Coal mine data acquisition method and system based on artificial intelligence

Abstract

The present application relates to the technical field of coal mine data acquisition, in particular to a coal mine data acquisition method and system based on artificial intelligence, comprising the following steps: S1, a collection node and a relay node generate communication reachability and data missing records in reporting, merge to form blind area representation data according to roadway space units, input a blind area learning model to generate a blind area fingerprint and a heat map; S2, based on the heat map, identify effective communication space-time combinations, generate a blind area opportunity window table, input a collection strategy generation model, output an opportunity collection strategy, and instruct the node to execute sampling and reporting within the opportunity window; S3, the node collects and reports according to the strategy, the gateway completes consistency verification and forms an effective data set, and the data and hit records are backfilled to the model to realize self-calibration and update of the blind area fingerprint and the strategy. The present application realizes the improvement of data acquisition accuracy, the optimization of energy consumption and the enhancement of system self-adaptation of the underground sensing network in the dynamic blind area environment.

Description

Technical Field

[0001] This invention relates to the field of coal mine data acquisition technology, and in particular to a coal mine data acquisition method and system based on artificial intelligence. Background Technology

[0002] With the advancement of intelligent transformation in coal mines, underground data acquisition systems have become a crucial foundation for safe production and equipment status monitoring. Low-power wireless sensing networks, due to their flexible deployment and low energy consumption, are widely used in underground coal mine environments to collect various monitoring data, including temperature, gas levels, equipment operation, and personnel location. This collected data not only supports real-time monitoring and remote decision-making but also provides vital data support for subsequent intelligent analysis and risk warning.

[0003] However, the complex and variable spatial structure of underground coal mines (such as tunnel turns, junctions, and equipment assembly areas) and the interfering environment (such as metal structure shielding, high humidity, and high dust) result in numerous communication blind spots in the sensing network. These blind spots cause unstable data reporting, frequent packet loss, or complete data loss, making the acquisition strategy lack specificity. Furthermore, existing sensing network acquisition scheduling mostly adopts a fixed periodic acquisition method, failing to intelligently adjust based on fluctuations in communication reachability and acquisition task priorities, leading to energy waste and the loss of critical data. In addition, blind spot models and acquisition strategies have long relied on static configurations, making it difficult to adapt to the dynamic evolution of the mine operating environment, resulting in complex system maintenance and low data reliability. Summary of the Invention

[0004] This invention provides a coal mine data acquisition method and system based on artificial intelligence. By collecting underground communication accessibility and data missing records, it constructs spatially merged blind zone characterization data, generates self-calibrating blind zone fingerprints and heat maps using a blind zone learning model, and dynamically generates opportunity acquisition strategies by combining task priorities and blind zone opportunity windows. This achieves intelligent control and adaptive optimization of data acquisition behavior, significantly improving acquisition coverage, communication success rate and data quality, reducing energy consumption and enhancing system robustness, and adapting to the actual needs of dynamic changes in the underground environment.

[0005] The artificial intelligence-based coal mine data acquisition method includes the following steps: S1. In the low-power wireless sensing network in the coal mine, communication reachability records and data loss records generated by each acquisition node and relay node during daily reporting are collected. The communication reachability records and data loss records are spatially merged according to roadway segments, turning points, roadway entrances and equipment operation positions to obtain blind zone characterization data corresponding to each spatial unit. The blind zone characterization data is input into the blind zone learning model to output the blind zone fingerprint of each spatial unit. Based on the blind zone fingerprint, a blind zone heat map is generated to indicate the strength and drift trend of the blind zone. S2, based on the blind zone heat map, identify the time period and location combination for effective communication, generate a blind zone opportunity window table corresponding to each spatial unit, and input the blind zone opportunity window table into the acquisition strategy generation model to generate an opportunity acquisition strategy that matches the priority of different acquisition tasks. The opportunity acquisition strategy instructs the acquisition node to perform a sequence of actions such as wake-up, sampling, buffering and short burst reporting within the blind zone opportunity window. S3, the acquisition node performs data acquisition and reporting according to the opportunity acquisition strategy. The gateway performs consistency verification and anomaly removal on the returned data to form a valid dataset. The valid dataset and its corresponding blind spot opportunity window hit status are fed back into the blind spot learning model and the acquisition strategy generation model to complete the self-calibration update of the blind spot fingerprint and the blind spot opportunity window table.

[0006] Optionally, S1 includes: S11, in the low-power wireless sensing network environment in coal mines, the acquisition nodes and relay nodes generate communication reachability records and data missing records during daily communication reporting, and divide the dimensions according to the preset spatial units, including roadway sections, turning points, roadway entrances and equipment operation locations, and form a blind area representation dataset that reflects spatial distribution characteristics through geospatial merging. S12, input the blind zone representation dataset into the blind zone learning model, identify the blind zone hidden state vector of each spatial unit, and output the blind zone fingerprint, including indicators describing the strength of communication capability of the region, stability indicators of communication success time window, and indicators of communication state time change trend. S13, visualize and map the blind zone fingerprint of each spatial unit to generate a blind zone heat map that reflects the communication coverage of the downhole sensing network. The blind zone heat map uses spatial units as grids and uses color depth to represent the severity of the blind zone.

[0007] Optionally, S11 includes: S111, In the coal mine, between the data acquisition nodes and relay nodes deployed underground, communication reachability records and data loss records are continuously recorded during the communication process. The data acquisition nodes include gas sensors, temperature and humidity meters, and equipment status acquisition devices. The relay nodes are responsible for data relay and calculate reachability scores based on the collected communication reachability records and data loss records. Data missing rate ; S112 matches the node coordinates with spatial units of different dimensions in the coal mine, assigning each node to a spatial unit. Establish spatial distribution mapping Furthermore, within each spatial unit, the reachability scores and data missing rates of its constituent nodes are weighted and aggregated to form the spatial unit. Blind zone characterization index ; S113, each spatial unit Corresponding blind zone characterization index Number of nodes, communication drift trend and the time trend of missing ratio Packed into blind zone representation vectors Ultimately, the blind zone representation vectors of all spatial units constitute the blind zone representation dataset for the entire mining area.

[0008] Optionally, S12 includes: S121, for each spatial unit The accessibility score and data missing ratio of the blind zone representation dataset within a continuous time window are extracted to construct the blind zone temporal input feature sequence. ; S122, input the blind zone time sequence into the feature sequence. The input is fed into the blind zone learning model to extract the hidden state vector of the spatial unit. ; The blind spot learning model employs a graph neural network (GCN) model, specifically including: S1221, for each spatial unit Obtain the corresponding blind zone temporal input feature sequence. Furthermore, these features are aggregated along the time dimension to construct the node input feature vector of the spatial unit. Input all spatial unit nodes into feature vectors Combined into a node feature matrix ; S1222, based on the spatial topological relationships in underground coal mines, constructs an undirected graph between spatial units. ,in, It is a set of nodes, that is, a set of spatial units. For edge sets, if spatial units and If they are physically adjacent or have communication links, then in and Establish edges between them, and the adjacency matrix of each edge is: By introducing self-loops and performing symmetric normalization, a normalized adjacency matrix is ​​obtained. ; S1223, The node feature matrix With normalized adjacency matrix Input a graph neural network, perform graph convolution operations, and output spatial units. Blind zone hidden state vector ; S123, based on the hidden state vector of the blind zone Blind zone fingerprints of spatial units are generated through feature decoupling mapping. ,in, As an indicator of communication capability, This is a stability indicator for the communication success time window. This is an indicator of the time-varying trend of communication status.

[0009] Optionally, S13 includes: S131, will be used to identify the blind zone fingerprint of the spatial unit. Mapped to blind spot strength score ; S132, standardize the blind spot intensity score to This forms a popularity index. , used to express the depth of color; S133, based on the heat index of each spatial unit Generate a blind spot heatmap. A green color indicates excellent communication, while green indicates the intensity of blind spots. A yellow-green color indicates good communication, while yellow-green indicates the heat of the blind spot. , indicating moderate to weak, and yellow indicates the blind spot heat level. A darker color indicates a more severe blind spot, and orange indicates the intensity of the blind spot. , indicates a severe blind spot, and red indicates the heat level of the blind spot.

[0010] Optionally, S2 includes: S21. By analyzing the heat evolution trend of each spatial unit in the blind zone heat map, the time period when the communication quality is relatively improved and the heat value decreases is extracted. Combined with the location of the spatial unit, the time-location combination of temporary enhancement of communication accessibility in the blind zone is identified and defined as the blind zone opportunity window. A blind zone opportunity window table is constructed for each spatial unit. S22, the blind zone opportunity window table and the acquisition task priority are input into the acquisition strategy generation model, and the opportunity acquisition strategy matching the task level is output, including whether the node is woken up, whether sampling is started, whether to enter the cache standby state, and whether to perform short burst upload.

[0011] Optionally, S21 includes: S211, for each spatial unit Obtain its popularity index sequence within the sliding time window. ; S212, based on heat index sequence Calculate the decrease in heat If satisfied and Then determine This is the starting point of a window of opportunity, where, The threshold for the decrease in popularity, To improve the upper limit of heat; S213, for each spatial unit Based on the starting point of the determined opportunity window, construct a blind spot opportunity window table. ,in, For the first Starting point of a blind spot opportunity window For the first The end time of the opportunity window in the blind spot. The initial popularity index of the window. This corresponds to the decrease in popularity. Score the current accessibility.

[0012] Optionally, S22 includes: S221, during the opportunity acquisition strategy generation phase, for each blind spot opportunity window and the priority of each data collection task The matching score is calculated by generating a model through the data collection strategy. ,when When this occurs, a data collection strategy is generated, in which... The scoring threshold is triggered by the data collection strategy; S222, based on matching score Construct policy action vector group ,in, To wake up a node, determine whether to activate it; 1 indicates waking up the node, and 0 indicates keeping it asleep. For sampling, determine whether to perform sampling; 1 indicates to perform sampling, 0 indicates to skip. For caching, it determines whether to temporarily store data after sampling; 1 indicates caching, and 0 indicates no caching. To upload, determine whether to perform a short burst upload; 1 indicates an attempt to upload, and 0 indicates no immediate upload. Based on matching score Set up an opportunity acquisition strategy, specifically including: like Then all actions are activated, and the complete process is executed, that is... ; like Then, wake-up, sampling, and buffering are performed, i.e. ; like Only wake-up and caching are performed, skipping sampling and uploading. ; like If the strategy is below the wake-up threshold, remain in sleep mode. ; S223, for each spatial unit Data collection task Blind spots and opportunity windows Output opportunity acquisition strategy .

[0013] Optionally, S3 includes: S31, based on the strategy action vector group Configure the execution rules for the data collection nodes, including: like Then the node is awakened; like Then collect sensor data (Collection Node) In the Within the blind spot opportunity window, for the task (Sensor data collected in real time) like Then the sensor data Add to the buffer queue; like The sensor data will then be transmitted via a temporary reachable link. Send to the gateway; S32, the gateway receives the uploaded sensor data stream. Perform consistency checks and anomaly removal to generate a valid dataset, specifically including: S321, Time Consistency Verification: Verify whether the timestamp of the sensor data is within the window range. Inside; S322, Spatial Consistency Verification: Verify whether the source of the sensor data matches the corresponding spatial unit. ; S323, Content validity filtering: Remove sensor data including dropped frames, null values, and abnormal formats; S324, Generate a valid dataset: ; S33, valid dataset The blind spot opportunity window sign The feedback to the blind zone learning model and the data acquisition strategy generation model specifically includes: S331, successfully recorded using in-window communication, generating the dynamic blind zone fingerprint of this spatial unit. ; S332, bias correction is performed on the weight coefficients in the model generated by the acquisition strategy based on the reinjection information.

[0014] An AI-based coal mine data acquisition system, used to implement the aforementioned AI-based coal mine data acquisition method, includes the following modules: Blind spot modeling module: Collects communication reachability records and data loss records generated by each acquisition node and relay node in the low-power wireless sensing network in the coal mine during the reporting process, and performs spatial merging according to preset spatial units to generate blind spot characterization data and input it into the blind spot learning model, and outputs the corresponding blind spot fingerprint and blind spot heat map. Strategy generation module: Based on the blind zone heat map, construct a blind zone opportunity window table, and combine it with the priority information of different collection tasks to generate an opportunity collection strategy for controlling the execution action sequence of the collection nodes; The data acquisition and execution module is deployed at each data acquisition node and is used to control the node to perform wake-up, sampling, caching and burst reporting operations within the blind zone opportunity window according to the opportunity acquisition strategy. Data verification and backfeed module: Performs consistency verification and anomaly removal on the data reported by the nodes, generates a valid dataset, and feeds the valid dataset and its blind spot opportunity window hit status back into the blind spot learning model and the data collection strategy generation model to complete the dynamic self-calibration update of the blind spot fingerprint and blind spot opportunity window tables.

[0015] The beneficial effects of this invention are: This invention constructs a blind zone characterization index that integrates communication reachability records and data missing records, and introduces a graph neural network to model the temporal characteristics of blind zones in underground coal mine spatial units. This enables the identification of blind zone fingerprints that reflect communication strength, window stability, and state change trends, and generates heat maps that reflect coverage differences. This provides data support for intelligent scheduling and blind zone compensation of sensing nodes, thereby enhancing the spatial coverage capability and dynamic adaptability of the sensing network under complex roadway topologies.

[0016] This invention constructs a data acquisition strategy generation model driven by a scoring function by combining the evolution trend of blind zone heat and the priority of data acquisition tasks. It dynamically generates the action sequence of activating nodes, sampling, caching and reporting within the temporary improvement window of communication status, and sets the action intensity level according to the scoring value. This achieves coordinated optimization scheduling of communication reachability and task urgency, effectively reducing node energy consumption and communication load, and improving the data acquisition success rate and data coverage integrity in boundary areas or fluctuating blind zones.

[0017] This invention uses the feedback signals of the returned data and the blind zone opportunity window hit status to update the communication capability vector of the blind zone learning model, and performs bias correction and normalization constraints on the scoring weight coefficients in the strategy generation model. It constructs a self-calibration closed-loop mechanism driven by backfeeding, so as to realize the continuous adaptive update of the system to the changes in blind zone distribution, communication state drift and task priority changes in the mining environment, thereby significantly improving the effectiveness of the acquisition strategy and data continuity in long-term operation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the data acquisition method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system functional modules according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] like Figure 1 As shown, the artificial intelligence-based coal mine data acquisition method includes the following steps: S1, in the low-power wireless sensing network in the coal mine, collect the communication reachability records and data loss records formed by each acquisition node and relay node in the daily reporting process, and spatially merge the communication reachability records and data loss records according to roadway segments, turning points, roadway entrances and equipment operation positions to obtain blind zone characterization data corresponding to each spatial unit. Input the blind zone characterization data into the blind zone learning model, output the blind zone fingerprint of each spatial unit, and generate a blind zone heat map based on the blind zone fingerprint to indicate the strength and drift trend of the blind zone. S2, based on the blind zone heat map, identifies the time period and location combination for effective communication, generates a blind zone opportunity window table corresponding to each spatial unit, and inputs the blind zone opportunity window table into the acquisition strategy generation model to generate an opportunity acquisition strategy that matches the priority of different acquisition tasks. The opportunity acquisition strategy instructs the acquisition node to perform a sequence of actions such as wake-up, sampling, buffering and short burst reporting within the blind zone opportunity window. S3, the data collection nodes perform data collection and reporting according to the opportunity collection strategy. The gateway side performs consistency verification and anomaly removal on the returned data to form a valid dataset. The valid dataset and its corresponding blind spot opportunity window hit status are fed back into the blind spot learning model and the data collection strategy generation model to complete the self-calibration update of the blind spot fingerprint and blind spot opportunity window table, thereby continuously improving the continuity and effectiveness of data collection in the blind spot area.

[0022] S1 includes: S11, in the low-power wireless sensing network environment in coal mines, the acquisition nodes and relay nodes generate communication reachability records and data missing records during daily communication reporting, and divide the dimensions according to the preset spatial units, including roadway sections, turning points, roadway entrances and equipment operation locations, and form a blind area representation dataset that reflects spatial distribution characteristics through geospatial merging. S12, input the blind zone representation dataset into the blind zone learning model, identify the blind zone hidden state vector of each spatial unit, and output the blind zone fingerprint, including indicators describing the strength of communication capability of the region, stability indicators of communication success time window, and indicators of communication state time change trend. S13. Visualize and map the blind zone fingerprint of each spatial unit to generate a blind zone heat map that reflects the communication coverage of the downhole sensing network. The blind zone heat map uses spatial units as grids and uses color depth to indicate the severity of the blind zone.

[0023] S11 includes: S111, in coal mines, continuously records communication reachability and data loss records between data acquisition nodes and relay nodes. Data acquisition nodes include gas sensors, thermometers, hygrometers, and equipment status acquisition devices. Relay nodes are responsible for data relay. Communication reachability records indicate whether a node successfully established a communication connection with the gateway or upstream relay within a certain time period. Data loss records indicate whether a node failed to report data as planned within a certain time period due to signal loss or power depletion. A reachability score is calculated based on the collected communication reachability and data loss records. Data missing rate , is represented as: ; ; in, For the time period internal nodes Number of successful communications For the time period Number of planned communications For the time period internal nodes Number of times data was not reported For the time period The total number of times data was reported by insiders; S112 matches the node coordinates with spatial units of different dimensions in the coal mine, assigning each node to a spatial unit. Establish spatial distribution mapping Furthermore, within each spatial unit, the reachability scores and data missing rates of its constituent nodes are weighted and aggregated to form the spatial unit. Blind zone characterization index , is represented as: ; in, spatial unit The average reachability score of all nodes in the network. spatial unit The average data missing rate across all nodes. , These are the corresponding weighting coefficients; S113, each spatial unit Corresponding blind zone characterization index Number of nodes, communication drift trend and the time trend of missing ratio Packed into blind zone representation vectors Ultimately, the blind zone representation vectors of all spatial units constitute the blind zone representation dataset of the entire mining area; ; ; in, The time window length, This represents the average value over the time period.

[0024] S12 includes: S121, for each spatial unit The accessibility score and data missing ratio of the blind zone representation dataset within a continuous time window are extracted to construct the blind zone temporal input feature sequence. , is represented as: ; ; in, Time period The corresponding blind zone feature vector, This is the first-order change in communication success rate. This represents the first-order change in the missing rate; S122, Input the blind zone time series into the feature sequence. The input is fed into the blind zone learning model to extract the hidden state vector of the spatial unit. ; The blind spot learning model employs a graph neural network (GCN) model, specifically including: S1221, for each spatial unit Obtain the corresponding blind zone temporal input feature sequence. Furthermore, these features are aggregated along the time dimension to construct the node input feature vector of the spatial unit. Input all spatial unit nodes into feature vectors Combined into a node feature matrix , is represented as: ; ; in, The time window length, This refers to the number of spatial units; S1222, based on the spatial topological relationships in underground coal mines, constructs an undirected graph between spatial units. ,in, It is a set of nodes, that is, a set of spatial units. For edge sets, if spatial units and If they are physically adjacent or have communication links, then in and Establish edges between them, and the adjacency matrix of each edge is: By introducing self-loops and performing symmetric normalization, a normalized adjacency matrix is ​​obtained. , is represented as: ; ; ; ; in, It is the identity matrix. To add a self-loop adjacency matrix, It is a degree matrix; S1223, The node feature matrix With normalized adjacency matrix Input a graph neural network, perform graph convolution operations, and output spatial units. Blind zone hidden state vector , is represented as: ; in, For the first The node input feature vector of each spatial unit In the adjacency matrix and The normalized adjacency matrix, This is the weight matrix. It is the ReLU activation function; S123, Based on the hidden state vector of the blind zone Blind zone fingerprints of spatial units are generated through feature decoupling mapping. ,in, As an indicator of communication capability, This is a stability indicator for the communication success time window. The communication status over time trend indicator is represented as: ; ; ; in, , , These are the corresponding weight parameters. This is the Sigmoid function.

[0025] S13 includes: S131, will be used to identify the blind zone fingerprint of the spatial unit. Mapped to blind spot strength score , is represented as: ; in, , , These are the corresponding weight coefficients; S132, standardize the blind spot intensity score to This forms a popularity index. Used to express the depth of color, represented as: ; in, , The minimum and maximum blind zone intensity scores are respectively assigned to all spatial units. S133, based on the heat index of each spatial unit Generate a blind spot heatmap. A green color indicates excellent communication, while green indicates the intensity of blind spots. A yellow-green color indicates good communication, while yellow-green indicates the heat of the blind spot. , indicating moderate to weak, and yellow indicates the blind spot heat level. A darker color indicates a more severe blind spot, and orange indicates the intensity of the blind spot. , indicates a severe blind spot, and red indicates the heat level of the blind spot.

[0026] S2 includes: S21. By analyzing the heat evolution trend of each spatial unit in the blind zone heat map, the time period when the communication quality is relatively improved and the heat value decreases is extracted. Combined with the location of the spatial unit, the time-location combination of temporary enhancement of communication accessibility in the blind zone is identified and defined as the blind zone opportunity window. A blind zone opportunity window table is constructed for each spatial unit. S22, input the blind zone opportunity window table and the collection task priority into the collection strategy generation model, and output the opportunity collection strategy that matches the task level, including whether the node is woken up, whether sampling is started, whether to enter the cache standby state, and whether to perform short burst upload.

[0027] S21 includes: S211, for each spatial unit Obtain its popularity index sequence within the sliding time window. ,in, For the first Each spatial unit at a given time point Popularity index, , The length of the time window; S212, based on heat index sequence Calculate the decrease in heat If satisfied and Then determine This is the starting point of a window of opportunity, where, The threshold for the decrease in popularity, To improve the upper limit of heat, the amount of heat reduction is expressed as: ; S213, for each spatial unit Based on the starting point of the determined opportunity window, construct a blind spot opportunity window table. ,in, For the first Starting point of a blind spot opportunity window For the first The end time of the opportunity window in the blind spot. The initial popularity index of the window. This corresponds to the decrease in popularity. Score the current accessibility. No. The end time of the blind spot opportunity window The strategy is set based on a fixed duration, with each blind spot opportunity window starting from the beginning. Starting from a fixed time period , is represented as: .

[0028] S22 includes: S221, during the opportunity acquisition strategy generation phase, for each blind spot opportunity window and the priority of each data collection task The matching score is calculated by generating a model through the data collection strategy. ,when When this occurs, a data collection strategy is generated, in which... The scoring threshold triggered by the data collection strategy is represented as follows: ; in, , , , These are the corresponding weight coefficients; ; in, This represents the average score of all current blind spot opportunity windows and data collection task combinations to be evaluated. The standard deviation of all current rating values. This is a sensitivity adjustment factor; ; ; ; ; in, As an indicator of mission urgency, Due to the urgency of the sampling deadline, For task importance level, , , These are the corresponding weighting parameters. For the task The current data is outdated for a certain period of time. For the task Maximum allowable data lag time For the task The remaining time until the latest sampling time, The total available sampling time window for the task. For the task The level values ​​(e.g., safety task = 5, equipment operation task = 3, environmental monitoring = 1, etc.). The highest level value across all task types; S222, based on matching score Construct policy action vector group ,in, To wake up a node, determine whether to activate it; 1 indicates waking up the node, and 0 indicates keeping it asleep. For sampling, determine whether to perform sampling; 1 indicates to perform sampling, 0 indicates to skip. For caching, it determines whether to temporarily store data after sampling; 1 indicates caching, and 0 indicates no caching. To upload, determine whether to perform a short burst upload; 1 indicates an attempt to upload, and 0 indicates no immediate upload. Based on matching score Set up an opportunity acquisition strategy, specifically including: like Then all actions are activated, and the complete process is executed, that is... ; like Then, wake-up, sampling, and buffering are performed, i.e. ; like Only wake-up and caching are performed, skipping sampling and uploading. ; like If the strategy is below the wake-up threshold, remain in sleep mode. ; S223, for each spatial unit Data collection task Blind spots and opportunity windows Output opportunity acquisition strategy .

[0029] S3 includes: S31, based on the strategy action vector group Configure the execution rules for the data collection nodes, including: like Then the node is awakened; like Then collect sensor data (Collection Node) In the Within the blind spot opportunity window, for the task (Sensor data collected in real time) like Then the sensor data Add to the buffer queue; like The sensor data will then be transmitted via a temporary reachable link. Send to the gateway; S32, the gateway receives the uploaded sensor data stream. Perform consistency checks and anomaly removal to generate a valid dataset, specifically including: S321, Time Consistency Verification: Verify whether the timestamp of the sensor data is within the window range. Inside; S322, Spatial Consistency Verification: Verify whether the source of the sensor data matches the corresponding spatial unit. ; S323, Content Validity Filtering: Removes sensor data including dropped frames, null values, and abnormal formats, specifically including: (1) Define a certain sensor data The validity determination function is: ; in, This indicates that the data is valid and will be retained and added to the valid dataset. This indicates that the data is invalid and should be discarded. (2) Combination of validity determination criteria: ; in, This is an indicator function that returns 1 if the condition is true, and 0 otherwise. To determine whether the sensor data is empty, To determine if frame drops have occurred, To determine whether the data conforms to the format specifications, such as whether it is a floating-point number, an integer, or a specified enumeration type; (3) If the sampling time of a certain sensor data is Let the previous normal data sampling time be... The expected maximum sampling interval is ; (4) Among all sensor data, filter and retain ; S324, Generate a valid dataset: ; S33, valid dataset The blind spot opportunity window sign The feedback to the blind zone learning model and the data acquisition strategy generation model specifically includes: S331, successfully recorded using in-window communication, generating the dynamic blind zone fingerprint of this spatial unit. , is represented as: ; in, A value of 0 indicates a successful submission within the window; otherwise, a value of 0 indicates a failure. For smoothing coefficients; S332, Based on the reinjection information, the weight coefficients in the model generated by the acquisition strategy are biased and corrected, specifically including: (1) Define the policy execution feedback quantity: ; in, For the first The first spatial unit, the first indivual Opportunity window, the The actual execution result label for each task; (2) Define prediction error: ; in, To account for the discrepancy between the predicted and the actual results; (3) For each weight coefficient Perform the update: ; in, , respectively corresponding , , , , Adjust the learning rate for the weights; (4) Apply normalization constraints to the corrected weight coefficients: .

[0030] like Figure 2 As shown, the AI-based coal mine data acquisition system, used to implement the aforementioned AI-based coal mine data acquisition method, includes the following modules: Blind spot modeling module: Collects communication reachability records and data loss records generated by each acquisition node and relay node in the low-power wireless sensing network in the coal mine during the reporting process, and performs spatial merging according to preset spatial units to generate blind spot characterization data and input it into the blind spot learning model, and outputs the corresponding blind spot fingerprint and blind spot heat map. Strategy generation module: Constructs a blind zone opportunity window table based on the blind zone heat map, and generates an opportunity acquisition strategy to control the action sequence of the acquisition nodes by combining the priority information of different acquisition tasks; The data acquisition and execution module is deployed on each data acquisition node and is used to control the node to perform wake-up, sampling, caching and burst reporting operations within the blind zone opportunity window according to the opportunity acquisition strategy. Data verification and backfeed module: Performs consistency verification and anomaly removal on the data reported by nodes, generates valid datasets, and backfeeds the valid datasets and their blind zone opportunity window hit status to the blind zone learning model and the data collection strategy generation model, completing the dynamic self-calibration update of the blind zone fingerprint and blind zone opportunity window tables.

[0031] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0032] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A coal mine data acquisition method based on artificial intelligence, characterized in that, Includes the following steps: S1. In the low-power wireless sensing network in the coal mine, communication reachability records and data loss records generated by each acquisition node and relay node during daily reporting are collected. The communication reachability records and data loss records are spatially merged according to roadway segments, turning points, roadway entrances and equipment operation positions to obtain blind zone characterization data corresponding to each spatial unit. The blind zone characterization data is input into the blind zone learning model to output the blind zone fingerprint of each spatial unit. Based on the blind zone fingerprint, a blind zone heat map is generated to indicate the strength and drift trend of the blind zone. S2, based on the blind zone heat map, identify the time period and location combination for effective communication, generate a blind zone opportunity window table corresponding to each spatial unit, and input the blind zone opportunity window table into the acquisition strategy generation model to generate an opportunity acquisition strategy that matches the priority of different acquisition tasks. The opportunity acquisition strategy instructs the acquisition node to perform a sequence of actions such as wake-up, sampling, buffering and short burst reporting within the blind zone opportunity window. S3, the acquisition node performs data acquisition and reporting according to the opportunity acquisition strategy. The gateway performs consistency verification and anomaly removal on the returned data to form a valid dataset. The valid dataset and its corresponding blind spot opportunity window hit status are fed back into the blind spot learning model and the acquisition strategy generation model to complete the self-calibration update of the blind spot fingerprint and the blind spot opportunity window table.

2. The coal mine data acquisition method based on artificial intelligence according to claim 1, characterized in that, S1 includes: S11, in the low-power wireless sensing network environment in coal mines, the acquisition nodes and relay nodes generate communication reachability records and data missing records during daily communication reporting, and divide the dimensions according to the preset spatial units, including roadway sections, turning points, roadway entrances and equipment operation locations, and form a blind area representation dataset that reflects spatial distribution characteristics through geospatial merging. S12, input the blind zone representation dataset into the blind zone learning model, identify the blind zone hidden state vector of each spatial unit, and output the blind zone fingerprint, including indicators describing the strength of communication capability of the region, stability indicators of communication success time window, and indicators of communication state time change trend. S13, visualize and map the blind zone fingerprint of each spatial unit to generate a blind zone heat map that reflects the communication coverage of the downhole sensing network. The blind zone heat map uses spatial units as grids and uses color depth to represent the severity of the blind zone.

3. The coal mine data acquisition method based on artificial intelligence according to claim 2, characterized in that, S11 includes: S111, In the coal mine, between the data acquisition nodes and relay nodes deployed underground, communication reachability records and data loss records are continuously recorded during the communication process. The data acquisition nodes include gas sensors, temperature and humidity meters, and equipment status acquisition devices. The relay nodes are responsible for data relay and calculate reachability scores based on the collected communication reachability records and data loss records. Data missing rate ; S112 matches the node coordinates with spatial units of different dimensions in the coal mine, assigning each node to a spatial unit. Establish spatial distribution mapping Furthermore, within each spatial unit, the reachability scores and data missing rates of its constituent nodes are weighted and aggregated to form the spatial unit. Blind zone characterization index ; S113, each spatial unit Corresponding blind zone characterization index Number of nodes, communication drift trend and the time trend of missing ratio Packed into blind zone representation vectors Ultimately, the blind zone representation vectors of all spatial units constitute the blind zone representation dataset for the entire mining area.

4. The coal mine data acquisition method based on artificial intelligence according to claim 3, characterized in that, S12 includes: S121, for each spatial unit The accessibility score and data missing ratio of the blind zone representation dataset within a continuous time window are extracted to construct the blind zone temporal input feature sequence. ; S122, input the blind zone time sequence into the feature sequence. The input is fed into the blind zone learning model to extract the hidden state vector of the spatial unit. ; The blind spot learning model employs a graph neural network model, specifically including: S1221, for each spatial unit Obtain the corresponding blind zone temporal input feature sequence. Furthermore, these features are aggregated along the time dimension to construct the node input feature vector of the spatial unit. Input all spatial unit nodes into feature vectors Combined into a node feature matrix ; S1222, based on the spatial topological relationships in underground coal mines, constructs an undirected graph between spatial units. ,in, It is a set of nodes, that is, a set of spatial units. For edge sets, if spatial units and If they are physically adjacent or have communication links, then in and Establish edges between them, and the adjacency matrix of each edge is: By introducing self-loops and performing symmetric normalization, a normalized adjacency matrix is ​​obtained. ; S1223, The node feature matrix With normalized adjacency matrix Input a graph neural network, perform graph convolution operations, and output spatial units. Blind zone hidden state vector ; S123, based on the hidden state vector of the blind zone Blind zone fingerprints of spatial units are generated through feature decoupling mapping. ,in, As an indicator of communication capability, This is a stability indicator for the communication success time window. This is an indicator of the time-varying trend of communication status.

5. The coal mine data acquisition method based on artificial intelligence according to claim 4, characterized in that, S13 includes: S131, will be used to identify the blind zone fingerprint of the spatial unit. Mapped to blind spot strength score ; S132, standardize the blind spot intensity score to This forms a popularity index. , used to express the depth of color; S133, based on the heat index of each spatial unit Generate a blind spot heatmap. A green color indicates excellent communication, while green indicates the intensity of blind spots. A yellow-green color indicates good communication, while yellow-green indicates the heat of the blind spot. , indicating moderate to weak, and yellow indicates the blind spot heat level. A darker color indicates a more severe blind spot, and orange indicates the intensity of the blind spot. , indicates a severe blind spot, and red indicates the heat level of the blind spot.

6. The coal mine data acquisition method based on artificial intelligence according to claim 5, characterized in that, S2 includes: S21. By analyzing the heat evolution trend of each spatial unit in the blind zone heat map, the time period when the communication quality is relatively improved and the heat value decreases is extracted. Combined with the location of the spatial unit, the time-location combination of temporary enhancement of communication accessibility in the blind zone is identified and defined as the blind zone opportunity window. A blind zone opportunity window table is constructed for each spatial unit. S22, the blind zone opportunity window table and the acquisition task priority are input into the acquisition strategy generation model, and the opportunity acquisition strategy matching the task level is output, including whether the node is woken up, whether sampling is started, whether to enter the cache standby state, and whether to perform short burst upload.

7. The coal mine data acquisition method based on artificial intelligence according to claim 6, characterized in that, S21 includes: S211, for each spatial unit Obtain its popularity index sequence within the sliding time window. ; S212, based on heat index sequence Calculate the decrease in heat If satisfied and Then determine This is the starting point of a window of opportunity, where, The threshold for the decrease in popularity, To improve the upper limit of heat; S213, for each spatial unit Based on the starting point of the determined opportunity window, construct a blind spot opportunity window table. ,in, For the first Starting point of a blind spot opportunity window For the first The end time of the opportunity window in the blind spot. The initial popularity index of the window. This corresponds to the decrease in popularity. Score the current accessibility.

8. The coal mine data acquisition method based on artificial intelligence according to claim 7, characterized in that, S22 includes: S221, during the opportunity acquisition strategy generation phase, for each blind spot opportunity window and the priority of each data collection task The matching score is calculated by generating a model through the data collection strategy. ,when At that time, the collection strategy is generated, in which, The scoring threshold is triggered by the data collection strategy; S222, based on matching score Construct policy action vector group ,in, To wake up a node, determine whether to activate it; 1 indicates waking up the node, and 0 indicates keeping it asleep. For sampling, determine whether to perform sampling; 1 indicates to perform sampling, 0 indicates to skip. For caching, it determines whether to temporarily store data after sampling; 1 indicates caching, and 0 indicates no caching. To upload, determine whether to perform a short burst upload; 1 indicates an attempt to upload, and 0 indicates no immediate upload. Based on matching score Set up an opportunity acquisition strategy, specifically including: like Then all actions are activated, and the complete process is executed, that is... ; like Then, wake-up, sampling, and buffering are performed, i.e. ; like Only wake-up and caching are performed, skipping sampling and uploading. ; like If the strategy is below the wake-up threshold, remain in sleep mode. ; S223, for each spatial unit Data collection task Blind spots and opportunity windows Output opportunity acquisition strategy .

9. The coal mine data acquisition method based on artificial intelligence according to claim 8, characterized in that, S3 includes: S31, based on the strategy action vector group Configure the execution rules for the data collection nodes, including: like Then the node is awakened; like Then collect sensor data ; like Then the sensor data Add to the buffer queue; like The sensor data will then be transmitted via a temporary reachable link. Send to the gateway; S32, the gateway receives the uploaded sensor data stream. Perform consistency checks and anomaly removal to generate a valid dataset, specifically including: S321, Time Consistency Verification: Verify whether the timestamp of the sensor data is within the window range. Inside; S322, Spatial Consistency Verification: Verify whether the source of the sensor data matches the corresponding spatial unit. ; S323, Content validity filtering: Remove sensor data including dropped frames, null values, and abnormal formats; S324, Generate a valid dataset: ; S33, valid dataset The blind spot opportunity window sign The feedback to the blind zone learning model and the data acquisition strategy generation model specifically includes: S331, successfully recorded using in-window communication, generating the dynamic blind zone fingerprint of this spatial unit. ; S332, bias correction is performed on the weight coefficients in the model generated by the acquisition strategy based on the reinjection information.

10. An artificial intelligence-based coal mine data acquisition system, used to implement the artificial intelligence-based coal mine data acquisition method as described in any one of claims 1-9, characterized in that, Includes the following modules: Blind spot modeling module: Collects communication reachability records and data loss records generated by each acquisition node and relay node in the low-power wireless sensing network in the coal mine during the reporting process, and performs spatial merging according to preset spatial units to generate blind spot characterization data and input it into the blind spot learning model, and outputs the corresponding blind spot fingerprint and blind spot heat map. Strategy generation module: Based on the blind zone heat map, construct a blind zone opportunity window table, and combine it with the priority information of different collection tasks to generate an opportunity collection strategy for controlling the execution action sequence of the collection nodes; The data acquisition and execution module is deployed at each data acquisition node and is used to control the node to perform wake-up, sampling, caching and burst reporting operations within the blind zone opportunity window according to the opportunity acquisition strategy. Data verification and backfeed module: Performs consistency verification and anomaly removal on the data reported by the nodes, generates a valid dataset, and feeds the valid dataset and its blind spot opportunity window hit status back into the blind spot learning model and the data collection strategy generation model to complete the dynamic self-calibration update of the blind spot fingerprint and blind spot opportunity window tables.

Images