Mine multi-source data fusion transmission method and device based on distributed edge agent collaboration

Abstract

The invention discloses a mine multi-source data fusion transmission method and device based on distributed edge agent collaboration, and relates to the field of data transmission. The method comprises the following steps: collecting heterogeneous data through a plurality of data collection terminals deployed in a mine production environment, and generating a data voucher based on the heterogeneous data; analyzing a value mark through a master agent to split a data processing task into task sub-pieces of different processing types, selecting a slave agent from bidding results of other edge computing nodes for bid invitation information, and constructing the master agent and the slave agent into an edge agent alliance for processing heterogeneous data; performing real-time interaction of event features in the edge agent alliance, and executing cross-modal joint coding on the associated heterogeneous data streams based on the event features; and aggregating and uploading the data subjected to joint coding to the cloud in a transmission window returned by the 5G base station. By implementing the technical scheme, mine multi-source heterogeneous data can be effectively processed, and on-demand distribution and dynamic scheduling of network resources can be realized.

Description

Technical Field

[0001] This application relates to the field of data transmission, and in particular to a method and apparatus for multi-source data fusion and transmission in mines based on distributed edge agent collaboration. Background Technology

[0002] In the mining industry, with the development of digitalization and intelligence, the fusion and transmission of multi-source data is of great significance for improving the efficiency, safety and management level of mining production.

[0003] To address the challenge of fusion and transmission of multi-source data in mines, existing technologies typically employ the following methods: One is a centralized data processing and transmission approach, where all collected data is aggregated at a central node, which then processes and transmits the data. Another is the traditional distributed data processing approach, which distributes data processing tasks across multiple nodes.

[0004] The shortcomings of existing technologies lie in the fact that centralized processing methods cannot meet the real-time processing and transmission needs of large amounts of heterogeneous data from multiple sources in mines, easily creating data processing bottlenecks. Traditional distributed processing methods lack effective data labeling and coordination mechanisms, making data prone to errors and loss during transmission and processing, affecting the effectiveness of data fusion. In addition, while the use of independent communication links can meet basic needs, it suffers from resource waste, network congestion, and high complexity in backend data parsing, leading to a waste of 5G frequency band resources, increasing the risk of network congestion, and the lack of uniformity in different types of data transmission protocols increases the complexity of cloud data parsing, making it difficult to meet the high requirements of real-time decision-making. Summary of the Invention

[0005] This application provides a method and apparatus for multi-source data fusion and transmission in mines based on distributed edge intelligent agent collaboration. It effectively processes multi-source heterogeneous data in mines, realizes cross-modal joint coding, reasonably predicts transmission demand, and realizes on-demand allocation and dynamic scheduling of network resources, ensuring the real-time performance of critical data and reducing the risk of network congestion.

[0006] Firstly, this application provides a method for multi-source data fusion and transmission in mines based on distributed edge agent collaboration, the method comprising: Heterogeneous data is collected by multiple data acquisition terminals deployed in the mining production environment. Based on the device type of the data source and the preset value rule base, a value tag is generated for the heterogeneous data through a 5G converged gateway. The hash value of the heterogeneous data is calculated as a digital fingerprint based on the value tag. Data credentials are generated based on the digital fingerprint and broadcast to a distributed ledger jointly maintained by multiple edge computing nodes deployed in the same area. The first edge computing node that verifies and records the data credentials in the distributed ledger is determined as the master agent. The master agent parses the value tag and splits the data processing task into task sub-slices of different processing types. It then broadcasts bidding information containing the resource requirements of each task sub-slice to other edge computing nodes. From the bidding results of the other edge computing nodes for the bidding information, slave agents are selected. The master agent and the slave agents are constructed into an edge agent alliance to process the heterogeneous data. Within the edge agent alliance, each slave agent constructs a data processing pipeline according to the logical order of the task sub-slices. During the pipeline processing, slave agents processing different modal data interact in real time with event features and perform cross-modal joint coding on the associated heterogeneous data streams based on the event features. Based on the historical task logs stored locally, the edge intelligence alliance predicts the transmission requirements of the data packets to be uploaded, generates a transmission request suggestion packet containing predicted traffic, priority and suggested transmission parameters, and sends it to the 5G base station. It receives the transmission window returned by the 5G base station and aggregates and uploads the jointly encoded data to the cloud within the transmission window.

[0007] By adopting the above technical solutions, value marking, digital fingerprinting, and distributed ledger mechanisms are used to ensure the traceability of the source and the verifiability of ownership of heterogeneous data, thereby enhancing data credibility and anti-tampering capabilities. Based on the construction of a master-slave intelligent agent alliance and a task bidding mechanism, flexible scheduling and efficient collaboration of edge computing resources are achieved, improving the utilization rate of computing resources and the efficiency of task execution. Real-time event feature interaction and cross-modal joint coding are introduced into pipeline processing to achieve semantic-level fusion of multi-source heterogeneous data, improving the consistency and information density of data expression. Based on historical task logs, transmission demand is predicted, and transmission windows are negotiated with 5G base stations to achieve on-demand allocation and dynamic scheduling of network resources, ensuring the real-time performance of critical data and reducing the risk of network congestion.

[0008] In some embodiments, generating value tags for heterogeneous data through a 5G converged gateway based on the device type of the data source and a preset value rule base includes: The 5G converged gateway parses the received heterogeneous data, identifies the device type and data type, and extracts the semantic units corresponding to the data type. Based on the device type, the content of the semantic unit, and the frequency with which historical similar data is called and used by the cloud analysis model to generate decisions, the unit value density score of the semantic unit is obtained. The semantic units whose unit value density scores are higher than a preset first threshold are aggregated into high-value data segments in the heterogeneous data, the semantic units whose unit value density scores are lower than a preset second threshold are aggregated into low-value data segments, and the rest are medium-value data segments. A segment-level value tag is generated for each data segment, and a segment mapping table is generated based on the segment-level value tag. Integrate all segment-level value tags to generate a global value tag for the heterogeneous data.

[0009] By employing the above technical solution, data is divided into high, medium, and low value segments through value density scoring of semantic units, achieving granular value identification and labeling of data content and providing a basis for subsequent differentiated management. Through segment-level value labeling and segment mapping tables, different strategies can be adopted for different value segments during transmission and processing, optimizing resource allocation. Combining historical call frequency with semantic content for value scoring enables value labeling to learn and adapt, better aligning with actual business needs and decision-making importance. By generating global value labels and segment mapping tables, a value management system from local to global is formed, facilitating system-level data scheduling, storage, and analysis optimization.

[0010] In some embodiments, calculating the hash value of the heterogeneous data as a digital fingerprint based on the value marker includes: The 5G converged gateway extracts the original data blocks corresponding to each data segment from the original byte stream of the heterogeneous data according to the byte range corresponding to each data segment in the segmentation mapping table. For each of the original data blocks, calculate the content hash as the block-level fingerprint of the data block; Based on the block-level fingerprints of multiple data blocks belonging to the target data segment, a Merkle subtree of the target data segment is constructed, and the root hash of the Merkle subtree is used as the segment integrity fingerprint of the target data segment, wherein the target data segment is any one of the heterogeneous data segments; Extract the segment-level value tag corresponding to the target data segment, and calculate the hash value of the segment-level value tag as the segment value digest fingerprint of the target data segment; Using the segment integrity fingerprints and segment value digest fingerprints of all data segments as leaf nodes, construct a global Merkle tree and generate a global root hash; The global root hash, the segment integrity fingerprint of each data segment, the segment value digest fingerprint, and the segment mapping table are jointly encapsulated to generate the digital fingerprint of the heterogeneous data.

[0011] By employing the above technical solution, a multi-level anti-tampering verification system is established from byte blocks to the complete data stream through hierarchical encapsulation of block-level fingerprints, segment-level integrity fingerprints, and a global Merkle tree, ensuring data integrity during transmission and processing. Encapsulating the segment-level value-tagged digest fingerprint with the data content fingerprint achieves an inseparable binding between data value attributes and the content itself, facilitating subsequent joint auditing of data value and authenticity. Through segmented mapping tables and hierarchical fingerprint structures, not only can the overall data be verified against tampering, but the specific modified data segments or even data blocks can be located, enhancing data credibility and security traceability. The encapsulated digital fingerprint contains complete hierarchical verification information and mapping relationships, facilitating direct use in different systems, nodes, or audit stages, supporting efficient data consistency comparison and trusted exchange.

[0012] In some embodiments, the slave agent processing different modal data interacts with event features in real time during pipeline processing, specifically including: When the first slave agent that processes time-series sensing data detects locally that the raw data value of the monitored target parameter exceeds a preset third threshold, it sends a spatiotemporal perception request packet to the master agent. The spatiotemporal perception request packet includes event features, which include a timestamp, an event type, and the physical location of the first slave agent. Based on the event type and the physical location, the main agent retrieves and wakes up multiple geographically or logically related secondary agents from the device topology and semantic relationship graph; The master agent broadcasts collaborative perception instructions to multiple second slave agents and receives confirmation signals and confidence levels returned by multiple second slave agents. The collaborative perception instructions include the event type and time window, and the confirmation signals are used to characterize the existence of collaborative verification features. If the number of confirmation signals and the weighted confidence value received by the master agent within a preset decision time exceed a preset fourth threshold, then the cross-modal collaborative perception event is determined to be established, a fusion event descriptor is generated, and the fusion event descriptor is distributed to all target second slave agents that return the confirmation signal. The fusion event descriptor includes event type, fusion timestamp, trigger source ID, list of collaborative verification source IDs, and data value confidence.

[0013] By employing the above technical solution, collaborative perception and cross-validation of multi-source, multi-modal data are achieved through event triggering by the first slave agent and dynamic wake-up of associated agents by the master agent, thereby improving the accuracy and reliability of abnormal event detection. The retrieval and wake-up of associated agents based on device topology and semantic relationship graphs ensures that the agents participating in the collaboration are physically or logically related, enhancing the contextual understanding capability of event recognition. A confidence-weighted and quantity threshold mechanism is used to quantitatively evaluate the collaborative perception results, assisting the master agent in quickly and objectively determining whether an event is valid and generating a structured fusion event descriptor, providing high-quality input for subsequent decision-making.

[0014] In some embodiments, performing cross-modal joint coding on the associated heterogeneous data streams based on the event features includes: Within the edge agent alliance, the target second slave agent parses the fused event descriptor to obtain the event type and the data value confidence, and obtains the available computing resources, prediction bandwidth and current mine production stage information of the target second slave agent in real time, and fuses them to form a dynamic context. Based on the dynamic context, a personalized encoding strategy vector is generated through a preset strategy generation model. The target second slave agent publishes a virtual resource offer within the edge agent alliance based on the resource requirements in the personalized coding strategy vector, and selects a third slave agent based on the bidding results of other slave agents within the edge agent alliance; The target second agent encodes the video stream, and the third agent locates the associated data segments from the non-video modal data stream corresponding to the list of co-verification source IDs that overlap in time or are logically related to the event features extracted from the video stream. The associated data segments are encoded using encoding parameters higher than a first quality threshold, and the unassociated data segments are compressed using encoding parameters lower than a second quality threshold, where the second quality threshold is lower than the first quality threshold. The video data encoded by the second agent and the non-video data encoded or compressed by the third agent are bound and encapsulated according to a unified spatiotemporal index to obtain a multimodal data unit.

[0015] By adopting the above technical solution, personalized encoding strategies are generated based on the data value confidence and dynamic context in the fused event descriptor. This ensures high-quality encoding of high-value associated data and low-quality compression of low-value unassociated data, achieving precise matching of resource allocation and data value. Through a virtual resource offer and bidding mechanism among intelligent agents, computing and bandwidth resources are flexibly allocated within the edge alliance, enhancing the distributed collaboration capability and resource utilization efficiency of encoding tasks. Video and non-video data are bound and encapsulated according to a unified spatiotemporal index, forming structured multimodal data units. This ensures spatiotemporal alignment of cross-modal data, facilitating subsequent joint analysis and retrieval. The encoding strategy is adjusted by combining real-time dynamic context and production stage information, enabling the encoding process to adapt to changes in network conditions, resource status, and business needs, improving the system's overall responsiveness and adaptability in complex environments.

[0016] In some embodiments, the step of predicting the transmission requirements of the data packets to be uploaded based on locally stored historical task logs by the edge agent alliance, and generating a transmission request suggestion packet containing predicted traffic, priority, and suggested transmission parameters, specifically includes: The traffic baseline is obtained by performing time-series pattern analysis on the historical task logs through the edge agent alliance. The number of currently active slave agents in the edge agent alliance, the current computing load of each slave agent, and the data value confidence carried by the fusion event descriptor are obtained. The data value confidence is mapped to a transmission priority weight. The real-time collaborative processing capability coefficient is calculated by combining the number of currently active slave agents and the computing load. The traffic baseline is corrected using the real-time collaborative processing capability coefficient to obtain the predicted traffic. The predicted traffic, the transmission priority weight, and the current average network latency and packet loss rate are used as query features and input into the transmission strategy knowledge base of the edge agent alliance to match a suggested transmission parameter set. The transmission parameter set includes coding redundancy, maximum retransmission count, and first packet transmission latency. The transmission strategy knowledge base stores a variety of preset mapping relationships between network states and task features and the optimal transmission parameter set. The predicted traffic, the transmission priority weight, and the transmission parameter set are assembled to generate the transmission request suggestion packet.

[0017] By employing the above technical solutions, a traffic baseline is established through time-series pattern analysis and corrected by incorporating real-time collaborative processing capability coefficients. This makes traffic prediction more closely aligned with the current task execution status, improving prediction accuracy and adaptability. Data value confidence is mapped to transmission priority weights, and combined with agent load and quantity calculation collaborative processing capabilities, intelligent matching of transmission priority with task importance and resource status is achieved. By querying the transmission strategy knowledge base, the optimal transmission parameter set (such as coding redundancy, retransmission strategies, etc.) is matched based on predicted traffic, priority, and real-time network status, improving transmission reliability and efficiency. Predictions, priorities, and transmission parameters are encapsulated into structured transmission request suggestion packets, providing 5G base stations with clear and executable transmission scheduling criteria, enhancing the system's collaborative decision-making capabilities in dynamic network environments.

[0018] In some embodiments, the method further includes: If the 5G base station receives multiple transmission request suggestion packets from different edge agent alliances in the same region within the same scheduling period, the 5G base station sorts the multiple transmission request suggestion packets according to the transmission priority weight in each transmission request suggestion packet and the first packet transmission delay, and performs preliminary non-overlapping window pre-allocation on the time-frequency resource grid based on the resources required by each transmission request suggestion packet to form multiple candidate transmission window schemes. The 5G base station will feed back the candidate window scheme, which includes the window start time, window duration, and the guaranteed bandwidth corresponding to the window, to the corresponding edge agent alliance. If the corresponding edge agent alliance confirms acceptance of the candidate window scheme, the 5G base station locks and reserves the corresponding radio resource block on the time-frequency resource grid before the start time of the window, and sends a transmission window confirmation signal to the corresponding edge agent alliance. The transmission window confirmation signal includes window parameters and a unique transaction identifier for this reservation, which is used for access verification during subsequent data upload.

[0019] By adopting the above technical solution, the base station uniformly sorts and pre-allocates transmission request suggestion packets from various alliances, avoiding congestion or conflicts caused by resource contention and ensuring that high-priority, low-latency tasks receive transmission resources first. The base station performs non-overlapping window pre-allocation on the time-frequency resource grid, forming a structured and executable candidate transmission window scheme, providing a systematic resource scheduling framework for multi-task concurrent transmission. Through window locking, resource reservation, and unique transaction identification mechanisms, it ensures that the transmission window is exclusively used within the agreed time, preventing resource preemption and conflicts, while supporting traceability and auditability of the transmission process. By feeding back candidate window schemes and receiving confirmation signals, a dynamic negotiation process is established between the base station and the edge agent alliance, improving the flexibility of resource allocation and system collaboration efficiency.

[0020] In a second aspect, embodiments of this application provide a computer system including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the method described in any possible implementation of the first aspect.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0022] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0023] It is understood that the computer system provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By using value marking, digital fingerprinting, and distributed ledger mechanisms, the source of heterogeneous data can be traced, its value can be assessed, and its content cannot be tampered with, thereby improving the credibility and security of data throughout the entire process of collection, transmission, and processing, and providing a technical foundation for mine data ownership confirmation and auditing; 2. Based on the task splitting and bidding mechanism of the main intelligent agent, an edge intelligent agent alliance is formed on demand and dynamically negotiated, realizing flexible scheduling and load balancing of computing resources, avoiding single point bottlenecks, and improving overall task execution efficiency and system scalability. 3. Through event-feature-driven real-time interaction and cross-modal joint coding, spatiotemporal alignment and semantic association of multi-source heterogeneous data such as video and sensors are achieved, improving the quality of data fusion and providing structured, high-information-density multimodal data units for subsequent intelligent analysis; 4. Based on historical task logs, predict transmission demand and negotiate transmission windows with 5G base stations to achieve proactive reservation and dynamic allocation of network resources, ensure the transmission priority of high-value, real-time data, reduce network congestion risks, and improve overall transmission efficiency and reliability. 5. From data collection and processing to transmission, a closed-loop system with full automation and intelligent collaboration is formed, reducing manual intervention and improving the real-time performance, adaptability, and overall operational efficiency of mine data fusion and transmission, providing a reliable data foundation for the construction of smart mines. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the multi-source data fusion and transmission method for mines based on distributed edge intelligent agent collaboration in this application embodiment; Figure 2 This is a schematic diagram of the process for generating a transmission request suggestion packet in an embodiment of this application; Figure 3 This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application. Detailed Implementation

[0026] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0028] The following is combined Figure 1 The method of the embodiments of this application will be described below.

[0029] Figure 1 This is a flowchart illustrating the multi-source data fusion and transmission method for mines based on distributed edge intelligent agent collaboration, as described in this application embodiment. Figure 1 As shown, the method includes the following steps: S101. Collect heterogeneous data through multiple data acquisition terminals deployed in the mining production environment. Based on the device type of the data source and the preset value rule base, generate value tags for the heterogeneous data through a 5G converged gateway. Calculate the hash value of the heterogeneous data as a digital fingerprint based on the value tags. Generate data credentials based on the digital fingerprint and broadcast them to a distributed ledger jointly maintained by multiple edge computing nodes deployed in the same area. S102. The first edge computing node that verifies and records the data certificate in the distributed ledger is determined as the master agent. The master agent parses the value tag and splits the data processing task into task sub-slices of different processing types. The master agent broadcasts bidding information containing the resource requirements of each task sub-slice to other edge computing nodes. The master agent selects a slave agent from the bidding results of the other edge computing nodes for the bidding information. The master agent and the slave agent are constructed into an edge agent alliance for processing the heterogeneous data. S103. Within the edge agent alliance, each slave agent constructs a data processing pipeline according to the logical order of the task sub-slices. During the pipeline processing, the slave agents processing different modal data perform real-time interaction of event features and perform cross-modal joint coding on the associated heterogeneous data streams based on the event features. S104. Based on the historical task logs stored locally, the edge intelligence alliance predicts the transmission requirements of the data packets to be uploaded, generates a transmission request suggestion packet containing predicted traffic, priority and suggested transmission parameters, and sends it to the 5G base station. It receives the transmission window returned by the 5G base station and aggregates and uploads the jointly encoded data to the cloud within the transmission window.

[0030] In mining production environments, various sensors, video surveillance systems, equipment controllers, and other terminals continuously generate multimodal, multi-protocol, and high-concurrency heterogeneous data streams. After receiving data from different terminals, the 5G converged gateway performs refined value labeling based on device type and a pre-built value rule base. Specifically, the gateway parses the data content, identifies its source device (such as gas sensors, high-definition cameras, motor controllers, etc.) and data type, and combines historical data analysis models to assign high, medium, and low value labels to different data segments based on the frequency of data access and its contribution to decision-making. This process essentially dynamically correlates data content with its business importance and real-time requirements, providing a basis for subsequent differentiated processing. Based on the generated value labels, the system generates a unique digital fingerprint for each piece of data. This fingerprint is not a simple overall hash calculation, but rather generates block hashes segment by segment based on the data segment mapping corresponding to the value label, constructs a Merkle tree structure, and ultimately forms a global root hash. This design ensures both data integrity and verifiability, and enables the ability to pinpoint specific data segments to where tampering occurred. The system encapsulates digital fingerprints into data credentials and broadcasts them to a distributed ledger jointly maintained by multiple edge computing nodes within the region. This mechanism effectively prevents data forgery, tampering, and source disputes, laying a solid foundation for the auditing, traceability, and definition of rights and responsibilities of mining data assets.

[0031] Once a data credential is successfully stored in the distributed ledger, the first edge computing node to successfully verify and record it is automatically designated as the master agent for processing the data. The master agent parses the value markers in the data credential, identifies key data processing steps (such as anomaly detection, video compression, feature extraction, and protocol conversion), and breaks down the overall task into multiple task sub-slices with specific resource requirements (such as computing power, memory, and I / O bandwidth). The master agent broadcasts bidding information containing the resource requirements of each task sub-slice to other edge nodes within the region. Each edge node submits resource bids and capability declarations based on its current load, resource availability, and task type matching, forming a competitive bidding response. The master agent comprehensively considers factors such as the responding node's computing power, historical reliability, and resource costs (such as energy consumption and latency) to select the optimal node as a slave agent from the bidders. It can also select different slave agents depending on the task. The master agent and all selected slave agents together form a temporary edge agent alliance for that data processing task.

[0032] Within the intelligent agent alliance, data processing is not simply parallelized, but rather a highly efficient data processing pipeline is constructed based on the logical dependencies of task sub-slices. Different intelligent agents undertake specific links in the pipeline, realizing the pipelining of the processing flow and reducing overall latency. More importantly, this application's embodiments introduce a real-time event feature interaction mechanism to address the need for multimodal data linkage in mining environments. For example, when an intelligent agent processing gas sensor data detects excessive concentration (event feature), it immediately reports information such as event type, timestamp, and location to the master intelligent agent. The master intelligent agent automatically wakes up other geographically or logically related intelligent agents (such as the intelligent agent responsible for video analysis in that area) based on a pre-set device topology and semantic relationship graph, triggering collaborative perception. After confirming the establishment of a cross-modal event, the relevant intelligent agents initiate cross-modal joint coding. Taking "gas anomaly + video linkage" as an example: the video intelligent agent uses high-fidelity encoding for video streams during key time periods before and after the event to ensure clear image details; while for video streams outside the event period or non-critical areas, it uses high compression rate encoding. Simultaneously, data segments related to the event in other sensor data streams are also identified and their encoding quality improved. The encoding process can even dynamically schedule computationally intensive encoding tasks to idle nodes through resource bidding within the consortium. Ultimately, all associated multimodal data are encapsulated according to a unified spatiotemporal index, forming structured multimodal data units. This approach changes the traditional model of independently encoding and transmitting each modality of data before laboriously fusing them, enabling fusion processing at the edge and significantly improving the effective information density and fusion efficiency of the data.

[0033] While processing data, the Edge Agent Consortium continuously analyzes locally stored historical task logs and uses time-series analysis models to predict the baseline traffic of data expected to be uploaded in the following week. Simultaneously, it dynamically adjusts the predicted traffic based on the number of active agents within the consortium, the real-time load of each node, and the confidence level of the data being processed, and calculates the comprehensive priority weight for this transmission. Based on the predicted traffic, priority weight, and real-time network conditions (such as average latency and packet loss rate), the Edge Agent Consortium queries its locally maintained transmission strategy knowledge base (which, through historical experience learning or preset rules, establishes a mapping relationship from network conditions and task characteristics to optimal transmission parameters—such as coding redundancy, maximum retransmission count, and first packet transmission latency) to generate a set of suggested transmission parameters. This information is encapsulated into a structured transmission request suggestion packet and proactively sent to the 5G base station responsible for its region. As a resource coordinator, the 5G base station, upon receiving multiple requests from different intelligent agent alliances in the same area, performs global optimization scheduling on the time-frequency resource grid based on the priority, latency requirements, and total resource needs of each request. It pre-allocates non-overlapping transmission windows for each alliance and provides candidate solutions, including window time and guaranteed bandwidth, back to the respective alliance. After alliance confirmation, the base station locks resources before the agreed-upon time and issues the final transmission window confirmation signal. Within the agreed-upon dedicated transmission window, the edge intelligent agent alliance aggregates jointly encoded and structured multimodal data units and efficiently and reliably uploads them to the cloud. This transmission mode achieves precise matching and on-demand guarantee of 5G network resources and edge computing task requirements, effectively avoiding congestion caused by random competition and ensuring the transmission quality of high-value, high-real-time data.

[0034] In some embodiments, generating value tags for heterogeneous data through a 5G converged gateway based on the device type of the data source and a preset value rule base includes: The 5G converged gateway parses the received heterogeneous data, identifies the device type and data type, and extracts the semantic units corresponding to the data type. Based on the device type, the content of the semantic unit, and the frequency with which historical similar data is called and used by the cloud analysis model to generate decisions, the unit value density score of the semantic unit is obtained. The semantic units whose unit value density scores are higher than a preset first threshold are aggregated into high-value data segments in the heterogeneous data, the semantic units whose unit value density scores are lower than a preset second threshold are aggregated into low-value data segments, and the rest are medium-value data segments. A segment-level value tag is generated for each data segment, and a segment mapping table is generated based on the segment-level value tag. Integrate all segment-level value tags to generate a global value tag for the heterogeneous data.

[0035] After receiving data streams from different terminals, the 5G converged gateway performs deep parsing. This includes not only traditional protocol conversion (such as unifying Modbus, TCP / IP, etc., into 5G standard protocols), but more importantly, content-aware parsing. Based on a pre-built device database, the system identifies the device type of the data source (e.g., "Gas Sensor_01 Roadway", "High-Definition PTZ Camera_Mining Face_East Side") and determines its data type (e.g., "Real-time Monitoring Value", "Video Stream I-Frame", "Equipment Status Log"). For unstructured data content (such as video frames and text logs), the gateway integrates a lightweight semantic analysis module capable of extracting semantic units with clear business meaning. For example, from the gas sensor data stream, it extracts units such as "Concentration Value: 1.5% CH4", "Status: Normal", and "Location: Roadway A". From keyframes of the video stream, through lightweight edge target detection, it extracts units such as "Number of Personnel: 3", "Equipment Type: Coal Mining Machine", and "Activity Status: Operating". These semantic units form the basis for data value analysis, transforming the raw byte stream into machine-understandable and business-related information fragments. For each extracted semantic unit, the system calculates its unit value density score using a dynamic scoring model. This model comprehensively considers equipment type weight, semantic content sensitivity, and historical utility feedback. Equipment type weight: Basic weights are pre-assigned based on the equipment's criticality in safe production; for example, the weight of "gas sensor" is much higher than that of "ambient temperature and humidity sensor." Semantic content sensitivity: The sensitivity of the unit's content itself is analyzed; for example, "concentration value: 1.5% CH4" (close to the warning value) is more sensitive than "concentration value: 0.2% CH4"; in the video, "personnel entering a dangerous area" is more sensitive than "equipment operating normally." Historical utility feedback: The system continuously tracks the frequency with which similar semantic units are analyzed by cloud-based analysis models (such as risk warning models and production scheduling models), and the frequency with which they are adopted and ultimately trigger effective decisions (such as issuing alarms or adjusting parameters). The higher the frequency, the greater the decision contribution of this type of data, and the higher its value score. By weightedly integrating the above factors, the model outputs a quantified value density score for each semantic unit. This process enables data value assessment to learn and evolve: data features that have proven useful historically will be assigned higher expected value in the future. After obtaining scores for all semantic units, the system stratifies them according to preset thresholds. High-value data segments: These are aggregated from the original byte ranges corresponding to semantic units whose value density scores are higher than the first threshold. This type of data typically contains critical alarms, serious anomalies, core features of security events, etc., requiring the highest priority processing and the highest reliability transmission. Low-value data segments: These are aggregated from the byte ranges corresponding to semantic units whose scores are lower than the second threshold.This type of data mainly consists of environmental background values, periodic normal status reports, and redundant log information, which can tolerate delayed processing, high-ratio compression, or even phased discarding. Medium-value data segments: Data between these two types constitutes the medium-value segment, including general monitoring data and routine operation records. Basic processing and transmission must be guaranteed, but resource allocation priority should be moderate. The system generates a segment-level value tag for each data segment, containing information such as value level, start and end byte offsets, and key semantic summaries. Subsequently, all segment-level tags are aggregated to generate a segment mapping table. The system integrates the segment-level value tags of each segment to generate a global value tag for this heterogeneous data. This tag is bound to the original data and the segment mapping table, and is used throughout all subsequent stages.

[0036] In some embodiments, calculating the hash value of the heterogeneous data as a digital fingerprint based on the value marker includes: The 5G converged gateway extracts the original data blocks corresponding to each data segment from the original byte stream of the heterogeneous data according to the byte range corresponding to each data segment in the segmentation mapping table. For each of the original data blocks, calculate the content hash as the block-level fingerprint of the data block; Based on the block-level fingerprints of multiple data blocks belonging to the target data segment, a Merkle subtree of the target data segment is constructed, and the root hash of the Merkle subtree is used as the segment integrity fingerprint of the target data segment, wherein the target data segment is any one of the heterogeneous data segments; Extract the segment-level value tag corresponding to the target data segment, and calculate the hash value of the segment-level value tag as the segment value digest fingerprint of the target data segment; Using the segment integrity fingerprints and segment value digest fingerprints of all data segments as leaf nodes, construct a global Merkle tree and generate a global root hash; The global root hash, the segment integrity fingerprint of each data segment, the segment value digest fingerprint, and the segment mapping table are jointly encapsulated to generate the digital fingerprint of the heterogeneous data.

[0037] The 5G converged gateway precisely segments the corresponding original data blocks from the original byte stream based on the start and end byte offsets of each value data segment (high, medium, and low value segments) recorded in the segmentation mapping table. This operation ensures that the structure of the digital fingerprint is strictly aligned with the value hierarchy structure. For each segmented original data block, its hash value (such as SHA-256) is calculated independently to generate a block-level fingerprint for that data block. This fingerprint uniquely represents the complete content of the data block at the time of generation. Any slight modification to the content of the data block will cause a drastic change in its block-level fingerprint. To achieve dual trust assurance of content and value at the segment level, the system constructs a segment integrity fingerprint and a segment value digest fingerprint for each data segment. The segment integrity fingerprint is used to verify the integrity of the data segment's content. The system constructs a Merkle subtree using the block-level fingerprints of all data blocks contained in a data segment as leaf nodes. The root hash of this subtree is used as the segment integrity fingerprint of the data segment. The characteristics of Merkle trees dictate that to verify whether a specific data block belongs to a segment and has not been tampered with, only the path hash value from the data block to the root node needs to be provided, without transmitting the entire data segment, thus achieving efficient verifiability. The segment value digest fingerprint is used to bind and verify the value attributes of the data segment. The system extracts the segment-level value tag corresponding to the data segment (containing information such as value level and key semantic digest) and calculates its hash value as the segment value digest fingerprint. This fingerprint ensures that the value description information of the data segment itself is also immutable and is associated with the data segment content through the subsequent global structure. After obtaining the two types of fingerprints (segment integrity fingerprint and segment value digest fingerprint) for all data segments, the system uses these fingerprints as new leaf nodes to jointly construct a final global Merkle tree. The system encapsulates the global root hash, the list of segment integrity fingerprints for all data segments, the list of segment value digest fingerprints for all data segments, and the initial segmentation mapping table into a structured data packet, which is the digital fingerprint of this heterogeneous data. Because the segment value digest fingerprint is incorporated into the global Merkle tree, any attempt to tamper with the data value tag after transmission or processing (e.g., forging low-value data as high-value data to gain priority transmission) will result in the global root hash verification failing. This ensures that the value attribute of the data is as trustworthy as its content. When exchanging data within an edge agent alliance or with the cloud, only the digital fingerprint and necessary path proofs need to be transmitted, rather than all the original data, for the other party to complete the trustworthy verification of the data content and value, greatly reducing the communication overhead incurred in establishing trust.

[0038] In some embodiments, the slave agent processing different modal data interacts with event features in real time during pipeline processing, specifically including: When the first slave agent that processes time-series sensing data detects locally that the raw data value of the monitored target parameter exceeds a preset third threshold, it sends a spatiotemporal perception request packet to the master agent. The spatiotemporal perception request packet includes event features, which include a timestamp, an event type, and the physical location of the first slave agent. Based on the event type and the physical location, the main agent retrieves and wakes up multiple geographically or logically related secondary agents from the device topology and semantic relationship graph; The master agent broadcasts collaborative perception instructions to multiple second slave agents and receives confirmation signals and confidence levels returned by multiple second slave agents. The collaborative perception instructions include the event type and time window, and the confirmation signals are used to characterize the existence of collaborative verification features. If the number of confirmation signals and the weighted confidence value received by the master agent within a preset decision time exceed a preset fourth threshold, then the cross-modal collaborative perception event is determined to be established, a fusion event descriptor is generated, and the fusion event descriptor is distributed to all target second slave agents that return the confirmation signal. The fusion event descriptor includes event type, fusion timestamp, trigger source ID, list of collaborative verification source IDs, and data value confidence.

[0039] The mechanism is initiated by the local anomaly detection of an agent within the pipeline. Taking the first slave agent processing time-series data on gas concentration as an example, it continuously monitors the data stream. When it detects that the raw data value of the monitored target parameter (e.g., "CH4 concentration") exceeds a preset third threshold (e.g., the warning concentration in safety procedures), the agent determines that a local suspected event has occurred. At this point, it immediately sends a structured spatiotemporal awareness request packet to the master agent of the consortium. This request packet contains a core event feature triple: timestamp, event type, and physical location. Timestamp: The precise moment the anomaly occurred. Event type: A predefined category based on the parameter and threshold, such as "Gas concentration exceeding limits_Warning". Physical location: The deployment location of the sensor associated with the first slave agent (e.g., "Underground-01 mining area-Return airway A point"). Upon receiving the request, the master agent performs an intelligent retrieval based on a pre-built and maintained equipment topology and semantic relationship graph. This data map defines geographical and logical associations. Specifically, it identifies sensors located in the same or adjacent work areas as the triggering event sensor, whose monitoring parameters are semantically related to the current event type. For example, a "gas exceedance" event is logically associated with "area personnel positioning cameras," "ventilation equipment status sensors," and "dust concentration sensors." Based on event type and physical location, the master agent quickly retrieves multiple geographically or logically highly related second-slave agents from the data map (e.g., the agent responsible for video analysis in the area, the agent responsible for fan status monitoring). The master agent then sends wake-up commands to these specific second-slave agents, transitioning them from standby or regular processing mode to a collaborative sensing ready state. The master agent broadcasts a collaborative sensing command to all awakened second-slave agents. This command clarifies the event type requiring verification and defines a critical time window (typically centered on the trigger timestamp, extending before and after a reasonable timeframe, such as 10 seconds before to 30 seconds after the event). The instruction requires each secondary agent to immediately backtrack and analyze the data stream it is responsible for (such as video streams or device status streams), searching within a specified time window for co-corroborating features that can confirm or supplement the triggering event. Upon receiving the instruction, each secondary agent initiates rapid analysis. For example, the video analysis agent checks for visual features such as unauthorized entry, open flames, or abnormal sparks from equipment in the video stream within the corresponding time window. The device status agent checks whether ventilation equipment has abnormally stopped or experienced a sudden drop in speed within that time window. After analysis, each secondary agent returns a confirmation signal and a confidence value to the primary agent. The confirmation signal indicates that "relevant corroborating features have been found"; the confidence value quantifies the prominence of the feature or its close association with the triggering event (e.g., a high confidence value for clearly seeing someone smoking in the video, and a low confidence value for only detecting blurry movement). The primary agent starts a decision timer to collect responses from all secondary agents within a preset decision time (e.g., 200 milliseconds).The system counts the number of second agents returning confirmation signals and then sums the confidence scores of all returned signals using a weighted average (different modalities of agents can be assigned different weights, such as video evidence having a higher weight than ordinary device status). If the combined result of the number of confirmation signals and the weighted confidence score exceeds a preset fourth threshold, the master agent formally determines that a cross-modal collaborative sensing event has been established. This event has higher credibility than a single sensor alarm. After the determination is established, the master agent immediately generates a structured fusion event descriptor as the authoritative digital summary of the event and distributes it to all target second agents that returned confirmation signals. This descriptor includes: event type, a refined type, such as "Gas anomaly accompanied by personnel violation_high risk"; fusion timestamp, a comprehensive time reference based on the trigger time; trigger source ID, the identifier of the first agent; a list of collaborative verification source IDs, a list of identifiers of all second agents providing verification, clarifying the source of the evidence; and data value confidence score, a comprehensive score calculated based on the number of verifications and the weighted confidence score, which directly affects the priority of subsequent data encoding and transmission.

[0040] In some embodiments, performing cross-modal joint coding on the associated heterogeneous data streams based on the event features includes: Within the edge agent alliance, the target second slave agent parses the fused event descriptor to obtain the event type and the data value confidence, and obtains the available computing resources, prediction bandwidth and current mine production stage information of the target second slave agent in real time, and fuses them to form a dynamic context. Based on the dynamic context, a personalized encoding strategy vector is generated through a preset strategy generation model. The target second slave agent publishes a virtual resource offer within the edge agent alliance based on the resource requirements in the personalized coding strategy vector, and selects a third slave agent based on the bidding results of other slave agents within the edge agent alliance; The target second agent encodes the video stream, and the third agent locates the associated data segments from the non-video modal data stream corresponding to the list of co-verification source IDs that overlap in time or are logically related to the event features extracted from the video stream. The associated data segments are encoded using encoding parameters higher than a first quality threshold, and the unassociated data segments are compressed using encoding parameters lower than a second quality threshold, where the second quality threshold is lower than the first quality threshold. The video data encoded by the second agent and the non-video data encoded or compressed by the third agent are bound and encapsulated according to a unified spatiotemporal index to obtain a multimodal data unit.

[0041] Once the second target agent receives the fused event descriptor distributed by the main agent (i.e., the agent participating in collaborative verification and processing data of this modality, such as a video analytics agent), the joint encoding process officially begins. The second target agent parses the descriptor and extracts two core decision inputs: event type and data value confidence level. For example, the event type is "gas anomaly accompanied by personnel violation," and the confidence level is "0.92" (high confidence). The second target agent collects its own dynamic context information in real time, including: available computing resources, which can be the current idle computing power of CPU, GPU, and memory; predicted bandwidth, which can be the recent uplink network bandwidth predicted based on historical data and base station feedback; and information on the current mine production stage, such as "blasting preparation period," "normal mining period," or "maintenance period," as different stages have different requirements for data real-time performance and completeness. After fusing these dynamic contexts with the event type and value confidence level, the data is input into a pre-defined policy generation model (this model can be a rule-based expert system or a lightweight machine learning model). This model, after training or configuration, can output a personalized encoding strategy vector based on the "type / confidence of the event to be processed" and the "current conditions (resources / network / stage)". This vector is a set of executable parameter instructions, for example, for video streams: instructing high frame rate, high resolution H.265 encoding for the "core event period" (e.g., 30 seconds before and after a warning); low frame rate, standard resolution encoding for the "non-core period"; and intelligent cropping or further compression of background areas. The personalized encoding strategy vector also includes the estimated computational resources required to execute this encoding strategy and the expected output bitrate. If the resource requirements assessed by the personalized encoding strategy vector exceed the capabilities of the current target agent, it will initiate a resource coordination mechanism within the alliance. The agent publishes a virtual resource offer within the alliance, explicitly listing the types of tasks to be subcontracted (e.g., high-quality encoding of associated segments in non-video modal data streams), the amount of computational resources required, and the "virtual cost" (a metric used internally by the system to adjust the load) it is willing to pay for using these resources. After receiving the offer, other idle or lightly loaded agents within the alliance (candidate third agents) submit bids based on their own circumstances (i.e., the amount of resources they can provide and the corresponding costs). The target second agent comprehensively considers the bidders' capabilities, reliability, and costs, and selects the optimal third agent as a partner, outsourcing some encoding tasks (usually the associated encoding of non-video modal data). This mechanism realizes secondary dynamic scheduling and load balancing of edge computing resources. After the task division is clear, joint encoding is carried out in parallel: (1) The target second agent performs video stream encoding. The target second agent encodes the video stream it is responsible for based on the policy vector. It uses the timestamps of event features to accurately locate event-related segments in the video stream (i.e., video segments that coincide with gas overruns and personnel violations in time).For associated segments, encoding parameters higher than the first quality threshold are used, such as lower quantization parameters and higher frame rates, to ensure high fidelity of key visual information of the event (such as details of personnel behavior and equipment status). For unassociated segments (regular monitoring footage before and after the event), encoding parameters higher than the second quality threshold are used for compression (second threshold < first threshold), such as higher quantization parameters, lower frame rates, or even intelligent frame extraction, to significantly reduce the bit rate with acceptable quality loss. (2) Non-video modal data encoding is performed by the third slave agent. The third slave agent finds the associated non-video data streams (such as specific wind speed sensors and equipment current data streams) under its responsibility according to the list of co-verification source IDs in the fusion event descriptor. It locates data segments that are precisely overlapped in time or strongly correlated in logic with video event segments from these data streams. For example, it locates wind speed data and equipment current fluctuation data during the period when gas concentration rises sharply. For these associated data segments, encoding parameters higher than the first quality threshold are also used (in non-video data, this may be reflected in higher sampling rates, lower compression rates, or lossless storage) to ensure the accuracy of key sensor data. For non-associated regular data segments, high-ratio compression is performed using encoding parameters higher than the second quality threshold, such as lossy compression or retaining only statistical features. After encoding, the system establishes a unified spatiotemporal index, including a temporal index and a spatial / logical index. The temporal index aligns the start and end times of all modal-related segments based on the fusion timestamp in the fusion event descriptor. The spatial / logical index establishes spatial or logical associations between data segments based on event location and participating device IDs. The encoded video data segments, encoded non-video data segments, and this unified spatiotemporal index are then bound and encapsulated to form a self-describing multimodal data unit.

[0042] Figure 2 This is a schematic diagram of the process for generating a transmission request suggestion packet in an embodiment of this application, such as... Figure 2 As shown, the step of predicting the transmission requirements of the data packets to be uploaded based on the locally stored historical task logs of the edge intelligence alliance, and generating a transmission request suggestion packet containing predicted traffic, priority, and suggested transmission parameters, specifically includes: S201. The traffic baseline is obtained by performing time-series pattern analysis on the historical task logs through the edge agent alliance. S202. Obtain the number of currently active slave agents in the edge agent alliance, the current computing load of each slave agent, and the data value confidence carried by the fusion event descriptor. Map the data value confidence to a transmission priority weight, and calculate the real-time collaborative processing capability coefficient by combining the number of currently active slave agents and the computing load. Use the real-time collaborative processing capability coefficient to correct the traffic baseline and obtain the predicted traffic. S203. The predicted traffic, the transmission priority weight, and the current average network latency and packet loss rate are used as query features and input into the transmission strategy knowledge base of the edge agent alliance to match the suggested transmission parameter set. The transmission parameter set includes coding redundancy, maximum retransmission count and first packet transmission latency. The transmission strategy knowledge base stores a variety of preset network states and task features to the optimal transmission parameter set. S204. Assemble the predicted traffic, the transmission priority weight, and the transmission parameter set to generate the transmission request suggestion packet.

[0043] The edge intelligence alliance continuously stores historical task logs locally. These logs not only record the amount of data but also associate information such as task type, processing results (e.g., whether an event was generated), and the time period of occurrence. By performing time-series pattern analysis on these logs (e.g., using sliding window averaging, seasonal decomposition, or lightweight time-series prediction models), the alliance can learn the patterns of data production in its region, thereby generating a traffic baseline. For example, the system may discover periodic data traffic peaks during the daily morning production rush (e.g., 8:00-10:00) and after equipment inspections (e.g., 15:00-16:00) due to active operations and concentrated equipment status reporting. This traffic baseline represents the expected normal data output under normal circumstances and serves as the starting point for transmission demand prediction. Actual transmission demand is significantly influenced by the complexity of real-time tasks. Therefore, the system introduces multi-dimensional real-time status information to dynamically correct the static traffic baseline. Task value weight input: Extracting data value confidence (a value between 0 and 1, representing the importance and reliability of the event) from the currently processed fusion event descriptor. This confidence is then converted into transmission priority weights through a pre-defined mapping function. High-confidence events (such as confirmed severe gas exceedances) are mapped to high-priority weights, directly indicating that their data packets should occupy a priority position in the transmission queue. The system obtains in real time the number of currently active slave agents (reflecting the overall processing scale of the consortium) and the current computing load of each slave agent (reflecting the busy / idleness level of the consortium). Combining these two indicators, a real-time collaborative processing capability coefficient is calculated. A high coefficient indicates that the consortium has sufficient computing power and can quickly complete complex data processing, possibly generating data to be transmitted in advance or in a concentrated manner; a low coefficient indicates that processing may be lagging and the data output rate will decrease. The real-time collaborative processing capability coefficient is used as a correction factor on the traffic baseline. The logic is: the stronger the processing capability, the more data can be completed and prepared for upload per unit time (positive correction); conversely, it may decrease (negative correction). At the same time, the high-priority weight itself also implies the importance of the data, which may indirectly affect the estimation of the traffic scale. Through correction, the static baseline is transformed into a dynamic predicted traffic reflecting the current actual production capacity. The system uses the predicted traffic, transmission priority weights, and the real-time monitored current average network latency and packet loss rate to form a multi-dimensional query feature vector. This vector is input into a locally maintained transmission strategy knowledge base for query matching. This knowledge base stores rule bases or experience bases that map various combinations of "network state-task characteristics" to "optimal transmission parameter sets". Example of the matching process: Querying features {predicted traffic: medium, priority: high, network latency: low, packet loss rate: low}, the knowledge base might match the parameter set: {encoding redundancy: low, maximum retransmissions: 2, first packet transmission latency: 0ms}. This means that when the network is good, high-priority data can be more aggressive, reducing redundancy and retransmissions to pursue low latency.Conversely, if the query characteristics are {Predicted Traffic: High, Priority: Medium, Network Latency: High, Packet Loss Rate: High}, then the system may match: {Coding Redundancy: High (increase FEC), Maximum Retransmissions: 5, First Packet Sending Delay: 50ms (slight buffering to avoid exacerbating congestion)}, to ensure reliability under poor network conditions. The system assembles the predicted traffic, transmission priority weights, and the matched set of suggested transmission parameters (including specific executable parameters such as coding redundancy, maximum retransmissions, and first packet sending delay) into a uniformly formatted transmission request suggestion packet.

[0044] In some embodiments, the method further includes: If the 5G base station receives multiple transmission request suggestion packets from different edge agent alliances in the same region within the same scheduling period, the 5G base station sorts the multiple transmission request suggestion packets according to the transmission priority weight in each transmission request suggestion packet and the first packet transmission delay, and performs preliminary non-overlapping window pre-allocation on the time-frequency resource grid based on the resources required by each transmission request suggestion packet to form multiple candidate transmission window schemes. The 5G base station will feed back the candidate window scheme, which includes the window start time, window duration, and the guaranteed bandwidth corresponding to the window, to the corresponding edge agent alliance. If the corresponding edge agent alliance confirms acceptance of the candidate window scheme, the 5G base station locks and reserves the corresponding radio resource block on the time-frequency resource grid before the start time of the window, and sends a transmission window confirmation signal to the corresponding edge agent alliance. The transmission window confirmation signal includes window parameters and a unique transaction identifier for this reservation, which is used for access verification during subsequent data upload.

[0045] Within a scheduling cycle (e.g., every 100 milliseconds), a 5G base station receives transmission request suggestion packets from multiple edge agent consortia within its coverage area. Each suggestion packet contains key information such as the consortium's predicted traffic, transmission priority weight, and first packet transmission latency suggestion. The base station first performs a global sorting, primarily based on transmission priority weight and first packet transmission latency. Transmission priority weight: directly reflects the importance of the data. A request from a consortium handling a "gas explosion risk" event will necessarily have a higher priority weight than a consortium handling "routine environmental monitoring." First packet transmission latency: reflects the urgency of the task. A request suggesting low latency implies extremely high real-time requirements. The base station sorts all requests based on these two core dimensions (usually with priority weight as the primary factor and latency requirement as a secondary factor), forming a global transmission queue. Requests at the front of the queue receive priority scheduling rights. As a resource manager, the base station needs to translate the abstract queue into concrete resource allocation. It calculates the approximate number of radio resource blocks required to satisfy the transmission needs of each request suggestion packet based on the predicted traffic. Then, on the abstract time-frequency resource grid (representing available time and frequency), an initial non-overlapping window pre-allocation is attempted for the requests in the queue. Non-overlapping is key to avoiding interference and ensuring performance. The base station uses intelligent algorithms to allocate non-conflicting transmission windows for different requests in both time and frequency dimensions. The pre-allocation forms candidate transmission window schemes, each precisely describing: start time, duration (window length), and provided bandwidth (guaranteed bandwidth). After completing the initial pre-allocation, the base station does not directly enforce it. Instead, it feeds back the calculated candidate transmission window schemes to the corresponding edge agent consortium. This step is crucial; it establishes a negotiation mechanism rather than a command mechanism. Upon receiving a candidate scheme, the edge agent consortium evaluates it: checking if the window start time meets its data readiness time and real-time requirements, confirming if the guaranteed bandwidth is sufficient to complete data upload within its predicted window length, and assessing whether the scheme is superior to other transmission opportunities it predicts. If the consortium accepts the scheme after evaluation, it sends an acknowledgment signal to the base station. This design grants the edge a degree of autonomy, enabling it to participate in scheduling decisions based on its own processing realities (such as unexpected delays in complex coding tasks), enhancing the system's flexibility and adaptability. When a base station receives an acceptance signal from a consortium for a candidate scheduling scheme, that scheme is upgraded from candidate to reserved. The base station immediately performs resource locking and reservation operations on the time-frequency resource grid for the upcoming transmission window. Locking: This means marking these resource blocks (specific time and frequency ranges) as occupied within the system for a future period, preventing them from participating in scheduling calculations for any other requests. Reservation: This is a preparatory action at the physical or MAC layer, ensuring that the network side is fully prepared to receive data when the window arrives, eliminating access contention.After resource reservation is complete, the base station sends a final transmission window confirmation signal to the edge agent alliance. This signal not only includes the final parameters of the window (start time, duration, bandwidth), but more importantly, it includes a unique transaction identifier for this reservation. The unique transaction identifier is a digital token for this reserved transmission; it is globally unique and used for access verification during subsequent data transmission. When the reserved transmission window start time arrives, the edge agent alliance begins uploading data. When initiating an access request, it must present the unique transaction identifier issued by the base station. The base station verifies that the identifier is valid and matches the current time window before allowing access and starting data transmission. Other nodes without a valid identifier or whose identifiers do not match will be denied access within this window, thus perfectly avoiding resource conflicts.

[0046] The above describes the mine multi-source data fusion and transmission method based on distributed edge agent collaboration in the embodiments of this application. The computer system in the embodiments of this application will be described in detail below in conjunction with the above-mentioned mine multi-source data fusion and transmission method based on distributed edge agent collaboration.

[0047] Please see Figure 3 This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application.

[0048] In some embodiments, the computer system 300 includes a computer device, which may be a terminal device. The computer device includes a processor 301, a memory 302, a sensor module 303, a communication module 304, an input device 305, and an output device 306 connected via a system bus. The processor 301 of the computer device provides computing and control capabilities. The memory 302 of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database is used to store data.

[0049] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0050] In some embodiments of this application, a computer-readable storage medium is provided, including instructions that, when executed on the computer system 300, cause the computer system 300 to execute the mining multi-source data fusion and transmission method based on distributed edge agent collaboration as described in this application.

[0051] In some embodiments of this application, a computer program product is also provided. When the computer program product is run on a computer system 300, the computer system 300 executes the mining multi-source data fusion and transmission method based on distributed edge intelligent agent collaboration in the embodiments of this application.

[0052] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0053] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0054] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for multi-source data fusion and transmission in mines based on distributed edge agent collaboration, characterized in that, include: Heterogeneous data is collected by multiple data acquisition terminals deployed in the mining production environment. Based on the device type of the data source and the preset value rule base, a value tag is generated for the heterogeneous data through a 5G converged gateway. The hash value of the heterogeneous data is calculated as a digital fingerprint based on the value tag. Data credentials are generated based on the digital fingerprint and broadcast to a distributed ledger jointly maintained by multiple edge computing nodes deployed in the same area. The first edge computing node that verifies and records the data credentials in the distributed ledger is determined as the master agent. The master agent parses the value tag and splits the data processing task into task sub-slices of different processing types. It then broadcasts bidding information containing the resource requirements of each task sub-slice to other edge computing nodes. From the bidding results of the other edge computing nodes for the bidding information, slave agents are selected. The master agent and the slave agents are constructed into an edge agent alliance to process the heterogeneous data. Within the edge agent alliance, each slave agent constructs a data processing pipeline according to the logical order of the task sub-slices. During the pipeline processing, slave agents processing different modal data interact in real time with event features and perform cross-modal joint coding on the associated heterogeneous data streams based on the event features. Based on the historical task logs stored locally, the edge intelligence alliance predicts the transmission requirements of the data packets to be uploaded, generates a transmission request suggestion packet containing predicted traffic, priority and suggested transmission parameters, and sends it to the 5G base station. It receives the transmission window returned by the 5G base station and aggregates and uploads the jointly encoded data to the cloud within the transmission window.

2. The method according to claim 1, characterized in that, The process of generating value tags for heterogeneous data based on device type and a preset value rule base through a 5G converged gateway includes: The 5G converged gateway parses the received heterogeneous data, identifies the device type and data type, and extracts the semantic units corresponding to the data type. Based on the device type, the content of the semantic unit, and the frequency with which historical similar data is called and used by the cloud analysis model to generate decisions, the unit value density score of the semantic unit is obtained. The semantic units whose unit value density scores are higher than a preset first threshold are aggregated into high-value data segments in the heterogeneous data, the semantic units whose unit value density scores are lower than a preset second threshold are aggregated into low-value data segments, and the rest are medium-value data segments. A segment-level value tag is generated for each data segment, and a segment mapping table is generated based on the segment-level value tag. Integrate all segment-level value tags to generate a global value tag for the heterogeneous data.

3. The method according to claim 2, characterized in that, The step of calculating the hash value of the heterogeneous data as a digital fingerprint based on the value marker includes: The 5G converged gateway extracts the original data blocks corresponding to each data segment from the original byte stream of the heterogeneous data according to the byte range corresponding to each data segment in the segmentation mapping table. For each of the original data blocks, calculate the content hash as the block-level fingerprint of the data block; Based on the block-level fingerprints of multiple data blocks belonging to the target data segment, a Merkle subtree of the target data segment is constructed, and the root hash of the Merkle subtree is used as the segment integrity fingerprint of the target data segment, wherein the target data segment is any one of the heterogeneous data segments; Extract the segment-level value tag corresponding to the target data segment, and calculate the hash value of the segment-level value tag as the segment value digest fingerprint of the target data segment; Using the segment integrity fingerprints and segment value digest fingerprints of all data segments as leaf nodes, construct a global Merkle tree and generate a global root hash; The global root hash, the segment integrity fingerprint of each data segment, the segment value digest fingerprint, and the segment mapping table are jointly encapsulated to generate the digital fingerprint of the heterogeneous data.

4. The method according to claim 1, characterized in that, The agent processing different modalities of data interacts with event features in real time during pipeline processing, specifically including: When the first slave agent that processes time-series sensing data detects locally that the raw data value of the monitored target parameter exceeds a preset third threshold, it sends a spatiotemporal perception request packet to the master agent. The spatiotemporal perception request packet includes event features, which include a timestamp, an event type, and the physical location of the first slave agent. Based on the event type and the physical location, the main agent retrieves and wakes up multiple geographically or logically related secondary agents from the device topology and semantic relationship graph; The master agent broadcasts collaborative perception instructions to multiple second slave agents and receives confirmation signals and confidence levels returned by multiple second slave agents. The collaborative perception instructions include the event type and time window, and the confirmation signals are used to characterize the existence of collaborative verification features. If the number of confirmation signals and the weighted confidence value received by the master agent within a preset decision time exceed a preset fourth threshold, then the cross-modal collaborative perception event is determined to be established, a fusion event descriptor is generated, and the fusion event descriptor is distributed to all target second slave agents that return the confirmation signal. The fusion event descriptor includes event type, fusion timestamp, trigger source ID, list of collaborative verification source IDs, and data value confidence.

5. The method according to claim 4, characterized in that, The step of performing cross-modal joint coding on the associated heterogeneous data streams based on the event features includes: Within the edge agent alliance, the target second slave agent parses the fused event descriptor to obtain the event type and the data value confidence, and obtains the available computing resources, prediction bandwidth and current mine production stage information of the target second slave agent in real time, and fuses them to form a dynamic context. Based on the dynamic context, a personalized encoding strategy vector is generated through a preset strategy generation model. The target second slave agent publishes a virtual resource offer within the edge agent alliance based on the resource requirements in the personalized coding strategy vector, and selects a third slave agent based on the bidding results of other slave agents within the edge agent alliance; The target second agent encodes the video stream, and the third agent locates the associated data segments from the non-video modal data stream corresponding to the list of co-verification source IDs that overlap in time or are logically related to the event features extracted from the video stream. The associated data segments are encoded using encoding parameters higher than a first quality threshold, and the unassociated data segments are compressed using encoding parameters lower than a second quality threshold, where the second quality threshold is lower than the first quality threshold. The video data encoded by the second agent and the non-video data encoded or compressed by the third agent are bound and encapsulated according to a unified spatiotemporal index to obtain a multimodal data unit.

6. The method according to claim 4, characterized in that, The process involves using the edge intelligence alliance to predict the transmission requirements of the data packets to be uploaded based on locally stored historical task logs, and generating a transmission request suggestion packet containing predicted traffic, priority, and suggested transmission parameters. Specifically, this includes: The traffic baseline is obtained by performing time-series pattern analysis on the historical task logs through the edge agent alliance. The number of currently active slave agents in the edge agent alliance, the current computing load of each slave agent, and the data value confidence carried by the fusion event descriptor are obtained. The data value confidence is mapped to a transmission priority weight. The real-time collaborative processing capability coefficient is calculated by combining the number of currently active slave agents and the computing load. The traffic baseline is corrected using the real-time collaborative processing capability coefficient to obtain the predicted traffic. The predicted traffic, the transmission priority weight, and the current average network latency and packet loss rate are used as query features and input into the transmission strategy knowledge base of the edge agent alliance to match a suggested transmission parameter set. The transmission parameter set includes coding redundancy, maximum retransmission count, and first packet transmission latency. The transmission strategy knowledge base stores a variety of preset mapping relationships between network states and task features and the optimal transmission parameter set. The predicted traffic, the transmission priority weight, and the transmission parameter set are assembled to generate the transmission request suggestion packet.

7. The method according to claim 6, characterized in that, The method further includes: If the 5G base station receives multiple transmission request suggestion packets from different edge agent alliances in the same region within the same scheduling period, the 5G base station sorts the multiple transmission request suggestion packets according to the transmission priority weight in each transmission request suggestion packet and the first packet transmission delay, and performs preliminary non-overlapping window pre-allocation on the time-frequency resource grid based on the resources required by each transmission request suggestion packet to form multiple candidate transmission window schemes. The 5G base station will feed back the candidate window scheme, which includes the window start time, window duration, and the guaranteed bandwidth corresponding to the window, to the corresponding edge agent alliance. If the corresponding edge agent alliance confirms acceptance of the candidate window scheme, the 5G base station locks and reserves the corresponding radio resource block on the time-frequency resource grid before the start time of the window, and sends a transmission window confirmation signal to the corresponding edge agent alliance. The transmission window confirmation signal includes window parameters and a unique transaction identifier for this reservation, which is used for access verification during subsequent data upload.

8. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

Images