Data management method and system for coal mine, and storage medium

Abstract

The invention relates to the technical field of coal mine informatization and data management, in particular to a data governance method and system for a coal mine and a storage medium, and provides a layered architecture from edge node preprocessing to central platform deep fusion in order to solve the problem of coal mine multi-source heterogeneous data governance. The edge nodes realize real-time anomaly judgment and output clustering data through an auto-encoder, a generative adversarial network and deep reinforcement learning; the central platform generates global collaborative model parameters and logic semantic data by using federated learning and a self-attention mechanism, then constructs a coal mine knowledge graph through unified semantic mapping and a graph convolutional network, and synchronously updates the digital twinborn platform with real-time verification data to generate real-time digital twinborn data. And finally, outputting an early warning decision under the action of a long short-term memory network and a self-attention mechanism, and performing traceability management in combination with a block chain technology, thereby realizing efficient and safe coal mine data management.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine informatization and data management, and in particular to a data governance method and system for coal mines, and a storage medium. Background Art

[0002] The data governance of coal mines is of great significance for ensuring mine safety production, optimizing dispatching management, and improving the intelligent level. The coal mine industry involves multi-source heterogeneous data such as geological exploration, equipment monitoring, environmental monitoring, and personnel positioning. Traditional methods are usually limited to single-dimensional data collection and isolated storage (Chinese invention patent, publication number: CN118820910B, title: A Heterogeneous Network Security Big Data Governance Method and System), lacking the ability to uniformly standardize and deeply integrate global data. Existing technologies mostly use serial processing and fixed rules to clean and classify data. Due to the inability to dynamically identify implicit complex associations, they often perform inadequately when dealing with emergencies or requiring real-time decisions. In addition, these traditional solutions are difficult to provide strong support for the automatic anomaly detection and online model update of large-scale multi-source data. The main reasons are the deficiencies in heterogeneous data preprocessing and fusion technologies, coupled with the lack of intelligent algorithms such as efficient federated learning and self-attention mechanisms, resulting in low overall governance efficiency and high misjudgment rates, and it is difficult to meet the stringent requirements of the coal mine industry for real-time and accuracy. Summary of the Invention

[0003] Aiming at the many problems existing in the above-mentioned prior art, the present invention provides a data governance method and system for coal mines, and a storage medium. The present invention relies on distributed edge node intelligent processing and central platform fusion algorithms, and through technologies such as autoencoders, generative adversarial networks, deep reinforcement learning, and federated learning, dynamically preprocesses and detects anomalies in coal mine multi-source data, and then combines the self-attention mechanism and graph convolutional network to construct a coal mine knowledge graph and update the digital twin model, and finally realizes real-time prediction and early warning and safety decision-making. The present invention effectively integrates data collection, intelligent recognition, and global fusion, and significantly improves the accuracy of data processing and decision-making efficiency.

[0004] A data governance method for coal mines includes the following steps:

[0005] Collect original data of geological exploration, equipment sensors, production scheduling, environmental monitoring, and personnel positioning, and convert them into standardized data, and preprocess to obtain real-time verification data;

[0006] On the edge node, use an autoencoder to extract features from the real-time verification data and calculate the reconstruction error to form reconstruction error data. At the same time, use a generative adversarial network to generate an enhanced data set, and use deep reinforcement learning to perform anomaly determination on fixed-time window data to output edge processing data, and then obtain clustering data through self-organizing clustering;

[0007] The edge processing data and clustering data are uploaded to the central platform. The central platform uses federated learning to fuse the parameters of each edge model, and combines the coal mine symbol rules with the self-attention mechanism to generate logical semantic data and global collaborative model parameters. The data is then mapped into unified semantic data and a coal mine knowledge graph is constructed. The knowledge graph and real-time verification data are synchronously updated on the digital twin platform to generate real-time digital twin data.

[0008] Based on the real-time digital twin data, global collaborative model parameters and coal mine knowledge graph, a long short-term memory network combined with a graph convolutional network and a self-attention mechanism is used to implement time series prediction and anomaly detection for key indicators to generate early warning decision data, and collect field feedback to close the loop and update the global collaborative model parameters. At the same time, the data and decision-making process are recorded to achieve data traceability and secure sharing.

[0009] Preferably, the step of extracting features from the real-time verification data using an autoencoder on the edge node and calculating the reconstruction error to form reconstruction error data includes:

[0010] A convolutional variational autoencoder is used to perform encoding and decoding operations on real-time verification data, and the abnormality of each data is determined by calculating the mean square error.

[0011] Preferably, the step of using a generative adversarial network on the edge node to generate an enhanced data set includes:

[0012] Synthetic samples are generated by inputting a latent variable vector of fixed length into the generator, and the synthetic samples are combined with real-time verification data to form an enhanced data set to expand the data distribution coverage.

[0013] Preferably, the step of using deep reinforcement learning on the edge node to perform anomaly determination on fixed time window data includes:

[0014] Real-time verification data, reconstructed error data and enhanced data sets are combined as state inputs, and the deep reinforcement learning model outputs the confidence of abnormal states and divides the data categories according to the confidence to generate edge processing data.

[0015] Preferably, the edge nodes perform global search optimization on the edge processing data through self-organizing clustering to determine the cluster center and output the cluster data. The self-organizing clustering uses quantum annealing and quantum tunneling principles to adaptively adjust the cluster center position.

[0016] Preferably, after uploading the edge processing data and clustering data to the central platform, the central platform first sets the weighting coefficients according to the data volume and the mean reconstruction error of each edge node, then fuses the model parameters of each edge node through the federated averaging algorithm, and combines the preset symbol rules in the coal mine field to perform self-attention calculation on the input rules and the neural network embedding vectors to generate logical semantic data and global collaborative model parameters. Subsequently, the logical semantic data is converted into unified semantic data according to the fixed mapping table, and the graph convolutional network is used to construct the graph structure and perform association reasoning on the unified semantic data, and finally a coal mine knowledge graph is formed and synchronized and updated with the real-time verification data at preset time intervals on the digital twin platform to generate real-time digital twin data.

[0017] Preferably, the step of performing time series prediction and anomaly detection on key indicators by using a long short-term memory network combined with a graph convolutional network and a self-attention mechanism based on the real-time digital twin data, the global collaborative model parameters, and the coal mine knowledge graph to generate early warning decision data includes:

[0018] Using the long short-term memory network to recursively model the key indicator sequence in the real-time digital twin data to generate prediction trend data, performing anomaly detection and analysis on the association relationships in the coal mine knowledge graph through the graph convolutional network in combination with the global collaborative model parameters, and comprehensively integrating the prediction trend data and historical data under the action of the self-attention mechanism, and finally fusing the prediction trend data and the anomaly detection results to output early warning decision data.

[0019] Preferably, after collecting the feedback of the on-site operator on the early warning decision data to form feedback data, the closed-loop feedback mechanism retrains and updates the global collaborative model parameters according to the misjudgment information and key samples marked in the feedback data, and submits the standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, as well as the feedback data and the decision-making process to the blockchain network for hash calculation and distributed storage through the blockchain technology based on Hyperledger Fabric, so as to generate data traceability records and secure shared data.

[0020] A coal mine data governance system for implementing the method for coal mine data governance, the system includes:

[0021] A data collection module, configured to collect geological exploration raw data, equipment sensor raw data, production scheduling raw data, environmental monitoring raw data, and personnel positioning raw data, and convert the foregoing raw data into standardized data and then preprocess it to generate real-time verification data;

[0022] An edge processing module is set on the edge node and is used to extract the features of real-time verification data by using an autoencoder and calculate the mean square error to form reconstructed error data. The generator in the generative adversarial network is used to generate synthetic samples with a fixed-dimensional latent variable vector as the input, so that the synthetic samples are combined with the real-time verification data to form an enhanced data set. The deep reinforcement learning model is used to perform anomaly determination on the data within a fixed time window and output edge processing data. Subsequently, the self-organizing clustering algorithm is used to globally optimize the clustering centers in a manner based on the principles of quantum annealing and quantum tunneling to generate clustering data;

[0023] A central processing module is set on the central platform. After uploading the edge processing data and clustering data, it uses the federated averaging algorithm to perform weighted fusion on the model parameters uploaded by each edge node according to the node data volume, and combines the preset symbol rules in the coal mine field to generate logical semantic data and global collaborative model parameters through the self-attention mechanism. The logical semantic data is converted into unified semantic data according to a fixed mapping relationship, and a coal mine knowledge graph is constructed in the graph database by using a graph convolutional network. The coal mine knowledge graph and the real-time verification data are synchronously updated on the digital twin platform at a set time interval to generate real-time digital twin data;

[0024] A prediction feedback module is set on the digital twin platform and is used to perform time series prediction and anomaly detection on key indicators by using a long short-term memory network combined with a graph convolutional network and a self-attention mechanism based on real-time digital twin data, global collaborative model parameters, and a coal mine knowledge graph to generate early warning decision data, and collect the feedback of on-site operators on the early warning decision data to form feedback data, use a closed-loop feedback mechanism to update the global collaborative model parameters, and at the same time record standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data and feedback data, and the decision-making process based on the Hyperledger Fabric blockchain technology to generate data traceability records and secure shared data.

[0025] A storage medium stores a computer program thereon, and when the program is executed by a processor, the steps of the data governance method for coal mines are implemented.

[0026] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0027] Through distributed edge node self-learning and central platform federated fusion technical means, real-time collection, deep preprocessing, and dynamic model update of large-scale coal mine data are realized;

[0028] By combining the self-attention mechanism with symbol rules, the domain safety regulations are fully integrated with the neural network model, enabling the system to efficiently identify potential threats and output early warning decisions in a multi-source heterogeneous data environment;

[0029] Through the linkage of the graph convolutional network and the digital twin platform, the semantic association and visual presentation of multi-dimensional data are realized, effectively enhancing the traceability and management transparency of coal mine data;

[0030] Through blockchain technology, the tamper-proof recording and secure sharing of key data and decision-making processes are realized. The combination of these core technologies solves the problems in the comparison scheme such as dealing with massive heterogeneous data, high-dimensional anomaly detection, and real-time adaptive update of the model, laying a solid data foundation for coal mine safety governance and intelligent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a schematic flowchart of the method of the present invention;

[0032] Figure 2 is a schematic diagram of the integration of the central platform in the present invention;

[0033] Figure 3 is a schematic diagram of the digital twin integration in the present invention;

[0034] Figure 4 is a block diagram of the structure of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth in order to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments can be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0036] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0037] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those of ordinary skill in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0038] As Figure 1 shown, a data governance method for coal mines includes the following steps:

[0039] Collect the original data from geological exploration, equipment sensors, production scheduling, environmental monitoring, and personnel positioning, and convert it into standardized data, and preprocess it to obtain real-time verification data;

[0040] In view of the problems of multi-source heterogeneity, inconsistent formats, and uneven quality of data in the field of coal mine data governance, the present invention proposes a complete process from original data collection to preprocessing to obtain real-time verification data. This process first collects the original data from geological exploration, equipment sensors, production scheduling, environmental monitoring, and personnel positioning. These data sources cover all key links in coal mine operations and reflect the geological structure of the mining area, equipment status, production plan, environmental parameters, and personnel location information. Subsequently, the collected original data is converted into a unified standardized data format to solve the inconsistency problems caused by differences in collection equipment, data formats, communication protocols, etc. among various data sources. Then, through preprocessing techniques, format verification, numerical range verification, and timestamp verification are performed on the standardized data to obtain real-time verification data, ensuring the data quality and accuracy in the subsequent data processing process. Overall, this step constitutes the basic link of data quality control in the coal mine data governance system, provides reliable data support for subsequent data analysis at all levels, model training, and decision-making support, and lays a solid data foundation for the system to achieve real-time monitoring and closed-loop feedback.

[0041] On the edge node, use an autoencoder to extract features from the real-time verification data and calculate the reconstruction error to form reconstruction error data. At the same time, use a generative adversarial network to generate an enhanced data set, and use deep reinforcement learning to perform anomaly determination on the fixed-time window data and output edge processing data, and then obtain clustering data through self-organizing clustering;

[0042] In view of the problems existing in the process of coal mine data governance, such as multi-source heterogeneity of data, inconsistent data formats, and uneven data quality, a complete process for intelligently processing real-time verified data on edge nodes is proposed. This process first collects original geological exploration data, original equipment sensor data, original production scheduling data, original environmental monitoring data, and original personnel positioning data, and uniformly converts the foregoing original data into standardized data through a pre-designed data conversion module. Subsequently, preprocessing techniques are used to verify the format, numerical range, and timestamp of the standardized data, so as to obtain verified data with high consistency and real-time performance. Next, on the edge node, an autoencoder is used to extract features from the real-time verified data, and the mean square error is calculated to generate reconstruction error data reflecting the data reconstruction quality. At the same time, a generative adversarial network is used to generate synthetic samples, and the synthetic samples are combined with the real-time verified data to form an enhanced data set to expand the data distribution. Then, a deep reinforcement learning model is used to determine the abnormal state of the data within a fixed time window and output edge-processed data. Finally, a self-organizing clustering algorithm is used to perform global search and optimization on the edge-processed data to determine the optimal clustering center, thereby generating clustering data.

[0043] This core process effectively integrates a variety of artificial intelligence algorithms and advanced computing technologies, not only realizing high-quality preprocessing of coal mine data, but also providing a reliable data basis and model input for subsequent central platform data fusion, knowledge graph construction, digital twin model update, and early warning decision-making. Finally, data traceability and secure sharing are realized, ensuring coal mine safety production and intelligent management.

[0044] Preferably, the step of using an autoencoder on the edge node to extract features from the real-time verified data and calculate the reconstruction error to form reconstruction error data includes:

[0045] Performing encoding and decoding operations on the real-time verified data using a convolutional variational autoencoder, and determining the abnormal degree of each data by calculating the mean square error.

[0046] On the edge node, in the step of using an autoencoder to extract features from the real-time verified data and calculate the mean square error, a convolutional variational autoencoder is used as the core technology. This variational autoencoder includes two parts: an encoder and a decoder. The encoder compresses the input real-time verified data (for example, gas concentration, temperature and humidity, and other key index data collected by each sensor) into a latent variable vector of a fixed length through multiple convolutional operations; the decoder takes this latent variable vector as the input and reconstructs an approximation of the original data. Specifically, let the input data vector be x = (x1, x2,..., x n ), and the encoder outputs the latent variable vector z = (z1, z2,..., z m), where \(n\) represents the dimension of the data, \(m\) is the dimension of the hidden layer, and \(m < n\); the decoder decodes \(z\) to generate the reconstructed data \(x'=(x'_1,x'_2,\cdots,x' n ). The reconstruction error is calculated using the mean squared error (MSE), and its calculation formula is:

[0047]

[0048] where \(x_1\) represents the \(i\)-th component in the original data, and \(x' i represents the corresponding component in the reconstructed data. The generation of the reconstruction error data can reflect the deviation degree between each data record and its reconstruction result. When the reconstruction error exceeds the preset threshold, it indicates that the data record is abnormal. In this step of practical application, local feature extraction is achieved by adopting convolution operations, and the variational autoencoder is used to reduce the data dimension while retaining the main features of the data, providing an effective low-dimensional representation for subsequent anomaly detection.

[0049] For example, assume that the real-time calibration data of a certain sensor is a sequence of temperature values. The encoder compresses this sequence into a 10-dimensional vector, then the decoder reconstructs the original sequence, and the mean squared error is calculated. If the mean squared error value is large, then the temperature data may have abnormal fluctuations, and this information can be transmitted to the deep reinforcement learning module as the basis for anomaly determination. By adopting this method, not only the data redundancy is reduced, but also the sensitivity and accuracy of anomaly detection are improved, thus playing a key role in coal mine safety monitoring.

[0050] Preferably, the steps of using a generative adversarial network to generate an enhanced data set on the edge node include:

[0051] Generating synthetic samples by inputting a latent variable vector of a fixed length into the generator, and combining the synthetic samples with the real-time calibration data to form an enhanced data set for expanding the coverage of the data distribution.

[0052] On the edge node, to expand the coverage of the data distribution and improve the robustness of anomaly determination, the present invention adopts generative adversarial network technology to generate an enhanced data set. Specifically, the generative adversarial network consists of a generator and a discriminator. The generator takes a latent variable vector of a fixed length as input and generates synthetic samples through several fully connected layers and non-linear activation functions. Let the latent variable vector be \(z\), and the generator function \(G(z)\) outputs the synthetic data \(x g . The training objective of the generator is to make the synthetic data as close as possible to the real real-time calibration data in distribution, while the discriminator distinguishes whether the input data is real data or synthetic data. The loss function of the generative adversarial network generally uses cross-entropy loss, and through adversarial training, the generator continuously improves the quality of the generated samples. Finally, the synthetic data output by the generator is fused with the original real-time calibration data to form an enhanced data set.

[0053] This enhanced data set not only supplements the sample-scarce regions but also captures the complex patterns hidden in the data, improving the accuracy of the deep reinforcement learning module in determining abnormal states. For example, for the gas concentration data recorded by a certain sensor, the distribution may have a scarcity of extreme values. Through the generative adversarial network, multiple synthetic samples can be generated to cover the low-density regions of the data distribution, thus making the determination of abnormal states more comprehensive in the subsequent model training process. When the present invention is actually implemented, a combination of a random noise vector and a conditional generator can be adopted to achieve more diverse synthetic sample generation. The specific parameters can be set according to experimental data, such as setting the latent variable vector dimension to 16 and using the ReLU activation function, etc., to ensure the stability and reproducibility of the generation process.

[0054] Preferably, the steps of using deep reinforcement learning to perform abnormal determination on the fixed-time window data on the edge node include:

[0055] Taking the real-time verification data, the reconstruction error data, and the enhanced data set as the state input, the deep reinforcement learning model outputs the abnormal state confidence level, and divides the data categories according to the confidence level to generate the edge processing data.

[0056] On the edge node, using deep reinforcement learning to perform abnormal state determination on the fixed-time window data is the key link in improving the sensitivity and response speed of abnormal detection in the present invention. This step takes the real-time verification data, the reconstruction error data, and the enhanced data set as the state input to form the state vector S. The deep reinforcement learning model adopts a multi-layer fully connected network. By processing the state S, it outputs the confidence level value and the determination result of the abnormal state.

[0057] Specifically, let the state vector S be composed of the real-time verification data X, the reconstruction error data E, and the enhanced data set A. The model is mapped to the output layer through a series of hidden layers, and outputs the abnormal confidence level Q(S). This confidence level value is used to divide the data categories. For example, setting a threshold τ (e.g., τ = 0.6), when Q(S) ≥ τ, it is determined as an abnormal state, otherwise it is a normal state. In practical applications, the fixed-time window is usually set to 30 seconds to ensure that the model can capture the dynamic changes of the data in real time. For example, in a coal mine environment, if the sensor data of a certain device shows a sharp fluctuation within 30 consecutive seconds, through the determination of the deep reinforcement learning model, this abnormal situation can be identified in time, and the result is output as the edge processing data. This process not only depends on the data features learned by the model during the training stage but also continuously optimizes the model strategy through online feedback, making the determination of abnormal states more accurate. In this way, the present invention has obvious advantages in improving the detection efficiency and accuracy of abnormal events in coal mine safety monitoring.

[0058] Preferably, on the edge node, the self-organizing clustering is used to globally search and optimize the edge processing data to determine the clustering center and output the clustering data. The self-organizing clustering adaptively adjusts the position of the clustering center by using the principles of quantum annealing and quantum tunneling.

[0059] On the edge node, after the edge processing data is judged for anomalies, in order to further optimize the data structure and subdivide the anomaly patterns, the present invention uses the self-organizing clustering algorithm to generate clustering data. This clustering algorithm uses the principles of quantum annealing and quantum tunneling to globally search and optimize the edge processing data and automatically determine the optimal clustering center.

[0060] In specific implementation, first, the edge processing data is mapped to a low-dimensional space, and then the quantum annealing algorithm is used to simulate the annealing process of the system energy function. By continuously reducing the system temperature, the global lowest energy state is searched to determine the position of the clustering center. The principle of quantum tunneling is used to jump out of local minima in this process to ensure the global optimality of the clustering result. Mathematically, assuming the edge processing data is X = {x1, x2, …, x N}, the objective function can be expressed as:

[0061]

[0062] where μ j is the clustering center. The quantum annealing algorithm gradually searches for the optimal μ k (α is the attenuation factor, k is the number of iterations) by setting the initial temperature T0 and the temperature decrease function T = T0·α j .

[0063] In actual operation, the present invention can perform fine-grained classification on the equipment status, environmental data, etc. collected in the coal mine, providing more accurate data grouping for subsequent data analysis. For example, if there are multiple different anomaly patterns in a certain equipment status data, after using self-organizing clustering, these data can be classified into different clusters respectively, providing a basis for anomaly cause analysis and early warning decision-making. This clustering method not only improves the robustness of anomaly detection but also enables the system to dynamically adapt to the changes in the coal mine site environment and realize the adaptive update of the clustering data.

[0064] As Figures 2 - 3 shown, the edge processing data and the clustering data are uploaded to the central platform. The central platform uses federated learning to fuse the parameters of each edge model and generates logical semantic data and global collaborative model parameters through the self-attention mechanism in combination with the coal mine symbol rules, and then maps them into unified semantic data and constructs a coal mine knowledge graph. The knowledge graph and the real-time verification data are synchronously updated on the digital twin platform to generate real-time digital twin data;

[0065] The present invention proposes a complete technical process for the key link of central data fusion in the process of coal mine data governance. After the data is processed at the edge nodes, the edge-processed data and clustering data are uploaded to the central platform, and then through a series of operations of fusion, mapping, and knowledge construction, real-time digital twin data for real-time monitoring and decision support is generated.

[0066] Specifically, the central platform first uses federated learning technology to fuse the model parameters uploaded by each edge node, and combines the preset coal mine symbol rules to calculate the input rules and neural network embedding vectors through the self-attention mechanism to generate logical semantic data and global collaborative model parameters; then, according to the fixed mapping relationship, the logical semantic data is converted into unified semantic data, and a coal mine knowledge graph is constructed in the graph database using the graph convolutional network; finally, the constructed coal mine knowledge graph and real-time verification data are synchronously updated in the digital twin platform at a set time interval to form real-time digital twin data for subsequent early warning, decision-making, and safety monitoring.

[0067] This process aims to combine the data processed by the edge nodes with the deep data fusion technology of the central platform, and through precise model fusion, rule-driven self-attention calculation, and graph neural network to construct a knowledge graph, realize the real-time integration and dynamic update of key index information of coal mines, so as to effectively support coal mine safety production and intelligent decision-making.

[0068] Preferably, after the edge-processed data and clustering data are uploaded to the central platform, the central platform first sets a weighted coefficient according to the data volume and mean reconstruction error of each edge node, then fuses the model parameters of each edge node through the federated averaging algorithm, and combines the preset coal mine domain symbol rules to perform self-attention calculation on the input rules and neural network embedding vectors to generate logical semantic data and global collaborative model parameters. Subsequently, the logical semantic data is converted into unified semantic data according to the fixed mapping table, and the graph convolutional network is used to construct a graph structure and perform association reasoning on the unified semantic data. Finally, a coal mine knowledge graph is formed and synchronously updated with the real-time verification data in the digital twin platform at a preset time interval to generate real-time digital twin data.

[0069] After the edge-processed data and clustering data are uploaded to the central platform, the central platform first performs weighted processing according to the uploaded data of each edge node. To ensure that the uploaded model parameters can reflect the data volume and quality of each edge node during the fusion process, the central platform sets a weighted coefficient, which depends on the data volume collected by each node and the mean reconstruction error. Suppose the model parameter uploaded by the i-th edge node is θ i , and the weighted coefficient of the node is denoted as w i ; the global model parameter θ after weighted fusion is calculated by the federated averaging algorithm, and its formula is:

[0070]

[0071] Among them, N is the number of edge nodes; w i is usually determined according to the mean reconstruction error μ of this node i and the data volume n i , for example, it can be set (where ∈ is a small positive number to prevent division by zero), to ensure that nodes with large data volume and low reconstruction error contribute more to the global model. After weighted fusion, the central platform will obtain the global model parameters reflecting the overall data characteristics of each edge node. Subsequently, to further extract the deep semantic information in the coal mine data, the central platform performs self-attention mechanism calculation in combination with the preset coal mine domain symbol rules. The basic calculation formula of the self-attention mechanism is:

[0072]

[0073] Among them, Q (query matrix), K (key matrix) and V (value matrix) are respectively generated by the neural network embedding vectors, and d k represents the dimension of the key vector. In the present invention, the preset coal mine symbol rules will, together with the embedding representation of the model parameters uploaded by the edge nodes, form the query, key and value, and then realize the joint calculation of the rules and statistical features within the self-attention layer. In this way, the central platform generates logical semantic data, which not only contains the statistical characteristics of the model parameters of each edge node, but also incorporates the domain knowledge expressed by the coal mine symbol rules, and then generates the global collaborative model parameters. In practical applications, the central platform standardizes this calculation process through a fixed mapping table, so that the logical semantic data is converted into unified semantic data.

[0074] For example, if the coal mine symbol rule stipulates that "when the gas concentration exceeds a certain threshold, it indicates a potential safety hazard", then in the self-attention mechanism calculation, the relevant data vectors will obtain higher attention weights, so that the safety hazard information is reflected in the logical semantic data. This process not only ensures the robustness of data fusion by using federated learning, but also fully combines the domain rules through the self-attention mechanism, realizes the deep expression of data semantics, and thus provides high-quality input for the subsequent construction of the knowledge graph.

[0075] After the logical semantic data is generated, the central platform converts the logical semantic data into unified semantic data according to a preset fixed mapping table. The unified semantic data refers to the standardized expression of the semantic information of the data of each edge node, so that the data from different edges are consistent in fields, units, and semantic tags. This conversion process requires that the mapping table clearly stipulates the unified semantic data format corresponding to each logical semantic data element. For example, the vector reflecting the gas concentration information in the logical semantic data will be mapped to the "gas concentration data" field in the unified semantic data, and its value is uniformly expressed as a value between 0 and 1 after normalization. After the conversion of the unified semantic data is completed, the central platform uses a graph convolutional network to construct a coal mine knowledge graph in the graph database. The graph convolutional network can iteratively update the node representation according to the adjacent relationship between nodes, and the basic formula is:

[0076]

[0077] where H (l) is the node feature matrix of the l-th layer, is the adjacency matrix with self-connection, is the degree matrix, W (l) is the learnable weight matrix, and σ is the activation function. In the present invention, the unified semantic data is input as the initial node feature of the graph convolutional network, and the weights between edges are set according to the spatio-temporal similarity and logical correlation between data.

[0078] Through multi-layer graph convolutional operations, the system gradually learns the deep correlation structure between the data of each coal mine, and finally forms a coal mine knowledge graph. The constructed coal mine knowledge graph not only contains the direct similarity between data records, but also can reveal the indirect relationship between data through graph reasoning. For example, during the coal mine production process, if there is an obvious correlation between the equipment status data and the environmental monitoring data in terms of time and space, the graph convolutional network can capture this relationship and form corresponding edges in the knowledge graph, and then provide a more detailed description of the on-site status in the subsequent digital twin platform.

[0079] The effects of the unified semantic data conversion and knowledge graph construction are to achieve the semantic standardization of multi-source data, promote the effective integration of cross-system data, and provide a solid data semantic foundation for the generation of real-time digital twin data. The central platform saves the constructed coal mine knowledge graph in the graph database, and synchronizes and updates it with the real-time verification data in the digital twin platform at a preset time interval (for example, every 10 minutes) to realize the dynamic update of the knowledge graph and the real-time reflection of on-site data, so as to form accurate real-time digital twin data.

[0080] The digital twin platform plays a key role after the central platform constructs the knowledge graph. Its purpose is to transform theoretical data into an intuitive virtual model that reflects the dynamic changes of key indicators at the coal mine site in real time. The digital twin platform receives real-time digital twin data from the central platform. This data is generated by updating the unified semantic data and the coal mine knowledge graph at regular intervals.

[0081] The digital twin platform uses a high-performance simulation engine to automatically convert the structural information, spatiotemporal relationships, and anomaly detection results described in the coal mine knowledge graph into virtual model parameters through a data interface. For example, data such as the operating status of each device in the coal mine, gas concentration, and ambient temperature are all displayed in 3D on the digital twin platform, allowing operators to intuitively view data changes in each area. The update mechanism of the digital twin platform requires the system to automatically obtain the latest real-time verification data and knowledge graph data at preset intervals (such as 10 minutes), and update the virtual model after data fusion. The core advantage of the digital twin platform is that it realizes the real-time linkage between field data and virtual models, thereby providing an intuitive basis for coal mine safety monitoring and early warning decision-making.

[0082] For example, if the gas concentration data in a certain area shows an anomaly after preprocessing, it will be highlighted on the digital twin platform with color or other visual signals, prompting operators to take timely action. This update process involves technical links such as automatic calling of data interfaces, data format conversion, and real-time rendering. Its implementation relies on standardized data transmission protocols and efficient simulation algorithms. Overall, the digital twin platform not only improves the intuitiveness and operability of coal mine data management, but also ensures the timeliness and accuracy of on-site decision-making through real-time updates, providing strong data support for coal mine safety management.

[0083] Based on the real-time digital twin data, global collaborative model parameters and coal mine knowledge graph, a long short-term memory network combined with a graph convolutional network and a self-attention mechanism is used to implement time series prediction and anomaly detection for key indicators to generate early warning decision data, and collect field feedback to close the loop and update the global collaborative model parameters. At the same time, the data and decision-making process are recorded to achieve data traceability and secure sharing.

[0084] In response to the problems of multi-source data heterogeneity, inconsistent data formats, and low data quality in the coal mine data governance process, this invention proposes a fusion technology solution that focuses on real-time digital twin data, global collaborative model parameters, and coal mine knowledge graphs.

[0085] First, the present invention collects a variety of original data from the site. After preprocessing to obtain real-time verification data, advanced algorithms such as autoencoders, generative adversarial networks, deep reinforcement learning, and self-organizing clustering are used on edge nodes to deeply process the data, generating edge processing data and clustering data. Subsequently, these data are uploaded to the central platform. The central platform uses federated learning to perform weighted fusion on the model parameters uploaded by the edge nodes, and combines the symbol rules in the coal mine field. Through the self-attention mechanism, logical semantic data and global collaborative model parameters are calculated and generated, and then converted into unified semantic data according to the fixed mapping relationship. The graph convolutional network is used to construct a coal mine knowledge graph in the graph database.

[0086] The constructed coal mine knowledge graph and the real-time verification data are synchronously updated on the digital twin platform at preset time intervals, thereby generating real-time digital twin data. Based on these real-time digital twin data, global collaborative model parameters, and the coal mine knowledge graph, the system further uses a long short-term memory network combined with a graph convolutional network and a self-attention mechanism to perform time series prediction and anomaly detection on key indicators, thereby generating early warning decision data.

[0087] Meanwhile, the feedback of the on-site operators on the early warning decision data is used through the closed-loop feedback mechanism to update the global collaborative model parameters, and the entire data and decision-making process is recorded using the Hyperledger Fabric blockchain technology to achieve data traceability and secure sharing. The overall solution effectively realizes the improvement of data quality, the real-time update of model parameters, and the closed-loop management of decision support through hierarchical fusion of edge intelligent processing and central deep fusion technology, providing reliable data and technical support for coal mine safety production and intelligent management.

[0088] Preferably, the step of using a long short-term memory network combined with a graph convolutional network and a self-attention mechanism to perform time series prediction and anomaly detection on key indicators to generate early warning decision data based on the real-time digital twin data, global collaborative model parameters, and the coal mine knowledge graph includes:

[0089] Using the long short-term memory network to recursively model the key indicator sequence in the real-time digital twin data to generate predicted trend data, combining the global collaborative model parameters to perform anomaly detection and analysis on the association relationships in the coal mine knowledge graph through the graph convolutional network, and comprehensively combining the predicted trend data and historical data under the action of the self-attention mechanism, and finally fusing the predicted trend data and the anomaly detection results to output early warning decision data.

[0090] In the central platform, by taking real-time digital twin data, global collaborative model parameters, and the coal mine knowledge graph as inputs and using a Long Short-Term Memory (LSTM) network to perform time series prediction on key indicators, it is an important means for the present invention to achieve coal mine status prediction. First, the LSTM network effectively captures long-term dependencies in the data using its gating structure, recursively models the key indicator sequence in the real-time digital twin data, and generates prediction trend data. Suppose a key indicator sequence in the real-time digital twin data is The LSTM dynamically adjusts the sequence through the forget gate, input gate, and output gate, and its state update process can be described as:

[0091] f t = σ(W f x t + U f h t-1 + b f )

[0092] i t = σ(W i x t + U i h t-1 + b i )

[0093]

[0094] o t = σ(W o x t + U o h t-1 + b o )

[0095] h t = o t ⊙ tanh(c t )

[0096] where, x t represents the input data at time t, h t is the hidden state, c t is the cell state, W f , W i , W c , W o are the input weight matrices, U f , U i , U c , U o are the hidden state weight matrices, b f , b i , b c , b ois the bias term, σ represents the Sigmoid function, and ⊙ represents element-wise multiplication. Using the above calculation process, the long short-term memory network can output predicted trend data, which reflects the change trend of key indicators in the future time period and provides a warning basis for coal mine on-site safety monitoring.

[0097] Meanwhile, the central platform uses the correlation relationship between the global collaborative model parameters and the coal mine knowledge graph, and adopts a Graph Convolutional Network (GCN) to perform anomaly detection on the data. In the graph convolutional network, the coal mine knowledge graph is constructed as a graph G=(V, E), where the node set V represents each data record, and the edge set E represents the spatio-temporal and logical associations between the data. The propagation formula of the graph convolutional network is:

[0098]

[0099] where H (l) is the node feature matrix of the l-th layer, is the adjacency matrix with self-connection, is the node degree matrix, and W (l) is the learnable weight matrix, and σ is the activation function. By iteratively updating the node features in the coal mine knowledge graph, the graph convolutional network can extract abnormal patterns and generate anomaly prediction data. Under the action of the self-attention mechanism, by calculating the relationship between the query matrix Q, the key matrix K, and the value matrix V:

[0100]

[0101] The system synthesizes the predicted trend data and historical data to further correct the abnormal state. The self-attention mechanism can identify the dependence relationships between key data points and enhance the model's response ability to abnormal changes. By fusing the predicted trend data generated by the long short-term memory network and the anomaly prediction data detected by the graph convolutional network, the central platform finally outputs warning decision data, which indicates the risk level of each region or device in numerical form and is accompanied by specific warning suggestions.

[0102] For example, in the prediction of the coal mine gas concentration index, if the LSTM predicts that the future gas concentration tends to rise, and the graph convolutional network detects data anomalies in the relevant areas of the coal mine knowledge graph, the system will output a high-risk warning and suggest taking corresponding safety measures. Overall, through clear mathematical formulas and model designs, this fusion process not only ensures the accuracy of prediction but also improves the sensitivity of anomaly detection, providing multi-level and dynamically updated decision support for coal mine safety management.

[0103] Preferably, after the on-site operator's feedback on the early warning decision data is collected to form feedback data, the closed-loop feedback mechanism re-trains and updates the global collaborative model parameters according to the misjudgment information and key samples marked in the feedback data, and submits the standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, feedback data and decision-making process to the blockchain network for hash calculation and distributed storage respectively through the blockchain technology based on Hyperledger Fabric, so as to generate data traceability records and secure shared data.

[0104] After the early warning decision data is generated, the system further collects the on-site operator's feedback on the early warning decision data to form feedback data. The on-site operator reviews the early warning data through the early warning terminal and records the observed misjudgment situations and key abnormal samples on the operation interface. This feedback data is transmitted to the central platform in a fixed format and is used in the closed-loop feedback mechanism. The closed-loop feedback mechanism mainly re-trains and updates the global collaborative model parameters based on the misjudgment information and key samples marked in the feedback data. The update process usually adopts an online learning algorithm to correct the original model parameters using the feedback data, so that the global collaborative model parameters are closer to the on-site actual situation.

[0105] Specifically, if a certain area in the feedback data is marked as misjudged and abnormal multiple times, the system will adjust the weight of the corresponding data in the global model in this area to reduce the misjudgment rate. This process can adopt an incremental learning algorithm, and its core formula form is:

[0106]

[0107] where θ old is the old global collaborative model parameter, θ new is the updated model parameter, η is the learning rate, L(θ; F) is the loss function calculated based on the feedback data F, is the gradient of the loss function with respect to the model parameter. Through this online update mechanism, the system can achieve the adaptive adjustment of the global collaborative model parameters, thereby improving the accuracy of early warning decisions.

[0108] Meanwhile, to ensure the security, transparency, and immutability of the entire data governance process, the central platform uses Hyperledger Fabric blockchain technology to record all data. Specifically, the system calculates the hash values of standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graphs, real-time digital twin data, early warning decision-making data, feedback data, and key information in the decision-making process to generate unique digital fingerprints, and then records these fingerprints in the blockchain network in a distributed storage manner. The blockchain network ensures the immutability of each record through a consensus algorithm and allows each participating party to conduct real-time queries and audits. For example, assume the hash value calculation formula for standardized data is:

[0109] H = SHA-256(D)

[0110] where D represents the standardized data and H is the corresponding hash value. Using blockchain technology, the hash values of all data are linked into a chain in chronological order to form a complete data traceability record. This process ensures the full-process traceability of data from collection, preprocessing to decision feedback, and provides legal and technical guarantees for secure sharing. Through closed-loop feedback and blockchain recording technology, the system not only realizes the real-time update of prediction and anomaly detection in data governance, but also provides a highly secure and transparent decision support platform for coal mine safety management, thereby effectively reducing safety risks and improving overall operational efficiency.

[0111] As Figure 4 shown, a coal mine data governance system for implementing the method for coal mine data governance includes:

[0112] A data collection module for collecting original geological exploration data, original equipment sensor data, original production scheduling data, original environmental monitoring data, and original personnel positioning data, and converting the foregoing original data into standardized data and then preprocessing to generate real-time verification data; this module includes a variety of sensors and terminals installed at the coal mine site for collecting original data on geological exploration, equipment sensors, production scheduling, environmental monitoring, and personnel positioning. The data collection module is equipped with industrial-grade communication interfaces (such as MQTT, OPC UA, and Modbus interfaces), capable of converting various types of original data into a predefined format in real time, and performing preliminary formatting and data storage through the on-site data collection unit. The collected original data is output as standardized data after hardware processing, and then the preprocessing unit performs format, numerical range, and timestamp verification on the data, and finally generates real-time verification data. This module uses a dedicated data conversion circuit and an embedded controller to ensure stable operation and accurate data transmission in the harsh environment of the coal mine.

[0113] The edge processing module is set on the edge node and is used to extract the features of real-time verification data by using an autoencoder and calculate the mean square error to form reconstructed error data. The generator in the generative adversarial network is adopted to generate synthetic samples with a fixed-dimensional latent variable vector as the input, so that the synthetic samples are combined with the real-time verification data to form an enhanced data set. Then, the deep reinforcement learning model is used to perform anomaly determination on the data within a fixed time window and output edge processing data. Subsequently, the self-organizing clustering algorithm is used to globally optimize the clustering center in a way based on the principles of quantum annealing and quantum tunneling to generate clustering data. This module is set on the edge node at the coal mine site, and its hardware platform is equipped with a high-performance embedded processor and a dedicated acceleration unit for running data processing algorithms. After receiving the real-time verification data, the edge processing module first extracts the features of the data by using the autoencoder unit, performs encoding and decoding operations on the input data by using a convolutional variational autoencoder, and generates reconstructed error data by calculating the mean square error. The module also integrates a generative adversarial network, which generates synthetic samples through the input of a fixed-dimensional latent variable and combines the synthetic samples with the real-time verification data to form an enhanced data set. Then, the deep reinforcement learning unit determines the abnormal state of the multi-dimensional data within a fixed time window and outputs edge processing data. Finally, the edge processing module globally optimizes the edge processing data through the self-organizing clustering unit by using a clustering algorithm based on the principles of quantum annealing and quantum tunneling, determines the clustering center, and generates clustering data. This module realizes real-time data processing and efficient anomaly detection by integrating a dedicated AI acceleration chip.

[0114] The central processing module is set on the central platform. After uploading the edge processing data and clustering data, it uses the federated averaging algorithm to perform weighted fusion on the model parameters uploaded by each edge node according to the node data volume, and generates logical semantic data and global collaborative model parameters through the self-attention mechanism in combination with the preset symbol rules in the coal mine field. The logical semantic data is converted into unified semantic data according to a fixed mapping relationship, and a coal mine knowledge graph is constructed in the graph database using a graph convolutional network. The coal mine knowledge graph and the real-time verification data are synchronously updated on the digital twin platform at a set time interval to generate real-time digital twin data. The central processing module is located in the central server or data center, and its hardware platform supports high-concurrency data storage and distributed computing. This module first receives the edge processing data and clustering data uploaded by each edge node, and ensures the complete transmission of data through a high-speed network interface. The central processing module has a built-in federated learning platform, which uses the federated averaging algorithm to perform weighted fusion on the model parameters uploaded by each edge node. The weighting coefficient is determined according to the data volume and reconstruction error of each node. The fused model parameters are combined with the preset coal mine symbol rules in the self-attention processing unit to realize the joint calculation of the rule vector and the neural network embedding vector, and generate logical semantic data and global collaborative model parameters. Subsequently, the central processing module converts the logical semantic data into unified semantic data according to a fixed mapping table, and constructs a coal mine knowledge graph in the graph database using a graph convolutional network. The constructed coal mine knowledge graph is updated on the digital twin platform with the real-time verification data through a timing synchronization mechanism to generate real-time digital twin data. This module uses a high-performance server cluster and GPU accelerators to achieve large-scale data fusion and complex model calculations, ensuring the accuracy and real-time performance of data processing.

[0115] The prediction feedback module is set on the digital twin platform. It is used to perform time series prediction and anomaly detection on key indicators by combining long short-term memory network with graph convolutional network and self-attention mechanism based on real-time digital twin data, global collaboration model parameters, and coal mine knowledge graph to generate early warning decision data. It also collects the feedback of on-site operators on the early warning decision data to form feedback data, uses the closed-loop feedback mechanism to update the global collaboration model parameters, and at the same time records standardized data, real-time verification data, edge processing data, clustering data, global collaboration model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, feedback data, and decision-making process based on Hyperledger Fabric blockchain technology to generate data traceability records and secure shared data. The prediction feedback module is set on the digital twin platform, and its hardware system integrates a high-performance simulation engine and a real-time data interface. This module analyzes the key indicators of the coal mine site by using the real-time digital twin data, global collaboration model parameters, and coal mine knowledge graph from the central processing module. Specifically, the prediction feedback module has a built-in long short-term memory network (LSTM) unit to recursively model the key indicator sequence in the real-time digital twin data and generate prediction trend data. At the same time, it performs anomaly detection on historical data and real-time digital twin data through the graph convolutional network and self-attention mechanism, and outputs anomaly prediction data. Subsequently, the prediction feedback module fuses the prediction trend data and the anomaly prediction data according to a fixed weight to generate early warning decision data for guiding coal mine safety early warning and emergency response. In addition, this module is also equipped with an on-site feedback collection terminal, through which on-site operators can give feedback on the early warning decision data. The feedback data is used to update the global collaboration model parameters through the closed-loop feedback mechanism to further optimize the model performance. At the same time, the prediction feedback module performs hash calculation and distributed storage on all key data (including standardized data, real-time verification data, edge processing data, clustering data, global collaboration model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, on-site feedback data, and decision-making process) through the blockchain recording unit (based on Hyperledger Fabric) to form data traceability records and secure shared data, thus ensuring the immutability and traceability of data and decision-making process. By comprehensively using advanced time series prediction, anomaly detection, and blockchain technology, this module realizes the real-time monitoring of coal mine site data and secure decision support.

[0116] A storage medium stores a computer program thereon, and when the program is executed by a processor, it implements the steps of the data governance method for coal mines.

[0117] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0118] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A data governance method for coal mines, characterized in that, It includes the following steps: Collect the original data of geological exploration, equipment sensors, production scheduling, environmental monitoring, and personnel positioning, convert it into standardized data, and preprocess it to obtain real-time verification data; On the edge node, use an autoencoder to extract features from the real-time verification data and calculate the reconstruction error to form reconstruction error data. At the same time, use a generative adversarial network to generate an enhanced data set, and use deep reinforcement learning to perform anomaly determination on the fixed-time window data to output edge processing data, and then obtain clustering data through self-organizing clustering; Upload the edge processing data and clustering data to the central platform. The central platform uses federated learning to fuse the parameters of each edge model, and combines the coal mine symbol rules to generate logical semantic data and global collaborative model parameters through a self-attention mechanism, and then maps them into unified semantic data and constructs a coal mine knowledge graph. The knowledge graph and the real-time verification data are synchronized and updated on the digital twin platform to generate real-time digital twin data; Based on the real-time digital twin data, global collaborative model parameters, and coal mine knowledge graph, use a long short-term memory network combined with a graph convolutional network and a self-attention mechanism to perform time series prediction and anomaly detection on key indicators to generate early warning decision data, and collect on-site feedback to update the global collaborative model parameters in a closed loop. At the same time, record the data and decision-making process to achieve data traceability and secure sharing.

2. The method according to claim 1, wherein The steps of using an autoencoder on the edge node to extract features from the real-time verification data and calculate the reconstruction error to form reconstruction error data include: Use a convolutional variational autoencoder to perform encoding and decoding operations on the real-time verification data, and determine the anomaly degree of each data by calculating the mean square error.

3. The method according to claim 1, wherein The steps of using a generative adversarial network on the edge node to generate an enhanced data set include: Generate synthetic samples by inputting a latent variable vector of a fixed length into the generator, and combine the synthetic samples with the real-time verification data to form an enhanced data set for expanding the coverage of the data distribution.

4. The method according to claim 1, wherein The steps of using deep reinforcement learning on the edge node to perform anomaly determination on the fixed-time window data include: Take the real-time verification data, reconstruction error data, and enhanced data set as state inputs, and the deep reinforcement learning model outputs the anomaly state confidence, and divides the data categories according to the confidence to generate edge processing data.

5. The method according to claim 1, characterized in that, On the edge node, perform global search optimization on the edge processing data through self-organizing clustering to determine the clustering center and output clustering data. The self-organizing clustering adaptively adjusts the position of the clustering center using the principles of quantum annealing and quantum tunneling.

6. The method according to claim 1, characterized in that, After uploading the edge processing data and clustering data to the central platform, the central platform first sets the weighting coefficients according to the data volume and the mean reconstruction error of each edge node, then fuses the model parameters of each edge node through the federated averaging algorithm, and combines the preset symbol rules in the coal mine field to perform self-attention calculation on the input rules and the neural network embedding vectors to generate logical semantic data and global collaborative model parameters. Subsequently, according to the fixed mapping table, the logical semantic data is converted into unified semantic data, and a graph convolutional network is used to construct a graph structure and perform association reasoning on the unified semantic data. Finally, a coal mine knowledge graph is formed and synchronized and updated with the real-time verification data at preset time intervals in the digital twin platform to generate real-time digital twin data.

7. The method according to claim 1, wherein The step of using a long short-term memory network combined with a graph convolutional network and a self-attention mechanism to perform time series prediction and anomaly detection on key indicators based on the real-time digital twin data, the global collaborative model parameters, and the coal mine knowledge graph to generate early warning decision data includes: Using a long short-term memory network to recursively model the key indicator sequence in the real-time digital twin data to generate predicted trend data, performing anomaly detection and analysis on the association relationships in the coal mine knowledge graph through a graph convolutional network in combination with the global collaborative model parameters, and comprehensively combining the predicted trend data and historical data under the action of the self-attention mechanism. Finally, the predicted trend data and the anomaly detection results are fused to output early warning decision data.

8. The method according to claim 1, wherein After collecting the feedback of on-site operators on the early warning decision data to form feedback data, the closed-loop feedback mechanism re-trains and updates the global collaborative model parameters according to the misjudgment information and key samples marked in the feedback data, and submits the standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, as well as the feedback data and the decision-making process to the blockchain network for hash calculation and distributed storage through the blockchain technology based on Hyperledger Fabric, so as to generate data traceability records and secure shared data.

9. A system for coal mine data governance, which is used to implement the method for coal mine data governance described in any one of claims 1 to 8, characterized in that, The system includes: A data acquisition module, which is used to acquire the original geological exploration data, the original equipment sensor data, the original production scheduling data, the original environmental monitoring data, and the original personnel positioning data, and convert the foregoing original data into standardized data and then preprocess it to generate real-time verification data; An edge processing module, which is set on the edge node, is used to extract the features of the real-time verification data by using an autoencoder and calculate the mean square error to form reconstruction error data, use the generator in the generative adversarial network to generate synthetic samples with a fixed-dimensional latent variable vector as the input, combine the synthetic samples with the real-time verification data to form an enhanced data set, and use a deep reinforcement learning model to perform anomaly determination on the data within a fixed time window and output edge processing data. Subsequently, the self-organizing clustering algorithm is used to globally optimize the clustering centers in a manner based on the principles of quantum annealing and quantum tunneling to generate clustering data; The central processing module is set on the central platform. After uploading the edge processing data and clustering data, it uses the federated averaging algorithm to perform weighted fusion on the model parameters uploaded by each edge node according to the node data volume, and combines the preset symbol rules in the coal mine field to generate logical semantic data and global collaborative model parameters through the self-attention mechanism. The logical semantic data is converted into unified semantic data according to the fixed mapping relationship, and the coal mine knowledge graph is constructed in the graph database using the graph convolutional network. The coal mine knowledge graph and the real-time verification data are synchronously updated on the digital twin platform at a set time interval to generate real-time digital twin data; The prediction feedback module is set on the digital twin platform. It performs time series prediction and anomaly detection on key indicators using the long short-term memory network combined with the graph convolutional network and the self-attention mechanism based on the real-time digital twin data, global collaborative model parameters, and coal mine knowledge graph to generate early warning decision data, and collects the feedback from on-site operators on the early warning decision data to form feedback data. It uses the closed-loop feedback mechanism to update the global collaborative model parameters. At the same time, it records the standardized data, real-time verification data, edge processing data, clustering data, global collaborative model parameters, coal mine knowledge graph, real-time digital twin data, early warning decision data, feedback data, and the decision-making process based on the Hyperledger Fabric blockchain technology to generate data traceability records and secure shared data.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the data governance method for coal mines according to any one of claims 1 to 8.

Images