A coal mine multi-source data aggregation management method and platform

Abstract

The application discloses a kind of coal mine multi-source data gathering management method and platform, it is related to coal mine data processing technical field, platform uses five progressive type architecture of sensing layer-edge layer-transmission layer-platform layer-application layer, method covers multi-source heterogeneous data acquisition and edge integrated processing, double-ring redundant low-delay transmission, data depth management, Kafka priority subscription distribution, whole-service collaborative control feedback whole process, by multi-protocol adaptation, 3 sigma criterion+linear interpolation data cleaning, hierarchical fusion intelligent labeling, hierarchical directional distribution and closed-loop feedback etc. Technology, solve coal mine multi-protocol access adaptation, data efficient flow, high-quality management and whole-service collaborative control industry problem.The application greatly improves data acquisition, flow, management efficiency and reliability, realizes coal mine whole-service data intercommunication and collaborative control, reduces the incidence of safety accidents, improves production and operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of coal mine data processing technology, and more specifically to a method and platform for the aggregation and management of multi-source data in coal mines. Background Technology

[0002] Large and medium-sized coal mines in China have gradually deployed various business subsystems, including gas extraction, ventilation scheduling, coal mining operations, personnel positioning, and business accounting. They have introduced programmable logic controllers (PLCs) from multiple manufacturers such as Siemens and Schneider Electric, as well as upper-level computers such as WinCC / Intouch, and various sensor terminals for gas, wind speed, temperature, and vibration, attempting to improve the level of coal mine safety production and business management efficiency through data-driven monitoring and control.

[0003] However, the existing technical system for coal mine data management and business control still has many prominent and urgent industry-wide problems that cannot be adapted to the actual needs of intelligent development in coal mines. These problems are specifically reflected in the following aspects: The data acquisition interfaces exhibit significant heterogeneity, resulting in extremely low equipment adaptation efficiency: The equipment used in various business subsystems comes from different manufacturers, and the communication protocols cover more than seven categories, including Profinet, Modbus, OPC UA, and DDE. Existing technologies lack universal protocol adaptation capabilities, requiring customized development of data access interfaces for each type of equipment. The debugging cycle for a single interface can be as long as 2-4 weeks. When adding new equipment, adaptation programs must be redeveloped, leading to extremely poor horizontal scalability. In 2023, a coal mine added three coal mining machines, and the interface development alone took 12 weeks, severely slowing down the overall progress of the mine's intelligent transformation.

[0004] Inefficient data flow and lack of priority scheduling lead to delayed safety alarm responses: Existing technologies generally employ a three-tier transmission architecture of "subsystem → local server → group server," with measured end-to-end data transmission latency exceeding 500ms. Furthermore, the absence of a data priority scheduling mechanism results in the mixed transmission of emergency alarm data such as gas exceedances and fan malfunctions with routine data like ambient temperature and equipment standby status, causing an average safety alarm response time of 12.5 minutes. In 2023, a coal mine experienced three gas concentration exceedance incidents, which nearly triggered major safety accidents due to delayed alarm responses, seriously threatening underground operational safety.

[0005] The platform functions are fragmented and data silos are serious: existing coal mine management and control platforms mostly focus on the development of single business scenarios and lack an integrated design concept. The five coal mines surveyed deployed an average of five independent business subsystems, including safety monitoring, production scheduling, business statistics, emergency rescue, equipment management, etc. The data standards of the various platforms are not unified and cannot be interconnected, forming serious information silos.

[0006] The lack of redundancy in the transmission link leads to insufficient reliability and can easily cause production interruptions: Existing coal mine data transmission networks mostly rely on a single-link architecture without redundancy backup and automatic fault switching mechanisms. Once problems such as underground cable damage or switch failure occur, it will directly lead to data transmission interruption, failure of on-site monitoring, and even force the coal mine to suspend production.

[0007] Weak data governance capabilities and inadequate data quality to support scientific decision-making: In the heterogeneous multi-source data of coal mines, outliers account for 3%-5% due to factors such as sensor failure, electromagnetic interference, and transmission fluctuations, while occasional missing values ​​account for 0.8%-1.2%. Existing technologies lack standardized governance methods for removing outliers and completing missing data, and a unified data fusion and identification system has not been established, resulting in poor data quality and high redundancy.

[0008] In summary, existing technologies have significant technical shortcomings in areas such as the acquisition, adaptation, transmission, governance, integration, and collaborative management of multi-source data in coal mines. They fail to address the core challenges of multi-protocol access adaptation, efficient and reliable data flow, high-quality governance, and comprehensive business collaborative management in the coal mining industry, becoming a key bottleneck restricting the deep intelligent transformation of coal mines. Against this backdrop, developing a multi-source data aggregation and governance method and platform adapted to the characteristics of the coal mining industry, and achieving high-quality control and collaborative linkage of coal mine data throughout the entire process, has become an urgent need for the intelligent development of coal mines. Summary of the Invention

[0009] In view of this, the present invention provides a method and platform for the aggregation and management of multi-source data in coal mines to solve the problems existing in the background art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A method for the aggregation and governance of multi-source data in coal mines includes: S1, collect multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data; S2, perform edge preprocessing on the multi-source heterogeneous data, encapsulate it, and add a unique device identifier; S3 performs edge data cleaning on the encapsulated data to obtain valid data; S4. Establish a primary and backup dual-ring network, and transmit effective data to the server through the primary and backup dual-ring network; S5, the server performs data fusion and tagging on the valid data, and stores it in layers, and distributes the layered data based on priority; S6, the hierarchically stored data is fed back to the full-service collaborative management and feedback center, and processed and mobilized based on the full-service collaborative management and feedback center.

[0011] Preferably, the coal mine-related equipment includes heterogeneous equipment and a host computer. The heterogeneous equipment connects to a Siemens S7-1500 programmable logic controller via a dedicated contrib-s7 communication node built into the gateway, and is compatible with the Profinet industrial Ethernet protocol; it connects to a Schneider M340 PLC via a contrib-modbus communication node, and is compatible with the Modbus RTU / TCP dual protocol, to collect heterogeneous data. The host computer connects to a WinCC host computer via a contrib-opcua node, and is compatible with the OPC unified architecture protocol, to collect production data; it connects to an Intouch host computer via a contrib-dde node, and is compatible with the dynamic data exchange protocol, allowing for data access from older equipment. The multi-source heterogeneous data includes heterogeneous data, data from older equipment, and production data.

[0012] Preferably, the edge preprocessing includes median filtering and Gaussian filtering. A threshold is set based on the sensor's inherent range. When the data exceeds the threshold, it is directly marked as invalid and discarded. Noise within the range is filtered by first using median filtering, then applying Gaussian filtering to the processed data, and finally encapsulating the data to obtain the encapsulated data.

[0013] Preferably, the edge data cleaning includes: removing abnormal data based on the statistical characteristics of normal distribution, performing double verification based on a secondary cleaning mechanism, and performing linear interpolation to complete the verified data to obtain valid data.

[0014] Preferably, the primary and backup dual-ring network includes: a primary and backup dual-ring network built through a ground core switch and an underground intrinsically safe switch, wherein the primary ring uses 4-core single-mode fiber and the backup ring uses 2-core single-mode fiber; the primary ring transmits valid data at a rate of 1000Mbps and the backup ring is hot-standby at a rate of 100Mbps; when a single point of failure is detected, the system automatically triggers the primary and backup ring switching, with a switching response time of ≤50ms; the system automatically performs link connectivity detection every 30s, and sends a link failure alarm through the platform layer system when an anomaly occurs.

[0015] Preferably, the data fusion and tagging specifically includes: constructing a two-level fusion architecture of feature layer and decision layer to adapt to the heterogeneous data characteristics of coal mines; feature layer fusion adopts a weighted feature projection algorithm to extract time-domain and frequency-domain features from multi-dimensional time-series data of the same equipment, and assigns weights to each feature through entropy weight method to map high-dimensional features to a unified feature space to eliminate data redundancy; decision layer fusion is based on DS evidence theory, fusion of multi-device associated data, construction of trust function and likelihood function, outputting fusion results, and synchronous data alignment during the fusion process, and time synchronization based on NTP / PTP protocol; character encoding is used to achieve unique identification of data in all dimensions, and tag fields are automatically adjusted based on changes in equipment operating status, with bidirectional synchronization between edge layer and platform layer; based on the coal mine safety production knowledge graph, a tag semantic association model is constructed to achieve cross-business tag interoperability, and automatic mapping rules between tags and Topic subscriptions are designed to automatically distribute data to the corresponding Kafka Topic partition according to the parameter type and priority field in the tag.

[0016] A coal mine multi-source data aggregation and governance platform includes: The perception layer, which has built-in sensors, collects multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data. The edge layer performs edge preprocessing on the multi-source heterogeneous data, encapsulates it, and adds a unique device identifier; it also performs edge data cleaning on the encapsulated data to obtain valid data. The transport layer is equipped with a primary and backup dual-ring network, through which effective data is transmitted to the server. At the platform layer, the servers perform data fusion and tagging on the valid data, store it in layers, and distribute the layered data based on priority. At the application layer, the hierarchically stored data is fed back to the full-service collaborative management and feedback center, which then processes and mobilizes the data.

[0017] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and platform for the aggregation and management of multi-source data in coal mines, with the following beneficial effects: 1. Strong multi-protocol access adaptability: It is compatible with 7 mainstream industrial protocols such as Profinet, Modbus, OPC UA, and DDE, enabling unified access of more than 100 heterogeneous devices. The debugging time for a single device is reduced from the traditional 2-4 weeks to 1 hour, the adaptation efficiency is improved by 97%, and the data acquisition success rate is stable at over 99.8%.

[0018] 2. High data flow efficiency and reliability: End-to-end data transmission latency ≤100ms, high-priority alarm data flow latency ≤50ms, alarm response time reduced from the traditional 12min to 1min, emergency response efficiency improved by 1100%; transmission link single-point failure switching time ≤50ms, annual transmission interruption duration ≤10min, reliability improved by 99.9% compared to traditional networks.

[0019] 3. Significantly improved efficiency of full-business collaborative management: Breaking down information silos and achieving data sharing among production, safety, operations, and emergency response, the incidence of coal mine gas exceeding limits has decreased by 68%, the downtime of coal mining machines due to malfunctions has decreased from 8 hours / month to 2 hours / month, and the overall utilization rate of equipment has increased by 15 percentage points; energy consumption per ton of coal production has decreased from 8.5 kWh to 7.2 kWh, saving approximately 2.4 million yuan in electricity costs annually, and reducing operating costs by 8%.

[0020] 4. High-quality data governance: Through the 3σ criterion and linear interpolation, the accuracy of outlier removal is ≥99%, and the data integrity is ≥99.5%; the standardized labeling system improves data retrieval efficiency by 60%, providing accurate data support for decision-making.

[0021] 5. Strong compatibility and scalability: Supports seamless integration of old and new equipment, with debugging time for new equipment ≤ 2 hours; the platform architecture supports on-demand expansion and can adapt to coal mines with different production capacities of 300,000 to 1.2 million tons / year without the need to reconstruct the core system. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 The flowchart of the multi-source data aggregation and governance method provided by this invention; Figure 2 The platform architecture diagram provided for this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention discloses a method for the aggregation and management of multi-source data in coal mines, such as... Figure 1 As shown, it includes: S1, collect multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data; S2 performs edge preprocessing on multi-source heterogeneous data, encapsulates it, and adds a unique device identifier; S3 performs edge data cleaning on the encapsulated data to obtain valid data; S4, set up a primary and backup dual-ring network, and transmit effective data to the server through the primary and backup dual-ring network; S5: The server performs data fusion and tagging on the valid data, stores it in layers, and distributes the layered data based on priority. S6 feeds back the hierarchically stored data to the full-service collaborative management and feedback center, and processes and mobilizes the data based on the full-service collaborative management and feedback center.

[0026] In specific embodiment 1: Step 1: Intelligent acquisition and edge processing of multi-source heterogeneous data (Execution subject: GeniusLinksGateway ZL-2025 intelligent gateway) 1. Heterogeneous Device Access: The gateway connects to a Siemens S7-1500 programmable logic controller (PLC) via its built-in contrib-s7 dedicated communication node, supporting the Profinet industrial Ethernet protocol and offering flexible frequency configuration from 100 to 1000ms (adapting to different device response speeds). It also connects to a Schneider M340 PLC via the contrib-modbus communication node, supporting Modbus RTU / TCP dual protocols to collect key parameters such as fan speed and damper opening (purpose: resolving protocol incompatibility issues between multiple vendors). 2. Host Computer Data Access: Connect to the WinCC host computer via the contrib-opcua node, adapting to the OPC Unified Architecture protocol (OPC UA: Open Platform Communications Unified Architecture), to collect production data such as coal mining machine position and cutting current; connect to the Intouch host computer (version ≤ 10.0) via the contrib-dde node, adapting to the Dynamic Data Exchange protocol (DDE: Dynamic Data Exchange), and ensuring compatibility with data access from older equipment (purpose: to achieve full data coverage for both new and old equipment); 3. Edge preprocessing: Gaussian filtering algorithm is used to filter out abnormal data outside the sensor range (such as invalid values ​​with gas concentration > 5%), and histogram equalization is used to optimize fuzzy data; multi-source heterogeneous data are uniformly encapsulated into JSON format (fields: device ID, acquisition timestamp, parameter name, value, data tag), and a unique device identifier is added (such as "gas-001" corresponding to gas sensor #1) (purpose: to reduce the data processing pressure of the platform layer and improve the flow efficiency).

[0027] (1) Two-layer filtering of abnormal data: First layer: Invalid values ​​outside the measurement range are directly eliminated. A threshold is set based on the sensor's inherent measurement range (e.g., for a gas sensor with a range of 0~5%CH4, the threshold range is set to [0,5], and when data X... When the value is [0,5], it is directly marked as invalid and discarded. The second layer: In-range noise filtering. A targeted strategy of "adapting filtering methods to sensor type + selecting algorithms based on noise type" is adopted. The core logic is a combined filtering approach of "median first, then Gaussian"—median filtering is used for impulse / salt-and-pepper noise, Gaussian filtering is used for continuous Gaussian noise, and a combination of both is used for mixed noise. This ensures effective noise reduction while avoiding noise propagation or data trend distortion. Specific adaptation is as follows: Single-point sensor (one-dimensional time series data) adaptation: For sensors deployed at a single point in underground coal mines (such as single-point gas sensors and single-point temperature sensors), whose output is one-dimensional time-series data, the corresponding noise types and filtering methods are as follows: Pulse-type noise (such as spikes / troughs caused by poor sensor contact or electromagnetic interference, e.g., a single-point gas sensor suddenly spikes 2.0% CH4 and then drops rapidly, deviating from the true concentration by 0.85%): One-dimensional median filtering is used, with a 3-5 point sliding window configured. The median value of the data within the window is taken to replace the core data, which can quickly remove spike noise without affecting the time series trend. Gaussian continuous noise (such as sensor electronic thermal noise, small continuous fluctuations caused by circuit drift, e.g., a single-point temperature sensor fluctuating within ±0.1℃ of its true value at 25℃): One-dimensional Gaussian filtering is used, with the core function being a one-dimensional Gaussian weight calculation function.

[0028] Where x is the temporal offset of the data within the window relative to the central data (x∈{-1,0,1} for a 3-point window, x∈{-2,-1,0,1,2} for a 5-point window), σ takes a value of 0.8~1.0 (the optimal value measured in the coal mine scenario), and e is a natural constant; during execution, the temporal data is convolved with a one-dimensional Gaussian kernel, with the central data having the highest weight and the edge data having decreasing weights, which smooths out noise while reducing temporal lag; Mixed noise (most single-point sensor scenarios, with both pulse and Gaussian noise): A combination of "one-dimensional median filtering + one-dimensional Gaussian filtering" is used. First, the pulse spikes are removed by median filtering, and then the remaining continuous noise is smoothed by Gaussian filtering, balancing the denoising effect and the data authenticity.

[0029] Sensor array (two-dimensional spatial data) adaptation: For underground sensor arrays deployed in a grid pattern (such as the wind speed sensor array in the 401 working face and the gas sensor array in the mining area), the output is a two-dimensional spatial data matrix. The corresponding noise types and filtering methods are as follows: Impulse noise (such as local anomalies caused by the failure of a single sensor in the array, for example: a node in a 5×5 wind speed sensor array suddenly drops to 0 m / s, while the wind speeds of surrounding nodes are all 1.2~1.5 m / s): Two-dimensional median filtering is used, a 3×3 sliding window is configured, and the median value of the data within the window is taken to replace the core node data, which accurately removes single-point impulse noise without destroying the spatial data distribution pattern. Gaussian continuous noise (such as array sensor consistency error, local continuous fluctuations caused by spatial environmental scattering, e.g., small spatial fluctuations of wind speed array data within ±0.05 m / s of the true value): Two-dimensional Gaussian filtering is used, with the core function retaining the original two-dimensional Gaussian convolution kernel weight calculation function.

[0030] Where (x,y) are the coordinates of the pixels within the convolution kernel relative to the center pixel (3×3 window x,y∈{-1,0,1}), σ is 0.8, and spatial continuous noise is smoothed by pixel-by-pixel convolution, while preserving the spatial gradients of wind speed and gas concentration (such as the wind speed gradient from the air intake side to the air return side of the working face). Mixed noise (mainstream array scenarios, e.g., wind speed arrays have both individual node pulse spikes and overall spatial continuous fluctuations): A combination of "two-dimensional median filtering + two-dimensional Gaussian filtering" is used. First, the two-dimensional median filtering removes single-point pulse fault values ​​in the array, and then the two-dimensional Gaussian filtering smooths the spatial continuous noise, avoiding the problem of spreading pulse noise when the two-dimensional Gaussian filtering is used alone.

[0031] (2) Data standardization and encapsulation: Multi-source heterogeneous data (sensor values, video frame feature values, etc.) are uniformly encapsulated into JSON format, and field definitions strictly follow the following rules: ① Rule 1: Device ID, using the encoding rule of "device type abbreviation-serial number" (e.g., "gas-001" corresponds to gas sensor #1, "fan-003" corresponds to fan #3), the encoding function is: DeviceID=TypeAbbr+"-"+String(SerialNo) (TypeAbbr is a 2~4 digit English abbreviation, SerialNo is a 3 digit numeric serial number); ② Rule 2: Collect timestamps in UTC time format, accurate to milliseconds, in the format "YYYY-MM-DD HH:MM:SS.sss"; ③ Rule 3: Core fields: Device ID, data collection timestamp, parameter name, value, and data tag (the tag consists of device type, work area, and parameter type, separated by "|", such as "coal mining machine|401 working face|cutting current"); (Purpose: to optimize data quality through algorithms, standardize encapsulation to reduce data processing pressure at the platform layer, and improve workflow efficiency).

[0032] 4. Edge Data Cleaning: Utilizing the gateway's built-in ARM Cortex-A76 computing module, data cleaning algorithms are executed synchronously to remove outliers and fill in missing values, forming an integrated closed loop of "preprocessing + cleaning," ensuring only valid data is uploaded. (Purpose: To reduce data volume at the source, decrease the load on the transmission layer, and avoid redundant processing at the platform layer.)

[0033] (1) 3σ criterion outlier identification model: Core principle: Based on the statistical characteristics of normal distribution, it is assumed that coal mine monitoring data (such as gas concentration and fan speed) follows a normal distribution N(μ,σ²), where 99.73% of the normal data will fall within the interval [μ-3σ,μ+3σ], and data outside this interval are judged as abnormal (abnormal, interference and other invalid data). Parameter calculation process: ① Sample data selection: Select the continuous valid data of the previous hour of the data to be judged X as the sample set S (sample size ≥ 30, to ensure statistical significance, collection frequency matching the sensing layer 100~1000ms, sample size range 3600~36000). ② Calculation of mean μ: The central tendency of the sample set is solved using the arithmetic mean formula, which is: ; Where n is the sample size, and S_i is the i-th valid data in the sample set; ③ Calculation of standard deviation σ: The dispersion of the data is solved by the sample standard deviation formula, which is: ; (Using an unbiased estimate of n-1 improves computational accuracy in small sample scenarios.) ④ Anomaly Detection and Handling: Substitute the data X to be judged into the judgment condition |X-μ|>3σ. If the condition is met, it is marked as abnormal data. For critical parameter anomalies such as sudden changes in gas concentration, an additional secondary verification is triggered (comparing the data from adjacent sensors in the same area during the same period to avoid misjudgment due to single-point sensor failure). After confirming the anomaly, an elimination operation is performed, and an anomaly log is recorded (including the anomaly time, data value, and sensor ID). Substitute the data X to be judged into the judgment condition |X-μ|>3σ. If the condition is met, it is marked as a candidate value for abnormal data. For critical parameters such as sudden changes in gas concentration, considering that the distance between adjacent sensors may lead to natural differences in gas concentration, comparing the data from adjacent sensors during the same period alone is prone to misjudgment. Therefore, a "double secondary verification" mechanism is triggered. First verification (comparison with adjacent sensors): Select adjacent sensors within 50m of the original sensor in the same area (if insufficient, select the nearest 2-3 sensors), and compare the difference in data from the same period. If the difference is ≤0.3%CH4 (the natural gradient threshold for coal mine gas concentration, optimized by actual measurement), it is temporarily judged as normal data (excluding concentration differences caused by distance); if the difference is >0.3%CH4, proceed to the second verification.

[0034] The second verification (historical mean comparison): Extract the valid monitoring data of the sensor from the same period (same time period, same operating conditions) of the previous day, calculate the historical mean μ_hist and the historical sample standard deviation σ_hist (calculation method is the same as the previous μ and σ, sample size ≥ 3600). If |X-μ_hist|>3σ_hist, the data is confirmed to be abnormal (excluding sudden changes caused by sensor failure, electromagnetic interference, etc.); if |X-μ_hist|≤3σ_hist, it is determined to be a natural fluctuation of gas concentration, temporarily classified as normal data and monitored continuously.

[0035] Once an anomaly is confirmed, a removal operation is performed, and an anomaly log is recorded (including anomaly time, data value, sensor ID, and dual verification result). For confirmed anomalies, a sensor fault troubleshooting prompt is triggered simultaneously and pushed to the terminal.

[0036] (2) Linear interpolation completion model: Prerequisites: This method is only applicable to occasional missing data (missing rate ≤ 0.5%), where the missing data is a single point or a discontinuous missing number of no more than 3 consecutive data points, and adjacent data (Xi-1, Xi+1) are all valid data (passed by the 3σ criterion). Core formula and parameter definitions: The completed formula is as follows

[0037] in: X i : The imputation value for the i-th missing data; Xi-1, Xi+1: The two valid data values ​​immediately before and after the missing data; ti, ti-1, ti+1: corresponding data collection timestamps (unit: ms, precisely matching the collection frequency of the sensing layer); Execution process: ① Missing data location: Traverse the data sequence and mark the location of missing data and the length of consecutive missing data; ② Validity verification: Determine whether the missing rate is ≤0.5% and the length of consecutive missing values ​​is ≤3. If not, trigger a sensor fault alarm and do not perform interpolation. ③ Interpolation calculation: Substitute into the formula to solve for the completed value X i ; ④ Result verification: Complete the value X i Substitute into the 3σ criterion model; if it falls within the interval [μ-3σ, μ+3σ], retain it; otherwise, mark it as an anomaly and trigger manual review. Example: Taking the No. 1 gas sensor (device ID: gas-001) in the 401 working face as an example, the acquisition frequency is configured to 1000ms (1 data point per second). At a certain moment, a single point in the data sequence is missing. The specific data is as follows: Previous valid data (X_i-1): Acquisition time t_i-1=2025-10-20 14:30:01.000, gas concentration value 0.82%CH4; Missing data (X_i): Data collection time t_i=2025-10-20 14:30:02.000 is missing; Subsequent valid data (X_i+1): Acquisition time t_i+1=2025-10-20 14:30:03.000, gas concentration value 0.86%CH4.

[0038] Step 1: Missing data location and validity verification. The missing data is a single point of failure, and the current data sequence missing rate is 0.2% ≤ 0.5%, which meets the applicable conditions. Step 2: Interpolation calculation, substituting into the formula: Time difference conversion (unit: ms): t_i - t_i-1 = 1000ms, t_i+1 - t_i-1 = 2000ms; X_i=0.82+(0.86-0.82)×(1000 / 2000)=0.82+0.04×0.5=0.84%CH4; Step 3: Result verification. Extract the data from the sensor for the previous hour and calculate μ=0.83%CH4, σ=0.02%CH4, and the 3σ interval is [0.77, 0.89]. The padded value 0.84%CH4 falls within the interval. This padded value is retained and written into the data sequence.

[0039] After cleaning is completed, the gateway adds a "cleaning identifier" (such as "cleaned - normal" or "cleaned - interpolation completion") to the data and uploads it synchronously to the transport layer to ensure that the platform layer can directly identify the data status.

[0040] (Overall function of the model: Through statistical modeling, it can accurately remove outlier data and reasonably complete missing data, ensuring data integrity ≥99.5% and accuracy ≥99%, providing a high-quality data source for subsequent data fusion.) Step 2: Dual-ring redundant low-latency transmission (Execution subject: Dual-ring redundant industrial Ethernet system) 1. Link Setup: A dual-ring network is set up using a ground core switch (Huawei S5735S-L48T4S-A) and an underground intrinsically safe switch (KTJ127, explosion-proof rating ExdI Mb, one unit deployed every 500m). The main ring uses 4-core single-mode fiber, and the backup ring uses 2-core single-mode fiber (function: to ensure link redundancy backup). 2. Data transmission: The main ring transmits core data (alarm data, real-time monitoring data) at a rate of 1000Mbps, and the backup ring is hot standby at a rate of 100Mbps; when a single point of failure is detected, the system automatically triggers the main and backup ring switching, with a switching response time of ≤50ms (purpose: to avoid data transmission interruption). 3. Link verification: Automatically check link connectivity every 30 seconds. If an anomaly is detected, send a link failure alarm through the platform layer system (purpose: to provide early warning of transmission risks).

[0041] Step 3: Multi-dimensional data in-depth governance (Execution subject: GeniusLinks IoT management system server, CPU: Intel Xeon Gold 6330, Memory: 32GB)

[0042] Based on the basic cleaning already completed at the edge layer, the focus is on in-depth data processing, eliminating the need to repeat the cleaning operation and improving governance efficiency.

[0043] 1. Data Fusion and Tagging: A standardized tagging system will be established in accordance with the requirements of the "Guiding Opinions of the State Administration of Coal Mine Safety on Accelerating the Construction of Coal Mine Safety Risk Monitoring and Early Warning System (Coal Mine Safety Supervision Office

[2019] No. 42)". A "layered fusion + intelligent coding + dynamic association" technical approach will be adopted to achieve multi-source data value mining and standardized management, providing support for business collaboration and precise scheduling. (Function: Supporting targeted distribution and business applications); Layered data fusion technology: A two-tiered fusion architecture, "feature layer - decision layer," is constructed to adapt to the heterogeneous data characteristics of coal mines (time-series monitoring data, equipment static parameters, and semantic business data). Feature layer fusion employs a weighted feature projection algorithm to extract time-domain and frequency-domain features from multi-dimensional time-series data of the same equipment (such as coal mining machine cutting current, vibration frequency, and bearing temperature). Entropy weighting is used to assign weights to each feature (avoiding subjective weighting bias), mapping high-dimensional features to a unified feature space and eliminating data redundancy. Decision layer fusion, based on DS evidence theory, fuses data from multiple devices (such as gas sensor and ventilation fan operation data, personnel positioning and working face equipment data). Trust functions and likelihood functions are constructed to resolve uncertainties and conflicts in multi-source data, outputting fusion results such as equipment health and working face safety level, with a fusion accuracy ≥98.5%. Data alignment is performed synchronously during the fusion process. Time synchronization is achieved based on NTP / PTP protocols (synchronization accuracy ≤1ms), and spatial calibration is performed based on the coal mine underground coordinate system to ensure data consistency across equipment and regions.

[0044] The intelligent tag coding system employs a standardized "five-level linkage" tag coding rule, using an 18-bit character code (letters + numbers) to achieve unique identification of data across all dimensions. The coding format is: [Equipment Type (2 digits) - Working Area (4 digits) - Parameter Type (3 digits) - Data Priority (1 digit) - Data Status (2 digits) - Timestamp Hash (6 digits)]. Specifically, the equipment type is coded according to the coal mine equipment classification standard (e.g., "CM" represents a coal mining machine, "FS" represents a ventilation fan); the working area uses the underground roadway numbering rule (e.g., "0401" represents the 401 working face); the data status is associated with the edge layer cleaning identifier (e.g., "01" represents cleaned - normal, "02" represents cleaned - interpolated completion); and the timestamp hash generates a unique checksum based on the collection timestamp to prevent tag duplication and data tampering. Simultaneously, a dynamic tag update mechanism is established, automatically adjusting tag fields based on changes in equipment operating status (e.g., equipment maintenance, working condition switching). Through bidirectional synchronization between the edge layer and the platform layer, real-time matching between tags and data status is ensured.

[0045] Tag Semantic Association and Mapping: Based on the knowledge graph of coal mine safety production, a tag semantic association model is constructed to achieve cross-business tag interoperability (e.g., associating "coal mining machine operation status tag" with "working face output tag" and "equipment maintenance tag"), breaking down data silos. Automatic mapping rules between tags and Topic subscriptions are designed. Based on the "parameter type" and "priority" fields in the tag, data is automatically distributed to the corresponding Kafka Topic partition (e.g., high-priority safety tag data is mapped to the gas extraction alarm Topic), requiring no manual configuration, with a data distribution response time ≤10ms.

[0046] 2. Tiered Storage: A Redis real-time database (128GB cache capacity) stores nearly 24 hours of real-time data and unprocessed alarms, using a hash structure to associate tag encoding with data content, achieving a query response time of ≤5ms. An InfluxDB time-series database (10TB storage capacity) stores historical data in partitions according to business dimensions (such as gas extraction, ventilation, etc.), building partition indexes based on tag encoding, supporting fast retrieval of historical data by tag (retrieval latency ≤30ms). Critical security data is retained for 3 years, and regular data is retained for 1 year. A dual backup mechanism of master-slave replication and off-site disaster recovery is configured, ensuring data storage reliability ≥99.99%. (Purpose: To balance data storage efficiency and security, meeting data access needs in different scenarios)

[0047] Step 4: Kafka priority subscription and distribution (Execution entity: Kafka message queue server, each node configured with 32GB memory and 2TB SSD)

[0048] 1. Topic Configuration: Divide into 8 core topics (gas extraction alarm, ventilation real-time data, etc.) according to "business subsystem - data type". Each topic is configured with 3 partitions, corresponding to independent user groups (purpose: to improve parallel processing efficiency). 2. Priority Scheduling: Data is divided into three levels: high priority (alarm data such as gas over-limit and fan failure), medium priority (real-time equipment operation data), and low priority (daily / weekly statistical data). By configuring the message.timestamp.type parameter to "CreateTime", high priority data triggers a priority writing mechanism with a flow delay of ≤50ms (purpose: to ensure rapid response to emergency alarms). 3. Targeted distribution: The downstream "eight centers" subscribe to the corresponding topics through dedicated consumer groups (such as the disaster prevention and control center subscribing to the "gas extraction alarm" and "ventilation alarm" topics). The GeniusLinksGateway gateway is configured with customized distribution rules, supports multi-terminal push on web / mobile terminals, and has a built-in 24-hour local caching and breakpoint resume mechanism (function: to ensure that data is not lost and is delivered accurately).

[0049] Step 5: Full-service collaborative management and feedback (Executing entity: Application layer "eight centers" collaborative management system)

[0050] 1. Specialized Management and Control: Each center implements its own functions based on subscribed data (Disaster Prevention and Control Center provides early warning 15 minutes in advance through gas concentration prediction model; Intelligent Dispatch Center coordinates the scheduling of personnel, machines, and tasks; Operation and Management Center calculates costs and outputs optimization solutions) (Function: Covering all business scenarios); 2. Closed-loop feedback: Feedback the control results (such as equipment failure handling results and energy consumption optimization suggestions) to the perception layer and platform layer, and dynamically adjust the collection frequency, alarm thresholds and governance rules (function: to achieve continuous optimization of the entire process). A coal mine multi-source data aggregation and governance platform, such as Figure 2 As shown, it includes: The perception layer, which has built-in sensors, collects multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data. The edge layer performs edge preprocessing on multi-source heterogeneous data, encapsulates it, and adds a unique device identifier; it also performs edge data cleaning on the encapsulated data to obtain valid data. The transport layer is equipped with a primary and backup dual-ring network, through which effective data is transmitted to the server. At the platform layer, the servers perform data fusion and tagging on the valid data, and store it in layers, distributing the layered data based on priority. At the application layer, data stored in layers is fed back to the full-service collaborative management and feedback center for processing and mobilization.

[0051] Product Solution: 1. Overall Structure: The platform adopts a five-layer progressive architecture: "Perception Layer - Edge Layer - Transmission Layer - Platform Layer - Application Layer". Each layer achieves bidirectional communication via Industrial Ethernet (TCP / IP protocol). Hardware devices and software systems are deeply integrated, forming a closed-loop management system of "Acquisition - Edge Preprocessing and Cleaning - Transmission - Deep Governance - Distribution - Application - Feedback". The edge layer is the core computing power support layer, undertaking the dual tasks of localized data preprocessing and cleaning, reducing data volume from the source and improving overall efficiency.

[0052] 2. Structure and connections of each layer: (1) Sensing layer: The core components are various heterogeneous devices and sensing terminals, including Siemens S7-1500 PLC, Schneider M340 PLC, WinCC / Intouch host computer, as well as gas sensors, wind speed sensors, personnel positioning terminals, vibration sensors, etc. All devices establish wired physical connections with the edge layer gateway through RJ45 network port and RS485 serial port to realize the collection and uploading of raw multi-dimensional data, with a collection coverage of 100%.

[0053] (2) Edge layer: The core component is the GeniusLinksGateway ZL-2025 smart gateway (interface configuration: 8 RJ45 network ports, 4 RS485 serial ports, 2 LC fiber optic interfaces), with a built-in dedicated edge computing module - equipped with an ARM Cortex-A76 quad-core processor (2.4GHz), 4GB DDR4 memory and 16GB eMMC local storage, with a single gateway computing power ≥20GFLOPS, which can support the integrated real-time processing of "edge preprocessing + data cleaning" of 500 data points per second. The gateway integrates preprocessing algorithms (dual-layer filtering for anomalies, fuzzy enhancement, and standardized encapsulation) and data cleaning algorithms (3σ criterion for anomaly removal and linear interpolation completion). It communicates bidirectionally with the perception layer devices via RJ45 Ethernet / RS485 serial ports, receiving raw data and simultaneously completing preprocessing and cleaning, providing feedback on the data processing status (including normal, anomaly removal, and interpolation completion indicators). It connects to the dual-ring network of the transport layer via an LC fiber optic interface, uploading only the cleaned, valid, standardized data, while simultaneously receiving unified processing rule instructions from the platform layer (including filter window parameters, 3σ thresholds, interpolation conditions, etc.) to ensure consistent processing standards across the entire gateway. A 16GB eMMC local storage reserve provides 24 hours of processing logs (including raw data, cleaning results, and rule versions), supporting fault traceability and verification.

[0054] (3) Transmission layer: The ground core switch (Huawei S5735S-L48T4S-A, configured with 48 1000Mbps electrical ports and 4 10G optical ports) and the underground intrinsically safe switch (KTJ127) form a dual-ring redundant network. The ground core switch is connected to the platform layer server through the 10G optical port, and the underground intrinsically safe switch is bidirectionally connected to the edge layer gateway and the ground core switch through single-mode optical fiber. The main and backup dual rings are redundantly backed up through optical fiber links to form a ring topology structure, ensuring low latency and high reliability transmission of effective data after edge layer cleaning.

[0055] (4) Platform layer: It includes 3 GeniusLinks IoT management system servers (cluster deployment, supporting load balancing), 3-node Kafka message queue servers, Redis real-time library servers (master-slave architecture) and InfluxDB time-series library servers (distributed deployment); each server is connected to the core switch of the transmission layer through a 10G optical port and communicates with the application layer terminal through an Ethernet interface. Among them, the Kafka server and the IoT management system server realize data interaction through an internal gigabit LAN, undertake the cleaned data uploaded from the edge layer, and focus on core tasks such as deep data fusion, tagging, and hierarchical storage, without repeating the basic data cleaning.

[0056] (5) Application layer: The “Eight Centers” consist of an industrial control computer (CPU: Intel Core i7-12700K, memory: 16GB, hard disk: 1TB SSD), a 55-inch monitoring screen (resolution 3840×2160), and a mobile APP (supporting Android 10.0+ / iOS 14.0+). The industrial control computer and the monitoring screen are connected via HDMI interface and connected to the platform layer network via RJ45 network port. The mobile APP communicates with the platform layer server via 5G / NB-IoT network to remotely receive alarms and dispatch instructions.

[0057] 3. Functions of each structure: (1) GeniusLinksGateway ZL-2025 Gateway (Edge Layer Core): Relying on the built-in ARM Cortex-A76 computing power module, it integrates 8 types of communication protocol adapter nodes, edge preprocessing algorithms and data cleaning algorithms to realize localized "preprocessing + cleaning" integrated processing of multi-source data. The total processing time for a single data item is ≤10ms (including preprocessing ≤5ms and cleaning ≤5ms), which is far below the 20ms delay limit and has sufficient computing power redundancy. The data acquisition success rate is ≥99.8%, the debugging time for a single device is ≤1h, and it can remove 3%-5% of outliers and 0.8%-1.2% of invalid supplementary data, reducing the transmission volume by 4%-6%, which greatly reduces the load on the transmission layer and the governance pressure on the platform layer.

[0058] (2) Dual-ring redundant switch cluster (transmission layer): The main ring ensures high-speed transmission of core data, and the backup ring achieves fault redundancy. The single-point fault switching time is ≤50ms, ensuring uninterrupted data transmission between the edge layer and the platform layer. With the simplification of edge layer data, the end-to-end transmission delay is reduced from ≤100ms to ≤80ms.

[0059] (3) Platform layer server cluster: It carries the core tasks of deep data integration, tagging, storage and subscription distribution. Since the edge layer has completed basic cleaning, the core business time is shortened by 15%-20%, supporting a peak throughput of 5000 records / s, and the high-priority data flow latency is ≤50ms. (4) "Eight Centers" Terminal (Application Layer): Receives targeted data distribution, realizes full business functions such as production monitoring, disaster early warning, emergency dispatch, and business accounting, and outputs visualized control results and operation instructions; (5) Auxiliary sensing equipment (sensing layer): including gas sensors (measurement range 0~10% CH4, accuracy ±0.01%), wind speed sensors (measurement range 0~20m / s, accuracy ±0.1m / s), etc., to provide high-quality raw data sources for the edge layer.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.

[0061] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for the aggregation and management of multi-source data in coal mines, characterized in that, include: S1, collect multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data; S2, perform edge preprocessing on the multi-source heterogeneous data, encapsulate it, and add a unique device identifier; S3 performs edge data cleaning on the encapsulated data to obtain valid data; S4. Establish a primary and backup dual-ring network, and transmit effective data to the server through the primary and backup dual-ring network; S5, the server performs data fusion and tagging on the valid data, and stores it in layers, and distributes the layered data based on priority; S6, the hierarchically stored data is fed back to the full-service collaborative management and feedback center, and processed and mobilized based on the full-service collaborative management and feedback center.

2. The method for multi-source data aggregation and management in coal mines according to claim 1, characterized in that, The coal mine-related equipment includes heterogeneous devices and a host computer. The heterogeneous devices connect to a Siemens S7-1500 programmable logic controller via a dedicated contrib-s7 communication node built into the gateway, and are compatible with the Profinet industrial Ethernet protocol; they also connect to a Schneider M340 PLC via a contrib-modbus communication node, and are compatible with Modbus RTU / TCP dual protocols, to collect heterogeneous data. The host computer connects to a WinCC host computer via a contrib-opcua node, and is compatible with the OPC unified architecture protocol, to collect production data. The contrib-dde node connects to the Intouch host computer, adapts to the dynamic data exchange protocol, and is compatible with data access from older devices; the multi-source heterogeneous data includes heterogeneous data, older device data, and production data.

3. The method for multi-source data aggregation and management in coal mines according to claim 1, characterized in that, The edge preprocessing includes median filtering and Gaussian filtering. A threshold is set based on the sensor's inherent range. When the data exceeds the threshold, it is directly marked as invalid and discarded. Noise within the range is filtered by first applying median filtering, then applying Gaussian filtering to the processed data, and finally encapsulating the data to obtain the encapsulated data.

4. The coal mine multi-source data aggregation and management method according to claim 1, characterized in that, The edge data cleaning includes: removing abnormal data based on the statistical characteristics of normal distribution, performing double verification based on a secondary cleaning mechanism, and performing linear interpolation to complete the verified data to obtain valid data.

5. The method for multi-source data aggregation and management in coal mines according to claim 1, characterized in that, The primary and backup dual-ring network includes: a primary and backup dual-ring network built through a ground core switch and an underground intrinsically safe switch; the primary ring uses 4-core single-mode fiber and the backup ring uses 2-core single-mode fiber; the primary ring transmits valid data at a rate of 1000Mbps and the backup ring is hot-standby at a rate of 100Mbps; when a single point of failure is detected, the system automatically triggers the primary and backup ring switching, with a switching response time of ≤50ms; link connectivity is automatically detected every 30s, and a link failure alarm is sent through the platform layer system when an anomaly occurs.

6. The method for multi-source data aggregation and management in coal mines according to claim 1, characterized in that, The data fusion and tagging specifically includes: constructing a two-level fusion architecture of feature layer and decision layer to adapt to the heterogeneous data characteristics of coal mines; feature layer fusion adopts a weighted feature projection algorithm to extract time-domain and frequency-domain features from multi-dimensional time-series data of the same equipment, and assigns weights to each feature through entropy weight method to map high-dimensional features to a unified feature space, eliminating data redundancy; decision layer fusion is based on DS evidence theory, integrates data related to multiple equipment, constructs trust functions and likelihood functions, outputs fusion results, and synchronously completes data alignment during the fusion process, and achieves time synchronization based on NTP / PTP protocol; character encoding is used to achieve unique identification of data in all dimensions, and the tag fields are automatically adjusted based on changes in equipment operating status, with bidirectional synchronization between the edge layer and platform layer; based on the coal mine safety production knowledge graph, a tag semantic association model is constructed to achieve cross-business tag interoperability, and automatic mapping rules between tags and Topic subscriptions are designed to automatically distribute data to the corresponding Kafka Topic partition according to the parameter type and priority fields in the tag.

7. A coal mine multi-source data aggregation and governance platform, employing the coal mine multi-source data aggregation and governance method according to any one of claims 1-6, characterized in that, include: The perception layer, which has built-in sensors, collects multi-source data from coal mine-related equipment to obtain multi-source heterogeneous data. The edge layer performs edge preprocessing on the multi-source heterogeneous data, encapsulates it, and adds a unique device identifier. Perform edge data cleaning on the encapsulated data to obtain valid data; The transport layer is equipped with a primary and backup dual-ring network, through which effective data is transmitted to the server. At the platform layer, the servers perform data fusion and tagging on the valid data, store it in layers, and distribute the layered data based on priority. At the application layer, the hierarchically stored data is fed back to the full-service collaborative management and feedback center, which then processes and mobilizes the data.

Images