Mining multi-agent collaborative operation method and device

Abstract

The invention relates to the technical field of mining multi-agent collaborative operation, and provides a mining multi-agent collaborative operation method and device, and the method comprises the steps: receiving a mining task; the mining task is analyzed, the cooperation type of an intelligent agent is determined, and the cooperation type comprises a single type and a mixed type; and an agent cooperation strategy is generated according to the cooperation type and the historical cooperation score of the agent, and the agent cooperation strategy at least comprises a mining basic cooperation framework, an agent role weight and a cooperation strategy parameter. According to the method, multiple cooperation types of the coal mine can be fused, the agent cooperation strategy containing the mining basic cooperation framework, the agent role weight and the cooperation strategy parameters is generated, interaction between the agents is optimized, the underground dynamic task requirement is met, resource conflicts are avoided, the cooperation efficiency and flexibility of the underground agents are improved, and the underground operation safety is improved.

Description

Technical Field

[0001] This application relates to the field of multi-agent collaborative operation technology in mining, specifically to a method and apparatus for multi-agent collaborative operation in mining. Background Technology

[0002] The current application of multi-agent systems in underground coal mines has obvious adaptability problems: In terms of cooperation mechanisms, the cooperation is simple and usually adopts static or predefined cooperation strategies, which is difficult to adapt to the dynamic task requirements underground and is prone to task conflicts or inefficiency. For example, inspection robots and transportation robots often have task conflicts and lack effective global coordination; the cooperation mechanism that relies on centralized orchestrators sacrifices system flexibility and cannot cope with the sudden changes in underground roadways. Summary of the Invention

[0003] The purpose of this application is to provide a method and apparatus for multi-agent collaborative operation in mining, so as to at least solve the technical problems of single collaboration, easy resource conflict or low collaboration efficiency and poor collaboration flexibility in underground coal mine collaboration scenarios.

[0004] To solve the above-mentioned technical problems, the embodiments of this application adopt the following technical solutions:

[0005] In a first aspect, embodiments of this application provide a multi-agent collaborative operation method for mining, including:

[0006] Receive mining tasks;

[0007] The mining task is analyzed to determine the collaboration type of the agent, wherein the collaboration type includes single type and mixed type;

[0008] An agent collaboration strategy is generated based on the collaboration type and the agent's historical collaboration scores. The agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

[0009] In some embodiments, the mining task is parsed to determine the agent's cooperation type, including:

[0010] The text of the mining task is parsed using the BERT-base model to generate collaboration type tags.

[0011] In some embodiments, an agent cooperation strategy is generated based on the cooperation type and the agent's historical cooperation scores, including:

[0012] The basic collaboration framework is determined through the rules layer;

[0013] Assign agent role weights to downhole agents through a role layer;

[0014] The parameters of the cooperative strategy are determined by using a proximal policy optimization algorithm at the model layer.

[0015] In some embodiments, the method further includes:

[0016] Obtain training metrics for the downhole agent;

[0017] The suitability of the agent is determined based on the training metrics.

[0018] If the agent is a defective agent, a corresponding training sample is generated based on the defective agent.

[0019] In some embodiments, the method further includes:

[0020] Mining collaboration experience is screened using a cosine similarity algorithm, and then pushed to intelligent agents with the same role.

[0021] In some embodiments, the method further includes:

[0022] Extract point cloud features of the downhole embodied environment;

[0023] Adjacent point clouds are grouped to obtain grouped features;

[0024] The grouping features are matched with the agent collaboration strategy to determine whether the agent collaboration strategy is suitable for the work scenario.

[0025] If a match is found, the agent collaboration strategy is executed; if a match is not found, the agent collaboration strategy is replanned using a preset collaborative operation model.

[0026] In some embodiments, the method further includes:

[0027] Based on the initial model of the intelligent agents and the task requirements, multiple intelligent agents in the mine are clustered and divided into core clusters and edge clusters.

[0028] The communication method of the intelligent agent is determined according to the classified cluster type, wherein the core cluster communicates via optical fiber and the edge cluster communicates via intrinsically safe wireless communication;

[0029] The cooperation strategy parameters are transmitted to the corresponding agent according to the communication method.

[0030] In some embodiments, the method further includes:

[0031] Constructing a knowledge graph of coal mines;

[0032] By pre-training the coal mine knowledge graph data in the coal mine knowledge graph through the backbone network, a general collaborative knowledge model for coal mines is obtained.

[0033] Based on the mining branch network, the general collaborative knowledge model of coal mines is fine-tuned to obtain a coal mine collaborative knowledge model applicable to preset operation scenarios.

[0034] In some embodiments, the method further includes:

[0035] Construct decision trees for high-frequency scenarios;

[0036] When the well is in a preset operating condition, the processing strategy for the preset operating condition is determined based on the high-frequency scenario decision tree.

[0037] Secondly, embodiments of this application provide a mining multi-agent collaborative operation device, comprising:

[0038] The receiving module is configured to receive mining tasks.

[0039] The parsing module is configured to parse the mining task and determine the collaboration type of the agent, wherein the collaboration type includes a single type and a mixed type;

[0040] The collaboration strategy generation module is configured to generate an agent collaboration strategy based on the collaboration type and the agent's historical collaboration scores. The agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

[0041] Thirdly, embodiments of this application provide an electronic device, including at least a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned multi-agent cooperative operation method for mining when executing the computer program in the memory.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned multi-agent cooperative operation method for mining.

[0043] This application provides a method and apparatus for multi-agent collaborative operation in mining. The method involves receiving mining tasks; parsing the mining tasks to determine the collaboration type of the agents, where the collaboration type includes single type and mixed type; and generating an agent collaboration strategy based on the collaboration type and the agents' historical collaboration scores. The agent collaboration strategy includes at least a basic mining collaboration framework, agent role weights, and collaboration strategy parameters. This approach can integrate multiple collaboration types in coal mines to generate an agent collaboration strategy that includes a basic mining collaboration framework, agent role weights, and collaboration strategy parameters. This optimizes the interaction between agents, adapts to dynamic underground task requirements, avoids resource conflicts, improves the efficiency and flexibility of underground agent collaboration, and enhances the safety of underground operations. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of a multi-agent collaborative operation method for mining, as described in an embodiment of this application.

[0046] Figure 2 This is another flowchart of the mining multi-agent cooperative operation method according to an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the structure of a mining multi-agent collaborative operation device according to an embodiment of this application. Detailed Implementation

[0048] Various embodiments and features of this application are described herein with reference to the accompanying drawings.

[0049] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.

[0050] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0051] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0052] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0053] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0054] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.

[0055] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0056] In the current development of artificial intelligence technology, multi-agent systems (MAS), as a distributed AI approach, have made significant progress in multiple fields. However, in the application scenario of multi-agent systems in underground coal mines, existing systems have the following core limitations:

[0057] (1) Problem of single collaboration: Existing systems usually adopt static or predefined collaboration strategies, which are difficult to adapt to the dynamic task requirements in the mine, such as the mixed scenario of "gas monitoring + coal transfer", which can easily lead to task conflicts or low efficiency.

[0058] (2) Inefficient communication problem: The communication architecture of traditional multi-agent systems lacks flexibility, does not take into account the characteristics of underground radio wave attenuation and dust interference, and cannot dynamically adjust the communication method according to the role of the agent (such as core monitoring node and edge transfer node), resulting in delay in the transmission of key information and potential safety hazards.

[0059] (3) High resource requirements: The existing training system adopts static resource allocation, which is not adapted to the computing power limitations of underground edge devices. The model size is too large (usually ≥2GB) and cannot be deployed. Furthermore, the training process does not integrate coal mine knowledge graphs, requiring repeated training to adapt to different roadway scenarios, resulting in serious resource waste.

[0060] (4) Insufficient scene adaptability: The existing system lacks an underground embodied knowledge alignment mechanism, cannot process point cloud data under low light conditions, and cannot design a fast response strategy for high-frequency mining scenarios (such as sudden gas surge, equipment jamming, roadway deformation, etc.), resulting in high scene adaptation costs and poor security.

[0061] To address the aforementioned technical problems, some improvements have been explored, but shortcomings still exist:

[0062] (1) In terms of collaboration mechanism, some studies have tried to optimize the collaboration strategy between agents by introducing reinforcement learning, which has improved the collaboration efficiency to a certain extent; however, it requires complex reward function design and only supports a single type of collaboration, which cannot cope with mixed tasks.

[0063] (2) In terms of communication architecture, research on elastic communication mechanisms is gradually emerging, such as dynamic bandwidth allocation models, which can dynamically adjust communication resources according to task priority. However, these methods mainly focus on the allocation of communication resources and do not fully consider the impact of agent roles and task types on communication strategies.

[0064] (3) Regarding the optimization of the training system, some studies compress mining models through knowledge distillation, but do not combine them with coal mine knowledge graphs (such as equipment failure association rules), resulting in poor model generalization ability and requiring a large amount of underground data for retraining.

[0065] (4) Regarding scenario adaptability, research on embodied intelligence has provided new ideas for solving the scenario adaptability problem. However, the application of existing methods in multi-agent collaborative scenarios is still in its early stages. Some studies have introduced lidar to improve downhole environmental perception, but have not achieved dynamic matching between point cloud features and collaborative strategies, and have not designed hardware acceleration schemes for high-frequency safety scenarios. The response latency still does not meet the downhole safety requirements.

[0066] In summary, existing underground agent collaboration scenarios suffer from simplistic collaboration, prone to resource conflicts and low efficiency, and lack flexibility. The agent training architecture typically treats multi-agent tasks as single-agent tasks, failing to fully consider the interaction characteristics between underground agents (such as explosion-proof inspection robots and transport robots). This results in a framework with limited versatility and an inability to support specific tasks like mine equipment fault diagnosis and gas early warning. Furthermore, the system's knowledge transfer capabilities are weak. In complex underground environments such as low light, high dust levels, and signal attenuation, the model often needs to be retrained before it can be used. In high-frequency, sudden scenarios like gas surges and equipment jams, response latency can exceed 50 milliseconds, making it difficult to simultaneously meet the demands of collaboration efficiency, resource utilization, and underground safety.

[0067] In view of this, embodiments of this application propose a method and apparatus for multi-agent collaborative operation in mining.

[0068] Example 1

[0069] Figure 1 A flowchart of a mining multi-agent cooperative operation method according to an embodiment of this application is shown, such as... Figure 1 As shown in the embodiment of this application, a multi-agent cooperative operation method for mining includes:

[0070] S101: Receive mining tasks.

[0071] The multi-agent collaborative operation method for mining is applied to a multi-agent collaborative operation system for mining (hereinafter referred to as the system), which includes intelligent agent devices, computing devices, network devices, and auxiliary devices. The intelligent agent devices include embodied intelligent agents, such as mining explosion-proof inspection robots and mining explosion-proof transport robots. The mining explosion-proof inspection robot is equipped with a dust-covered depth camera (e.g., resolution 1280×720, frame rate 30fps) and dust-resistant IMU sensors (e.g., acceleration measurement range ±16g, angular velocity measurement range ±2000° / s), used for underground inspection. The mining explosion-proof transport robot (e.g., load capacity 500kg, maximum speed 1m / s) is used for underground gas monitoring, equipment inspection, and coal transfer tasks. Preferably, the communication distance of the mining explosion-proof transport robot is ≤30m. The intelligent agent devices may also include virtual intelligent agents, which are generated through software simulation and can be deployed on cloud servers and edge node devices. Their number can be dynamically expanded according to underground tasks, used to assist the embodied intelligent agents in data simulation and strategy pre-playing.

[0072] The computing equipment is the computing processing equipment for the multi-agent collaborative operation method in mining, including a cloud server, edge node devices, and terminals. The cloud server is used to run training tasks for the mining backbone network and core cluster (working face control nodes), and it is preferably a mining explosion-proof server, for example, with a protection level of IP65 and an operating temperature range of -20℃ to 40℃. The edge node devices are used for offloading training tasks of the underground edge cluster, and are preferably intrinsically safe mining edge devices with a protection level of IP67. The terminal computer is deployed in the ground control center (ground dispatch center) for operators to control underground agents, monitor data, and issue tasks.

[0073] Network equipment includes switches and wireless routers. The switches can be mine-use explosion-proof gigabit Ethernet switches to ensure high-speed communication between the core cluster and edge nodes; the wireless routers can be mine-use intrinsically safe gigabit wireless routers for wireless communication of underground edge intelligent agents (such as inspection robots and transport robots).

[0074] The auxiliary equipment includes an FPGA acceleration chip and a display. In this embodiment, the FPGA acceleration chip is used for high-frequency scene decision processing in real-time response enhancement technology. Preferably, its response latency is ≤20ms. The display can be deployed at the ground dispatch center to display information such as the operating status and training data of the downhole intelligent agent, downhole gas concentration data, and equipment fault warnings.

[0075] The cloud server receives mining tasks from the ground control center, such as "coordinating the monitoring of gas concentration, inspection of scraper conveyors, and coal transfer in a 3,000-meter longwall mining roadway".

[0076] Before implementing the above methods, the intelligent agent is first deployed. The embodied intelligent agent is deployed in the target area underground, such as the mining face or a 3000-meter tunnel, and connected to the intrinsically safe power supply and communication module to ensure the stability of test communication and the accuracy of sensors. Virtual intelligent agent software can be deployed on the cloud server and edge node devices to realize the deployment of virtual intelligent agents and configure the basic parameters of the intelligent agent (such as name, number, initial capability value, etc.).

[0077] S102: Analyze the mining task to determine the collaboration type of the agent.

[0078] After receiving mining tasks, the cloud server analyzes them and determines the type of collaboration. Collaboration types include single and mixed types. A single type can be either competitive or collaborative, while a mixed type can involve both competition and collaboration simultaneously. For example, the task of "collaboratively completing gas concentration monitoring, scraper conveyor inspection, and coal transfer in a 3000-meter longwall roadway" includes a mixed type consisting of cooperation in gas monitoring and competition for transfer resources.

[0079] In some embodiments, step S102 involves parsing the mining task to determine the agent's collaboration type, including:

[0080] S1021: The text of the mining task is parsed using the BERT-base model to generate collaboration type tags.

[0081] In this step, the BERT-base model is used to parse the downhole task text and generate collaboration type tags. Collaboration type tags can include three types: competition, cooperation, and hybrid. For example, 0 represents the competition type, mainly used for resource scheduling of transport robots; 1 represents the cooperation type, used for zone monitoring by inspection robots; and 2 represents the hybrid type.

[0082] The specific steps may include: word segmentation → vector representation → fully connected layer classification → selection of the type with the highest probability.

[0083] BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model based on the Transformer architecture, understanding text semantics through bidirectional context. In this embodiment, a pre-trained BERT-base model is used to segment the task text, which is then input into the BERT-base model to obtain a vector representation of the task text. This representation is then classified using a fully connected layer and a softmax function, outputting a probability distribution of collaboration types. The type with the highest probability is taken as the final collaboration type label; for example, the collaboration type label for the mining task mentioned above is 2. The model uses a cross-entropy loss function during training, with a learning rate set to 2e. -5 The training consists of 3 rounds.

[0084] S103: Generate an agent collaboration strategy based on the collaboration type and the agent's historical collaboration scores, wherein the agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

[0085] After determining the collaboration type of the underground agents, a collaboration strategy is generated based on the collaboration type and the agents' historical collaboration scores. The historical collaboration scores include the agents' historical collaboration accuracy rate of 92% and the load efficiency of each agent. In this embodiment, dynamic generation of hybrid collaboration types for mining can be achieved, adapting to multi-task requirements such as "monitoring + transportation," thus improving collaboration efficiency and flexibility.

[0086] In some embodiments, step S103, generating an agent cooperation strategy based on the cooperation type and the agent's historical cooperation scores, includes:

[0087] S1031: Determine the basic collaboration framework through the rule layer;

[0088] S1032: Assign agent role weights to downhole agents through the role layer;

[0089] S1033: The parameters of the cooperation strategy are determined by using the PPO algorithm through the model layer.

[0090] In this embodiment, a ternary strategy fusion is adopted to generate specific collaboration strategies based on collaboration type labels and historical collaboration scores of the agents. The ternary strategy fusion includes a rule layer, a role layer, and a model layer. The rule layer is used to determine the basic collaboration framework for mining. For example, the basic collaboration framework for inspection robots is a zonal monitoring collaboration mode with cross-validation with transportation robots, while the basic collaboration framework for transportation robots is a competitive coordination framework of "full load priority, nearby transportation". The role layer is used to dynamically allocate role weights, for example, assigning different weights to inspection robots and transportation robots respectively. The model layer generates strategy adjustment gradients through the PPO (Proximal Policy Optimization) algorithm to adapt the collaboration strategy parameter settings of underground tasks, such as the zonal monitoring parameters of inspection robots.

[0091] PPO is a reinforcement learning algorithm that improves training stability by limiting the magnitude of policy updates. In this embodiment, the PPO algorithm uses Generalized Advantage Estimation (GAE) as the advantage function, with a discount factor γ set to 0.99, GAE parameter λ set to 0.95, policy clipping parameter ε set to 0.2, and the policy network is updated every 2048 time steps.

[0092] After receiving the cooperation strategy parameters, each agent executes the corresponding task in either a physical or virtual environment. The physical agent collects environmental data through sensors, while the virtual agent generates simulated data based on a pre-set environmental model. For example, an inspection robot collects downhole environmental data using a dust-proof depth camera and an IMU sensor, monitoring it in a "zonal synchronization + cross-validation" mode; a transport robot executes transfer tasks according to load priority, and the virtual agent synchronously simulates the task process to help correct strategy deviations.

[0093] The mining multi-agent collaborative operation method provided in this application receives mining tasks; parses the mining tasks to determine the collaboration type of the agents, wherein the collaboration type includes single type and mixed type; and generates an agent collaboration strategy based on the collaboration type and the agents' historical collaboration scores. The agent collaboration strategy includes at least a basic mining collaboration framework, agent role weights, and collaboration strategy parameters. It can integrate multiple collaboration types in coal mines to generate an agent collaboration strategy containing a basic mining collaboration framework, agent role weights, and collaboration strategy parameters, optimizing the interaction between agents, adapting to dynamic underground task requirements, avoiding resource conflicts, improving the efficiency and flexibility of underground agent collaboration, enhancing underground operation safety, and solving problems such as the singleness, inefficiency, and poor flexibility of underground collaboration.

[0094] In some embodiments, such as Figure 2 As shown, the method further includes:

[0095] S201: Based on the initial model of the intelligent agents and the task requirements, cluster the multiple intelligent agents in the well and divide the multiple intelligent agents into core clusters and edge clusters;

[0096] S202: Determine the communication method of the intelligent agent according to the divided cluster type, wherein the core cluster communicates via optical fiber and the edge cluster communicates via intrinsically safe wireless communication;

[0097] S203: Transmit the cooperation strategy parameters to the corresponding agent according to the communication method.

[0098] In this embodiment, agent clustering is performed first when building the communication cluster. Based on the initial model of the agents and task requirements, the DBSCAN clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, an unsupervised clustering algorithm) is used to cluster agents based on their location and task relevance, thus dividing the cluster into core clusters and edge clusters.

[0099] The DBSCAN algorithm is a density-based spatial clustering algorithm that divides regions with sufficient density into clusters and discovers clusters of arbitrary shapes in noisy data. In this embodiment, by setting an appropriate radius (ε) and minimum number of samples (MinPts), agents that are close to each other and closely connected are clustered into one class. Here, ε is set to half the communication distance between agents (i.e., 25m), and MinPts is set to 3 to ensure that each cluster has enough agents to work collaboratively.

[0100] After dividing the communication clusters, a flexible communication architecture is configured. The core cluster uses a mining-grade fiber optic wired communication method (fiber optic communication) to achieve low-latency data transmission through switches; the edge clusters use a mining-grade intrinsically safe wireless communication method (intrinsically safe wireless communication) to exchange data through wireless routers. Then, based on the divided communication cluster type, the cooperation strategy parameters are transmitted to the corresponding agents to improve underground communication efficiency and security.

[0101] The agent role can be a role determined by the agent type, such as an inspection robot and a transport robot; the agent role can also be a role determined by the communication cluster mentioned above, such as a core cluster and an edge cluster.

[0102] In this embodiment, based on the agent role weight (e.g., ≥70% for the core cluster) and collaboration type, a flexible communication architecture can be used to dynamically switch the triggering conditions and execution flow of fiber optic or intrinsically safe wireless communication for mining, efficiently transmitting the generated collaboration strategy parameters to each agent. The core cluster uses intrinsically safe wireless communication, while the edge cluster uses it.

[0103] Compared to a static star topology, the flexible communication architecture (fiber optic + intrinsically safe wireless) in this embodiment can adapt to downhole signal attenuation environments, reducing critical information transmission latency from 50ms to ≤20ms and communication packet loss rate from 5% to ≤0.1%, effectively improving downhole communication efficiency and security.

[0104] In some embodiments, the method further includes:

[0105] S301: Constructing a knowledge graph for coal mines;

[0106] S302: Pre-train the coal mine knowledge graph data in the coal mine knowledge graph through the backbone network to obtain a general collaborative knowledge model for coal mines;

[0107] S303: Based on the mining branch network, the general coal mine collaborative knowledge model is fine-tuned to obtain a coal mine collaborative knowledge model applicable to preset operation scenarios.

[0108] In this embodiment, relevant underground coal mine data can be collected in advance, including underground equipment parameters (coal mining machine, scraper conveyor models and failure cases, etc.) over the past 5 years, historical gas monitoring data (concentration change curves, sudden increase warning records, etc.), roadway topology (size of mining face, tunneling face and support parameters), and coal transfer process (transportation route, load threshold), etc. After acquiring the data, data preprocessing can be performed. Based on the characteristics of underground data, image filtering algorithms are used to remove dust interference, signal smoothing technology is used to suppress vibration and noise, and abnormal gas concentrations / equipment failure types are labeled to generate sample data suitable for model training.

[0109] Then, a coal mine knowledge graph is constructed using relevant underground coal mine data. The coal mine knowledge graph data is stored in the Neo4j database and includes entity relationships such as "equipment-fault-handling solution", "gas concentration-ventilation-early warning level", and "roadway-support-inspection frequency". The coal mine knowledge graph can be iteratively updated based on underground operation data.

[0110] After constructing the coal mine knowledge graph, collaborative training of mining models can be carried out. During model training preparation, a backbone network (such as a lightweight version of GPT-4, which needs to be compressed to 1.2GB) is deployed on a cloud server. Pre-training is performed using the pre-processed coal mine knowledge graph data to obtain a general collaborative knowledge model for coal mines. At the same time, mining branch networks are set up: branch network structures are designed for specific underground scenarios such as gas early warning, scraper conveyor fault diagnosis, and roadway support detection. Feature layers such as dust interference factors and vibration compensation layers are added, and scenario-specific data (such as 100+ simulated gas surge data and 500+ equipment fault images) are prepared for fine-tuning.

[0111] As shown above, this embodiment constructs a resource-sensitive training engine adapted to the computing power of underground edge devices by leveraging a coal mine knowledge graph. It employs a collaborative mechanism of backbone network pre-training and branch network fine-tuning. Specifically, the cloud-based backbone network pre-trains general coal mine knowledge, while branch networks are fine-tuned for specific scenarios (such as dust interference and vibration compensation). This improves training efficiency, addresses the issues of high resource requirements and limited scale, and yields a more reasonable and reliable collaborative strategy. The resource-sensitive training engine makes model training more efficient and reduces the computing burden on edge devices.

[0112] In some embodiments, the method further includes:

[0113] S401: Obtain the training metrics of the downhole agent;

[0114] S402: Determine whether the agent is qualified based on the training indicators;

[0115] S403: If the agent is a defective agent, generate corresponding training samples based on the defective agent.

[0116] In this embodiment, the resource-sensitive training engine described above can be constructed by combining large-scale model supervision. Large-scale model supervision includes: real-time collection of training metrics of downhole agents (such as policy execution error, task completion rate, etc.), displaying them on the terminal display of the ground control center, and generating personalized training samples when the training metrics fail to meet the standards (the agent is an unqualified agent) and inputting them into the resource-sensitive training engine to improve its training difficulty and relevance.

[0117] In some embodiments, the method further includes:

[0118] S501: Screen mining collaboration experience through cosine similarity algorithm, and push the mining collaboration experience to intelligent agents with the same role.

[0119] In this embodiment, effective mining collaboration experience is screened using cosine similarity (e.g., ≥0.8) and pushed to agents with the same role to achieve knowledge sharing. The cosine similarity algorithm measures the similarity between two vectors; the closer the value is to 1, the more similar the two vectors are. In this embodiment, the collaboration experience of agents is represented as a vector. By calculating the cosine similarity of the collaboration experience vectors of different agents, if the similarity is ≥0.8, the collaboration experience is considered valuable to other agents and is shared. This shared knowledge is then used to implement large-scale model supervision.

[0120] In this embodiment, a resource-sensitive training engine and scalable architecture are constructed for mining, combined with large-scale model supervision to adapt to the computing power of underground edge devices, thus solving the problems of high training resource requirements and limited scale. For example, the resource-sensitive training engine adapts to the computing power of underground edge devices (≤16TOPS), compresses the model size to 40% of the original, shortens the training time by 50%, and can adapt to different roadways without repeated training, thereby optimizing the efficiency of underground resource utilization.

[0121] In some embodiments, the method further includes:

[0122] S601: Extract point cloud features of the downhole embodied environment;

[0123] S602: Group adjacent point clouds to obtain grouped features;

[0124] S603: Match the grouping features with the agent collaboration strategy to determine whether the agent collaboration strategy is suitable for the work scenario;

[0125] S604: If a match is found, the agent collaboration strategy is executed; if a match is not found, the agent collaboration strategy is replanned using a preset collaborative operation model.

[0126] In this embodiment, underground embodied knowledge alignment deeply couples the large mining model with multiple agents to achieve underground scene adaptation. The PointNet++ model processes the point cloud data of the embodied environment. First, a sampling layer extracts point cloud features of the underground embodied environment (such as point clouds of tunnel walls and scraper conveyors). Then, a grouping layer groups adjacent points, transforming them into 128-dimensional symbol vectors (128-dimensional point cloud features) that the large model can resolve. These vectors are then matched with a collaborative strategy to ensure that the agent's actions conform to the physical constraints of the environment. For example, feature matching uses a cosine similarity threshold of ≥0.8. When the similarity reaches the threshold, the agent's collaborative strategy is triggered; otherwise, a collaborative operation model is initiated to re-plan the agent's collaborative strategy.

[0127] In some embodiments, the method further includes:

[0128] S701: Constructing decision trees for high-frequency scenarios;

[0129] S702: When the well is in a preset operating condition, determine the processing strategy for the preset operating condition based on the high-frequency scenario decision tree.

[0130] In this embodiment, a high-frequency scenario decision tree mechanism for mining enables rapid adaptation and response of the intelligent agent to the scenario. For example, a high-frequency scenario decision tree can be constructed using the C4.5 algorithm and burned into an intrinsically safe FPGA acceleration chip to achieve a response time of ≤20ms. When encountering emergencies such as equipment failure, sudden increase in gas concentration, or roadway deformation, the system quickly matches and outputs the optimal solution from the pre-generated solution library, with a response latency of ≤20ms.

[0131] Decision trees are a tree-structured decision-making algorithm that constructs models capable of classifying or regressing data by continuously splitting features. This algorithm selects splitting features based on information gain ratio, can handle both continuous and discrete features, and exhibits good generalization ability. The constructed decision tree is then programmed into an FPGA acceleration chip, leveraging the FPGA's parallel computing capabilities to achieve rapid decision-making and response in various scenarios.

[0132] During the collaborative operation of intelligent agents, real-time data collection can be performed on the agents underground, including collaboration logs, training metrics, and scenario feedback. This data can be organized and analyzed to evaluate the collaborative performance and identify any problems. Specifically, by constructing a coal mine-specific knowledge graph and integrating its rules (such as gas explosion limit correlations) into the agent's action planning, the fusion of coal mine knowledge (e.g., equipment-fault-environment) with the agent is achieved. This enables adaptation to underground scenarios and rapid response, improves the reusability of coal mine knowledge, and reduces collaboration costs. For example, knowledge transfer efficiency is improved by 70%, and the cost of adapting to new roadway scenarios is reduced by 60%.

[0133] Through the aforementioned embodied knowledge alignment (steps S601 to S604), real-time response enhancement (steps S701 to S702), and coal mine knowledge fusion, the large model and mining agent are deeply coupled, improving the adaptability and collaborative foresight of underground scenarios and enhancing the safety of underground operations. For example, it improves the agent's adaptability to complex environments such as low light and high dust levels, while also reducing the time for identifying gas hazards and handling equipment failures, and improving coal transfer efficiency. In a specific embodiment, through mining embodied knowledge alignment (point cloud feature extraction accuracy ≥92%) and high-frequency response of mining FPGA, the gas hazard identification rate is increased from 60% to 98%, and the equipment failure handling time is shortened to 67%, meeting the requirements for safe coal mine operations.

[0134] This application also enables model and strategy optimization. Specifically, it can modify the PPO algorithm reward function of the dynamic collaboration mechanism and adjust the weights of the ternary strategy based on the collaboration success rate data output by the scenario adaptation, thereby adjusting the collaborative operation model and improving the effectiveness of the collaboration strategy. Furthermore, this application utilizes newly collected agent operation data to retrain and fine-tune the backbone network and branch networks, updating the agent's capability matrix and improving the overall performance of the system.

[0135] In specific implementation, system configuration adjustments can also be made, including: (1) Adjusting the resource allocation ratio and optimizing the utilization efficiency of computing resources according to the new collaboration rules and model parameters. For example, adjusting the pruning ratio of the large model supervision module according to the computing resource limitations of the industry scenario. (2) Checking the operating status of the communication architecture, and re-clustering and reconstructing the communication cluster according to the changes in the number of agents and communication load to ensure the efficiency and stability of communication.

[0136] As can be seen from the above, in this embodiment, the iterative update of the multi-agent collaborative operation method is achieved through the collection and analysis of agent operation data, model and strategy optimization, and system configuration adjustment.

[0137] Furthermore, embodiments of this application can also be used for maintenance and troubleshooting of multi-agent collaborative operation systems.

[0138] During routine maintenance of the multi-agent collaborative operation system, mining equipment can be inspected periodically. For example, explosion-proof surfaces of explosion-proof equipment and intrinsically safe power supply voltage can be checked weekly; dust covers of agent sensors can be cleaned monthly to ensure normal equipment operation, and aging or faulty equipment can be replaced promptly. Routine maintenance includes not only hardware but also software maintenance, such as regularly (e.g., quarterly) updating the mining operating system and model, regularly (e.g., monthly) backing up the coal mine knowledge graph and model parameters stored on the ground control center terminal server to prevent data loss or corruption; and regularly (e.g., monthly) promptly patching underground data software vulnerabilities to improve system security and stability. Routine maintenance also includes communication testing, such as monthly testing of underground fiber optic and intrinsically safe wireless communication to ensure the entire link packet loss rate is ≤0.1% and latency is ≤20ms, and aging communication cables can be replaced promptly.

[0139] Handling common faults in multi-agent collaborative work systems includes:

[0140] 1) Communication failure

[0141] If communication between the underground intelligent agent and the ground dispatch center is interrupted, first check the operating status of the intrinsically safe router (restart to restore); if the fault persists, check the fiber optic connector (whether it has become loose due to vibration), re-plug and test; if the problem still persists, check the communication protocol configuration and IP address settings to ensure that the network parameters are correct. If necessary, reconfigure the communication cluster and switch to the backup communication channel (the intrinsically safe wireless backup frequency band for mining).

[0142] 2) Agent failure

[0143] For embodied intelligent agents, if a hardware failure occurs (such as sensor damage or motor failure), repair or replace the parts in a timely manner; if a software failure occurs, redeploy the intelligent agent software and restore the initial parameters.

[0144] If a virtual agent malfunctions, check the resource usage of the deployed devices, close unnecessary processes to release resources, or restart the virtual agent software.

[0145] 3) Model training failure

[0146] If the model training process results in slow convergence or failure to converge, check the quality and quantity of the training data, increase the number of downhole training samples, or adjust the data preprocessing method. Adjust the model's hyperparameters (such as learning rate, batch size, etc.), or change the training algorithm and retrain the model.

[0147] In summary, the multi-agent collaborative operation system of this application constructs a three-layer closed-loop architecture with "a dynamic collaboration mechanism for mining as the core, resource-sensitive training as a guarantee, and underground scenario adaptation as the goal": The dynamic collaboration mechanism optimizes the interaction mode of underground agents, providing an efficient foundation for training; resource-sensitive training achieves computing power adaptation based on underground data; and scenario adaptation transforms the collaboration and training results into underground operational capabilities, while simultaneously providing feedback on data optimization strategies. These three elements form a "collaboration-training-adaptation-iteration" closed loop, effectively solving the problems of singular collaboration, inefficient communication, high resource requirements, and insufficient scenario adaptability in underground coal mine collaboration scenarios.

[0148] The following specific implementation example illustrates the multi-agent collaborative operation method for coal mine longwall faces.

[0149] Application Scenario: The 30201 longwall face of a large coal mine is 200m long, 600m deep, and has a coal seam thickness of 2.5m. It faces problems such as low light (illuminance ≤50lux), high dust (concentration ≤100mg / m³), and signal attenuation (communication distance ≤30m). It is necessary to achieve coordinated operation of gas monitoring, equipment inspection and coal transfer. Traditional manual methods are inefficient and have high safety risks.

[0150] System deployment: Deploy 8 mining explosion-proof inspection robots (inspection 01-08), 3 mining explosion-proof transport robots (transport 01-03), 1 mining explosion-proof switch, 2 intrinsically safe wireless routers, and deploy 2 cloud servers and 3 monitoring displays in the ground dispatch center.

[0151] (1) Application of dynamic collaboration mechanism

[0152] Task assignment: The ground control center assigns the following tasks: conduct gas concentration monitoring of the 30201 longwall face every 5 minutes and complete scraper conveyor inspection and coal transfer every 10 minutes (each time requiring a full load of 500kg).

[0153] Collaboration type label generation: The BERT model analyzes the task and generates a "hybrid collaborative" label (2). At the same time, it identifies that the task includes "inspection robot cooperative monitoring" and "transport robot resource competition".

[0154] Collaboration strategy generation: The ternary strategy fusion technology combines historical collaboration scores (the historical accuracy rate of Inspection 01-03 is 92%, and Transportation 01 has the highest load efficiency) to output the following collaboration strategy: Inspection robots collaborate in a "3-unit group for zoned monitoring (1-70m, 71-140m, 141-200m) + cross-validation" mode (weight 70%); Transportation robots compete and coordinate in a "full load priority, nearest transportation" manner (weight 30%).

[0155] Communication assurance: The core cluster (inspection 01-03, transportation 01) transmits data via fiber optic cable, while the edge cluster (other inspection and transportation robots) transmits data via intrinsically safe wireless cable (dynamic routing avoids blind spots in the working face fault zone), ensuring real-time uploading of gas data.

[0156] (2) Application of resource-sensitive training engines

[0157] Coal mine knowledge graph construction: including the topology of the 30201 working face (200m×5m roadway), scraper conveyor failure cases (jamming, chain breakage symptoms, etc.), gas concentration diffusion model (ventilation volume requirement at 0.5% CH4), and more than 3,500 entity relationships;

[0158] Model training: The cloud backbone network is pre-trained based on 5 years of historical data from the mining area. The branch network adds a "dust interference factor" feature layer and uses 120 simulated gas surges and 80 equipment jamming data to enhance the training samples. During the training process, the edge nodes compress the inspection robot inference model to 400MB (40% of the original size) to match its 16TOPS computing power.

[0159] Resource optimization: When the computing power of the downhole edge node is insufficient, some training tasks are automatically offloaded to the cloud to ensure training efficiency.

[0160] (3) Application of deep coupling between large models and multi-agent systems

[0161] Embodied knowledge alignment: The underground inspection robot collects point cloud data of the working face through dustproof lidar, extracts 128-dimensional features through the PointNet++ model, and matches them with the "equipment fault early warning" strategy. When the cosine similarity is 0.85≥0.8, the inspection robot is triggered to focus and detect.

[0162] Real-time response: When the inspection robot 02 detects a sudden increase in methane concentration to 0.9% CH4, the FPGA acceleration chip outputs an emergency plan within 15ms: three inspection robots form a triangular monitoring array to fully cover the methane area, the transport robot immediately switches to the working face emergency escape route, and at the same time triggers the ground dispatch center to activate the ventilation equipment and accelerate it to 20m³ / s.

[0163] Improved efficiency: After the multi-agent collaborative operation system was put into operation, the gas hazard identification rate increased from 60% to 98%, the fault handling time of the scraper conveyor was shortened from 30 minutes to 10 minutes, and the coal transfer efficiency increased from 500 tons / day to 700 tons / day, fully meeting the needs of safe and efficient operation in underground coal mines.

[0164] Example 2

[0165] Figure 3 A schematic diagram of the structure of a mining multi-agent collaborative operation device according to an embodiment of this application is shown, such as... Figure 3 As shown in the figure, this application provides a mining multi-agent collaborative operation device, including:

[0166] Receiver module 10 is configured to receive mining tasks;

[0167] The parsing module 20 is configured to parse the mining task and determine the collaboration type of the agent, wherein the collaboration type includes a single type and a mixed type;

[0168] The collaboration strategy generation module 30 is configured to generate an agent collaboration strategy based on the collaboration type and the agent's historical collaboration score. The agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

[0169] In some embodiments, the parsing module 20 is further configured to:

[0170] The text of the mining task is parsed using the BERT-base model to generate collaboration type tags.

[0171] In some embodiments, the collaboration strategy generation module 30 is further configured to:

[0172] The basic collaboration framework is determined through the rules layer;

[0173] Assign agent role weights to downhole agents through a role layer;

[0174] The parameters of the cooperative strategy are determined by using a proximal policy optimization algorithm at the model layer.

[0175] In some embodiments, the mining multi-agent collaborative operation device further includes an optimization module configured as follows:

[0176] Obtain training metrics for the downhole agent;

[0177] The suitability of the agent is determined based on the training metrics.

[0178] If the agent is a defective agent, a corresponding training sample is generated based on the defective agent.

[0179] In some embodiments, the optimization module is further configured as follows:

[0180] Mining collaboration experience is screened using a cosine similarity algorithm, and then pushed to intelligent agents with the same role.

[0181] In some embodiments, the mining multi-agent collaborative operation device further includes a scene adaptation module, configured as follows:

[0182] Extract point cloud features of the downhole embodied environment;

[0183] Adjacent point clouds are grouped to obtain grouped features;

[0184] The grouping features are matched with the agent collaboration strategy to determine whether the agent collaboration strategy is suitable for the work scenario.

[0185] If a match is found, the agent collaboration strategy is executed; if a match is not found, the agent collaboration strategy is replanned using a preset collaborative operation model.

[0186] In some embodiments, the mining multi-agent collaborative operation device further includes a communication module, configured as follows:

[0187] Based on the initial model of the intelligent agents and the task requirements, multiple intelligent agents in the mine are clustered and divided into core clusters and edge clusters.

[0188] The communication method of the intelligent agent is determined according to the classified cluster type, wherein the core cluster communicates via optical fiber and the edge cluster communicates via intrinsically safe wireless communication;

[0189] The cooperation strategy parameters are transmitted to the corresponding agent according to the communication method.

[0190] In some embodiments, the mining multi-agent collaborative operation device further includes a model building module, configured as follows:

[0191] Constructing a knowledge graph for coal mines;

[0192] By pre-training the coal mine knowledge graph data in the coal mine knowledge graph through the backbone network, a general collaborative knowledge model for coal mines is obtained.

[0193] Based on the mining branch network, the general collaborative knowledge model of coal mines is fine-tuned to obtain a coal mine collaborative knowledge model applicable to preset operation scenarios.

[0194] In some embodiments, the optimization module is further configured to: construct a high-frequency scenario decision tree;

[0195] When the well is in a preset operating condition, the processing strategy for the preset operating condition is determined based on the high-frequency scenario decision tree.

[0196] The mining multi-agent collaborative operation device provided in this application corresponds to the mining multi-agent collaborative operation method in the above embodiments. Any option in the mining multi-agent collaborative operation method embodiment is also applicable to the embodiment of the mining multi-agent collaborative operation device, and will not be repeated here.

[0197] Example 3

[0198] This application also provides an electronic device, which includes at least a memory and a processor. The memory stores a computer program, and the processor executes the computer program in the memory to implement the above-described multi-agent cooperative operation method for mining.

[0199] In some embodiments, the processor executing a computer program may be a processing device that includes one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor that runs other instruction sets, or a processor that runs a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), system-on-a-chip (SoCs), etc.

[0200] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, PHP, and Python, as well as conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the passenger's computer, partially on the passenger's computer, as a standalone software package, partially on the passenger's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the passenger's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0201] The memory may be a read-only memory (ROM), random access memory (RAM), phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), electrically erasable programmable read-only memory (EEPROM), other types of random access memory (RAM), flash drives or other forms of flash memory, cache, registers, static memory, optical disc read-only memory (CD-ROM), digital versatile optical disc (DVD) or other optical storage, magnetic tape cassette or other magnetic storage devices, or any other possible non-transitory medium used to store information or instructions that can be accessed by computer equipment.

[0202] The electronic devices in this application may include, but are not limited to, fixed terminal devices such as servers, desktop computers, and digital TVs, as well as mobile terminal devices such as in-vehicle devices, handheld devices (e.g., mobile phones, tablets, etc.), and wearable devices (e.g., smartwatches, smart bracelets, etc.).

[0203] Example 4

[0204] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-agent cooperative operation method for mining.

[0205] The computer-readable storage medium of this application embodiment can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. The computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device; for example, it can be the aforementioned memory.

[0206] The computer programs of embodiments of this application can be organized into one or more computer-executable components or modules. Various aspects of this application can be implemented with any number and combination of such components or modules. For example, aspects of this application are not limited to the specific computer-executable instructions or specific components or modules shown in the drawings and described herein. Other embodiments may include different computer-executable instructions or components having more or fewer functions than those shown and described herein.

[0207] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

[0208] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. Multitasking and parallel processing may be advantageous in certain environments. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

Claims

1. A multi-agent collaborative operation method for mining, characterized in that, include: Receive mining tasks; The mining task is analyzed to determine the collaboration type of the agent, wherein the collaboration type includes single type and mixed type; An agent collaboration strategy is generated based on the collaboration type and the agent's historical collaboration scores. The agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

2. The method according to claim 1, characterized in that, The mining task is analyzed to determine the cooperation type of the agent, including: The text of the mining task is parsed using the BERT-base model to generate collaboration type tags.

3. The method according to claim 2, characterized in that, Generate agent cooperation strategies based on the cooperation type and the agent's historical cooperation scores, including: The basic collaboration framework is determined through the rules layer; Assign agent role weights to downhole agents through a role layer; The parameters of the cooperative strategy are determined by using a proximal policy optimization algorithm at the model layer.

4. The method according to claim 1, characterized in that, The method further includes: Obtain training metrics for the downhole agent; The suitability of the agent is determined based on the training metrics. If the agent is a defective agent, a corresponding training sample is generated based on the defective agent.

5. The method according to claim 1, characterized in that, The method further includes: Mining collaboration experience is screened using a cosine similarity algorithm, and then pushed to intelligent agents with the same role.

6. The method according to claim 1, characterized in that, The method further includes: Extract point cloud features of the downhole embodied environment; Adjacent point clouds are grouped to obtain grouped features; The grouping features are matched with the agent collaboration strategy to determine whether the agent collaboration strategy is suitable for the work scenario. If a match is found, the agent collaboration strategy is executed; if a match is not found, the agent collaboration strategy is replanned using a preset collaborative operation model.

7. The method according to claim 1, characterized in that, The method further includes: Based on the initial model of the intelligent agents and the task requirements, multiple intelligent agents in the mine are clustered and divided into core clusters and edge clusters. The communication method of the intelligent agent is determined according to the classified cluster type, wherein the core cluster communicates via optical fiber and the edge cluster communicates via intrinsically safe wireless communication; The cooperation strategy parameters are transmitted to the corresponding agent according to the communication method.

8. The method according to claim 1, characterized in that, The method further includes: Constructing a knowledge graph for coal mines; By pre-training the coal mine knowledge graph data in the coal mine knowledge graph through the backbone network, a general collaborative knowledge model for coal mines is obtained. Based on the mining branch network, the general collaborative knowledge model of coal mines is fine-tuned to obtain a coal mine collaborative knowledge model applicable to preset operation scenarios.

9. The method according to claim 1, characterized in that, The method further includes: Construct decision trees for high-frequency scenarios; When the well is in a preset operating condition, the processing strategy for the preset operating condition is determined based on the high-frequency scenario decision tree.

10. A multi-agent collaborative operation device for mining, characterized in that, include: The receiving module is configured to receive mining tasks. The parsing module is configured to parse the mining task and determine the collaboration type of the agent, wherein the collaboration type includes a single type and a mixed type; The collaboration strategy generation module is configured to generate an agent collaboration strategy based on the collaboration type and the agent's historical collaboration scores. The agent collaboration strategy includes at least a mining basic collaboration framework, agent role weights, and collaboration strategy parameters.

Images