e., bidirectional LSTM) solves the defect that the LSTM model can only capture information transmitted from the front to the back, and can capture both forward and reverse information at the same time, making the utilization of text information more comprehensive and effective. CRF (Conditional Random Fields) is a conditional probability distribution model that solves the output sequence under the condition of a given input sequence. The combination of the two models can better complete the task of entity and attribute recognition.

[0119] Figure 7 The structure of the BiLSTM-CRF model used in some specific embodiments is shown as follows: Figure 7As shown in the figure, the input layer of the BiLSTM-CRF model is the Embedding layer, which primarily converts the input text sequence into the feature vectors required for model training. The BiLSTM layer of the BiLSTM-CRF model consists of two LSTMs. The BiLSTM uses the forward LSTM layer to learn information about the previous context and the backward LSTM layer to learn information about the following context, integrating this contextual information for training. The vectors obtained by the forward and backward LSTM layers are then concatenated in the hidden layer according to their position and then fed into the next layer for processing. The CRF layer in the BiLSTM-CRF model primarily adds constraints to the prediction results input by the hidden layer to reduce the probability of model prediction errors.

[0120] In some specific embodiments of the present application, the BIO annotation tool Daccano can be used to annotate the texts and coal mining data composed of multiple documents contained in the corpus. The annotated information is the entities and related attribute information contained in the text and data. Tables 3 and 4 respectively show the coal mining entity types and attribute definitions in some specific embodiments. Figure 8 The following illustrates how text is annotated in some specific embodiments.

[0121] Table 3 Coal mine entity types

[0122] Label Entity Type KQ mining area MX coal measures MC coal seam

[0123] Table 4 Coal mine attribute definitions

[0124] Label Attribute Type Label Attribute Type DLGM Quantitative scale MXLX Coal rock type DC Strata JG Gangue SJ time JGYX Gangue rock properties YS color MCJG Coal seam structure KJ space MCCG Coal seam occurrence CJX Sedimentary facies MYGZ Coal rock structure MCXT Coal seam morphology MYZF Coal rock components DBYX Roof lithology MCWDX Coal seam stability DBYX1 Roof lithology 1 MZ Coal Type DK fracture MCKXC Coal seam mineability LX rift YX Lithology

[0125] After data labeling is completed, the training set, validation set, and test set are divided into training set, validation set, and test set in a ratio of 6:2:2, and the above BiLSTM-CRF model is trained. Finally, the tested BiLSTM-CRF model can be used for entity / attribute joint extraction.

[0126] In some preferred embodiments, relation extraction is performed using a deep learning model based on R-BERT. The R-BERT model is a derivative of the BERT model for relation extraction and can be applied to a variety of relation extraction tasks. Figure 9 The structure of the R-BERT model used in some specific embodiments is shown.

[0127] like Figure 8As shown in the figure, the R-BERT model consists of three parts. The first part is the BERT model, which is mainly used to generate text vectors. The second part is the FCLayer layer, which processes the sentence vectors obtained from the BERT model in order to obtain entity and attribute vector representations (it should be noted that since entity and attribute vectors may contain more than one word, the R-BERT model sums and averages the word vectors contained in the entity to represent the entity and attribute vector representation). The third part is the label_classifier, which concatenates the vector representations of entities and attributes to obtain a matrix, passes the matrix through a fully connected layer followed by a softmax, and obtains its loss function through cross entropy.

[0128] In some specific embodiments of the present application, multiple sentences containing at least two entities and attributes are selected from the data of completed entity\attribute joint extraction for data annotation. The format of data annotation is (entity 1, entity 2, relationship type, original sentence), where the relationship type reflects the relationship between different entities. Table 5 shows the coal mine relationship types summarized based on the entity and attribute data characteristics of coal mine literature.

[0129] Table 5 Coal mine relationship types

[0130] Relationship Type Interpretation of relationship Containment relationship The relationship between the three coal mine entities Spatial attribute association Three types of coal mine entities and spaces Time attribute association Three types of coal mine entities and time Basic feature association Three types of coal mine entities and quantitative scale, structure, morphology, occurrence, etc. Correlation between coal quality characteristics The color, fracture, crack, structure, and texture of three types of coal mine entities and coal rocks Sedimentary environment Three coal deposit entities and sedimentary facies Structural Control Three types of coal mine entities and geological structures

[0131] According to the above relationship type, the sentence can be annotated to obtain annotated data in the form of: "(Qiquanhu mining area, Taipei sag in Turpan Depression, spatial attribute association relationship, Qiquanhu mining area is located in the northwest corner of the Tuha Basin and belongs to the Taipei sag in Turpan Depression)".

[0132] After obtaining the above-mentioned multiple annotated data, the training set, validation set, and test set are constructed using the above-mentioned annotated data and the above-mentioned R-BERT model is trained. Finally, the tested R-BERT model can be used to extract relations from sentences containing at least two entities and attributes and generate corresponding coal mine knowledge triples.

[0133] In the embodiments of the present application, the corpus used by the knowledge graph unit to construct the knowledge graph comes not only from various coal mining literature, but also from coal mining data obtained from actual mining. By constructing coal mining knowledge triples using actual coal mining data and incorporating them into the coal mining resource knowledge graph, the coal mining resource knowledge graph can further include information such as the geological structure and quality attributes of the coal seams in the mining area being mined, as well as their associated relationships, further improving the accuracy of reasoning about the vein trends in the mining area.

[0134] In the embodiments of the present application, the knowledge fusion module is used to perform a knowledge fusion operation on the generated coal mine knowledge triples. The reason for performing the knowledge fusion operation is that the coal mine literature resources are from various sources, and the same entity or attribute has different aliases, abbreviations, etc. in different literature, which needs to be processed to make the description information about the same entity or attribute from multiple sources fused together.

[0135] Specifically, in some embodiments of the present application, a similarity algorithm can be used to calculate the similarity of two entities or attributes to perform entity and attribute alignment, and ensure the uniqueness of a coal mine entity or attribute in the knowledge base. For example, a cosine similarity algorithm can be used to perform attribute alignment operation, an entity name similarity and attribute similarity weighting algorithm can be used to perform entity alignment operation, and an artificial auxiliary means can be added in the fusion process to control the fusion accuracy. The above knowledge fusion method based on similarity algorithm is known to those skilled in the art, and will not be repeated here.

[0136] In the embodiments of the present application, the knowledge management module is used to import and store the coal mine knowledge triples obtained through knowledge extraction and knowledge fusion, to form a structured coal mine domain knowledge graph, and to perform editing, deleting, etc. on the above coal mine knowledge triples. Figure 10 The specific process of importing the coal mine knowledge triples and constructing the coal mine domain knowledge graph by the knowledge management module in some specific embodiments is shown, Figure 11a 、 11b The local schematic diagrams of the coal mine domain knowledge graph displayed visually in some specific embodiments are shown respectively.

[0137] As explained in detail above, in the embodiments of the present application, the mining data unit and the knowledge graph unit respectively obtain actual coal mine mining data of different coal seams according to actual mining progress, and obtain the relationship between entities and attributes of different mining areas and different coal seams according to coal mine literature data and form a structured coal mine domain knowledge graph. Further, in the embodiments of the present application, the reasoning unit comprehensively utilizes the above coal mine mining data and the coal mine domain knowledge graph, and performs deductive reasoning based on a preset reasoning rule to complete the reasoning of the coal mine vein trend.

[0138] Specifically, in some embodiments, the reasoning unit performs reasoning of the coal mine vein trend by the following steps:

[0139] C1. Construct an initial coal seam node;

[0140] C2. Construct a reasoning node and associate it with the initial coal seam node, the reasoning node including a reasoning vein trend of the associated coal seam node and a reasoning coal proportion and a reasoning quality coefficient of the coal seam along the reasoning vein trend;

[0141] C3. Set the initial coal seam node to the current coal seam node, read the initial mining coal seam mining data and set it to the current mining data;

[0142] C4 based on the current mining data to update the current coal seam node;

[0143] C5. Execute preset inference rules and update inference nodes according to the current coal seam node and the coal mine field knowledge graph;

[0144] C5. Generate a new coal seam node and associate the inference node with it;

[0145] C6. Set the newly generated coal seam node as the current coal seam node, read the coal mining data of the next mining coal seam and set it as the current mining data;

[0146] C7. Return to step C4 until mining is completed.

[0147] Furthermore, the preset inference rules include:

[0148] Rule 1: If the coal-rock ratio of the current coal seam node is greater than the inferred coal-rock ratio in the inference node, then update the inferred vein trend in the inference node based on the current mining direction of the coal seam node, and update the inferred coal-rock ratio and inferred quality distribution information based on the coal mining data of the current coal seam node. Otherwise, execute Rule 2.

[0149] Rule 2: Based on the coal mining data of the current coal seam node and the inference quality distribution information, the most likely vein trend is inferred from the coal mine field knowledge graph through the path sorting algorithm, and the inference vein trend in the inference node is updated based on the inference result.

[0150] Specifically, when the coal-rock ratio obtained by statistically analyzing the coal mining data of a certain mined coal seam node is greater than the inferred coal-rock ratio in the inference node, it is considered that the current mining direction is closer to the correct vein direction than the inferred vein trend. At this time, the current mining direction should be kept unchanged, and the inferred vein direction in the inference node should be corrected using the current mining direction. At the same time, the inferred coal-rock ratio and the inference quality distribution information in the inference node should be updated to the actual mining data of the current coal seam node; when the coal-rock ratio obtained by statistically analyzing the coal mining data of a certain mined coal seam node is not greater than the inference coal-rock ratio in the inference node, it can be considered that the current mining direction deviates from the vein direction, so that the overall coal content of the mined raw coal fails to reach the optimal value. At this time, reasoning should be performed based on the coal mine field knowledge graph to obtain the vein trend that is most likely to mine raw coal with inference quality distribution information.

[0151] In an embodiment of the present application, the above-mentioned reasoning of the vein trend based on the coal mine knowledge graph is implemented by a path ranking algorithm. The path ranking algorithm (PRA) is a data-driven reasoning algorithm that establishes and trains a relational learning model based on the knowledge graph. It then uses the trained relational learning model to infer possible valid association paths between entities (i.e., possible relationships between entities) and determines the most likely valid association path (i.e., the most likely relationship between entities) through ranking.

[0152] Specifically, in some embodiments of the present application, the vein trend can be inferred by the following steps:

[0153] First, based on multiple coal mine knowledge triples (h, r, t) in the knowledge graph, we construct the first entity set E, the relationship set R, and the tail entity set T, where r is the relationship connecting the entity pair (h, t), and the local subgraph of the knowledge graph connected by r is G. i , the entity set associated with r is E i , by h k ∈E i The set of tail entities associated with r is T k , subgraph G i The associated valid path of the subgraph G is defined as i The feature set ψ i .

[0154] Secondly, define the knowledge graph entity relationship cluster {Re}, calculate the similarity between clusters, merge clusters with high similarity, and update the shared feature set of the merged clusters. The similarity between clusters is measured as follows:

[0155]

[0156] Among them, Re n and Re m The coal-rock ratio of clusters n and m based on a single coal-rock target cluster c Description, b n is the cluster deviation, w c is the individual bias weight, which can be calculated by the shared weight matrix of the relational learning model. n and Re m When the similarity is high, the clusters will be merged and the feature set of the merged clusters will be updated, that is, ΠRe=ΠRe n ∪ΠRe m .

[0157] On the basis of completing the above-mentioned cluster fusion, the most likely effective association path between the coal mining data of the current coal seam mining node (coal stone ratio, quality distribution information, mining location, etc. abstracted in the form of triples) and the inference data in the inference node (inferenced coal stone ratio, inferenced quality distribution information also abstracted in the form of triples) is searched based on the path sorting algorithm. The effective association path is the inference result of the vein trend. Specifically, the entity and attribute information of the knowledge graph is represented as an embedding vector, and the possibility of the existence of association between entity pairs is judged by calculating the entity similarity, and some invalid path searches are filtered out. The entity similarity information is used as a constraint condition for random walks, and a target-driven heuristic random walk is performed on the global graph corresponding to the knowledge graph to obtain the path feature set of the relationship. The modeling of the relationship learning model is completed by calculating and training the path features. In some specific embodiments, the relationship learning model can be defined as:

[0158]

[0159] Among them, x i,c ,y i,c are the embedding vectors of the two entities / attribute information whose relationship is to be inferred, L(x i,c ,y i,c ) is the logistic regression form of the training loss function, and its specific expression is L(x ic ,y ic )=log(1+exp(-y ic f c (x ic ))), where f c (x ic )=(w0+w c )x ic +b0, μ1 and μ2 are regularization parameters used to control the degree of parameter sharing between different relations.

[0160] Finally, the above-mentioned coal mining data and reasoning data, which are abstracted into entity and attribute information, are respectively input into the trained relational learning model as entity / attribute information embedding vectors to output the optimal vein trend inference results. Figure 12 A schematic diagram of inferring the trend of coal mine veins according to a specific embodiment is shown.

[0161] From the above detailed description of the reasoning method of the reasoning unit, it can be seen that the comprehensive use of actual mining data and coal mining knowledge graphs to reason about vein trends can not only use the a priori coal seam location, geological structure, quality attribute information and their mutual relationships of each mining area that has completed exploration or mining to predict and infer the vein trends of the area to be mined, thereby effectively reducing the investment and cost of large-scale surveys in the mining area, but also can timely correct the reasoning situation based on the real data obtained with the actual mining progress, effectively avoiding the impact of early survey errors on subsequent mining.

[0162] The above is a detailed introduction to the specific implementation methods of the present application. For those skilled in the art, several improvements and modifications can be made to the present application without departing from the principles of the present application. These improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A coal vein trend inference system that infers coal vein trends based on coal mine literature and coal mine mining data, including a knowledge graph unit, a mining data unit, and an inference unit, characterized by: The mining data unit classifies the mined raw coal of each mining coal seam into coal rock and generates coal mining data, wherein the coal mining data includes the coal rock ratio, quality distribution information, mining location, mining period and mining direction of each mining coal seam; The knowledge graph unit generates a coal mine field knowledge graph based on coal mine literature and coal mine mining data; The reasoning unit infers the trend of coal veins based on the coal mining data and the coal mining domain knowledge graph; The inference unit infers the trend of the coal vein by the following steps: C1. Construct the initial coal seam node; C2 constructs an inference node and associates it with the initial coal seam node, the inference node includes the inference vein trend of the associated coal seam node and the inference coal stone ratio and inference quality distribution information along the inference vein trend of the coal seam; C3. Set the initial coal seam node to the current coal seam node, read the initial mining coal seam mining data and set it to the current mining data; C4 based on the current mining data to update the current coal seam node; C5. Execute preset inference rules and update inference nodes according to the current coal seam node and the coal mine field knowledge graph; C5. Generate a new coal seam node and associate the inference node with it; C6. Set the newly generated coal seam node as the current coal seam node, read the coal mining data of the next mining coal seam and set it as the current mining data; C7. Return to step C4 until mining is completed; The preset inference rules include: Rule 1: If the coal-rock ratio of the current coal seam node is greater than the inferred coal-rock ratio in the inference node, then update the inferred vein trend in the inference node based on the current mining direction of the coal seam node, and update the inferred coal-rock ratio and inferred quality distribution information based on the coal mining data of the current coal seam node. Otherwise, execute Rule 2. Rule 2: Based on the coal mining data of the current coal seam node and the inference quality distribution information, the most likely vein trend is inferred from the coal mine field knowledge graph through the path sorting algorithm, and the inference vein trend in the inference node is updated based on the inference result.

2. The coal mine vein trend inference system according to claim 1, characterized in that: The mining data unit includes: an identification module, which generates a transmission image of the mined raw coal in each mined coal seam based on the X-ray transmission data of the mined raw coal in the coal seam and identifies multiple coal rock targets therefrom; a classification module, for determining a coal rock category of each coal rock target based on the X-ray transmission data; A statistical module collects statistics on the coal stone categories of each coal stone target, obtains the coal stone ratio and quality distribution information of each mined coal seam and generates the coal mining data; The mining data management module stores, edits and deletes the coal mining data.

3. The coal mine vein trend reasoning system according to claim 2, characterized in that: The X-ray transmission data is generated based on the high-energy transmission intensity and the low-energy transmission intensity of the X-ray.

4. The coal mine vein trend inference system according to claim 2, characterized in that: The classification module determines the coal rock category of each coal rock target by the following steps: A1. Determine the energy detection vector of the coal rock target based on the X-ray transmission data Where m is the number of sampling points in the coal rock target, is the low-energy transmission intensity of X-rays at the i-th sampling point, R i is the classification feature of the i-th sampling point determined by the following formula: in, are the initial values ​​of the high-energy transmission intensity and low-energy transmission intensity of X-rays, is the high-energy transmission intensity of X-rays at the i-th sampling point; A2. Calculate the coal classification coefficient r of the coal target based on the energy detection vector c Classification coefficient r g , Among them, c is coal, g is gangue, for The mean value of I i,c To R i Substitute the coal fitting curve The estimated value obtained, For I i,c The mean value of I i,g To R i Substitute the gangue fitting curve The estimated value obtained, For I i,g The mean of A3. Based on the r c 、r g Determine the coal rock type of the coal rock target.

5. The coal mine vein trend reasoning system according to claim 4, characterized in that: The statistical module determines the coal-rock ratio of each mined coal seam based on the following formula: c , Where M is the number of coal rock targets classified as coal in the mined raw coal of the mining coal seam, L j 、W j are the length and width of the jth coal rock target classified as coal, S is the number of coal rock targets classified as gangue, L′ k 、W k ′ are the length and width of the kth coal rock target classified as gangue.

6. The coal mine vein trend reasoning system according to claim 4, characterized in that: The quality distribution information of each mined coal seam includes the distribution of rich coal, normal coal and lean coal in the mined coal seam, and the quality distribution information of each mined coal seam is determined by clustering each coal rock target of the mined raw coal in the mined coal seam.

7. The coal mine vein trend inference system according to claim 2, characterized in that: The identification module further optimizes the identified coal rock targets through the following steps: B1. Divide each coal target into multiple sampling rings with equal spacing from the center to the edge; B2. Delete the outermost sampling ring of the coal target; B3. Select the sampling ring to be retained based on the area of ​​the remaining portion of the coal rock target; B4. Traverse each retained sampling ring; B5. Generate optimized coal rock targets based on the traversal results.

8. The coal mine vein trend inference system according to claim 1, characterized in that: The knowledge graph unit includes: A preprocessing module, configured to preprocess the coal mine literature and the coal mine mining data to construct a corpus; a knowledge extraction module, performing entity / attribute joint extraction and relationship extraction on the corpus to form coal mine knowledge triples; A knowledge fusion module, performing knowledge fusion on the coal mine knowledge triples; The knowledge management module performs structured storage on the coal mine knowledge triples to form the coal mine field knowledge graph, and performs editing and deletion operations on the coal mine knowledge triples.

9. The coal mine vein trend reasoning system according to claim 8, characterized in that: The entity / attribute joint extraction is performed through a deep learning model based on BiLSTM-CRF; The relationship extraction is performed through a deep learning model based on R-BERT.

10. The coal mine vein trend inference system according to claim 1, characterized in that: The coal mine vein trend reasoning system also includes: A retrieval unit, configured to query and retrieve the coal mining data, the coal mining domain knowledge graph, and the inference results of the coal mine vein trend; A display unit, configured to display the coal mining data, the coal mining domain knowledge graph, and the inference results of the coal mine vein trend on a display device; User management unit, used to set and manage user information; A database is used to store the coal mining data, coal mining knowledge graph and the reasoning results of coal vein trends.

Images