Multi-objective multi-constraint hybrid decision-making agent planning method for coal mining process

Abstract

The present application relates to a kind of coal mining process multi-objective multi-constraint hybrid decision-making agent planning method, belong to coal mine intelligent technology field.It includes: global system level planning type organization Agent plans the target output of current production cycle;Each local equipment level planning type organization Agent obtains the four-dimensional target optimization function and its reward function of respective corresponding scene task;Multiple multi-objective joint task strategies are obtained by using MAPPO algorithm planning;After local equipment level planning type arbitration Agent obtains the conflict-free multi-objective joint task strategy, local equipment level planning type organization Agent calculates the multi-objective joint task strategy of optimal value based on the arbitration strategy of cooperation game according to corresponding reward function, and the multi-objective joint task strategy of optimal value is issued to Agent alliance.The present application can improve the intelligent degree of decision-making, not only can reduce labor cost, but also can improve decision efficiency and reduce security risks.

Description

Technical Field

[0001] This invention relates to the field of intelligent coal mining technology, and in particular to a multi-objective, multi-constraint hybrid decision-making intelligent agent planning method for coal mining processes. Background Technology

[0002] Coal mining involves numerous aspects, including geological conditions, environmental safety, production efficiency, and equipment coordination. Its optimization decisions aim to achieve safe, efficient, cost-effective, and intelligent coal mining with reduced manpower, enabling dynamic adjustment and automatic adaptation of the mining process. Furthermore, the execution of coal mining operations requires comprehensive consideration of constraints such as resources, timing, and safety. Therefore, achieving multi-objective optimization while satisfying multiple constraints necessitates a hybrid decision-making process that combines optimization objectives with constraints.

[0003] Currently, intelligent decision-making technology systems for coal mining still rely mainly on human experience, which is not only costly in terms of manpower and inefficient, but may also pose certain safety hazards. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a multi-objective, multi-constraint hybrid decision-making intelligent agent planning method for coal mining processes. The technical solution of this invention is as follows: A multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes, the method comprising: S1, the global system-level planning organization agent plans the target output of the current production cycle based on the relationship between the historical output and the target output plan of the previous production cycle, and distributes the target output of the current cycle to multiple local equipment-level planning organization agents of the coal mining task process; among them, one scenario task corresponds to one local equipment-level planning organization agent and one local equipment-level planning arbitration agent. S2, each local device-level planning organization agent obtains its corresponding scenario task's four-dimensional objective optimization function and its reward function; S3, each local device-level planning organization agent uses the MAPPO algorithm to plan multiple multi-objective connection task strategies based on the four-dimensional objective optimization function and reward function of its corresponding scenario task, and then passes the multiple multi-objective connection task strategies to the corresponding local device-level planning arbitration agent; S4, the local device-level planning arbitration agent detects whether there are conflicts among the multiple multi-objective connection task strategies. If each multi-objective connection task strategy has a conflict, the detection result is fed back to the corresponding local device-level planning organization agent. The local device-level planning organization agent modifies some constraint parameters of the four-dimensional objective function under the corresponding scenario task and re-plans the multi-objective connection task strategy until a conflict-free multi-objective connection task strategy is obtained. Then, the local device-level planning organization agent calculates the optimal multi-objective connection task strategy based on the arbitration strategy of cooperative game based on the corresponding reward function and issues the optimal multi-objective connection task strategy to the agent alliance.

[0005] Preferably, before step S2, the method further includes: S0, based on the principles of safety, efficiency improvement, cost reduction, and intelligent coal mining with fewer personnel, predefines a four-dimensional objective function and its reward function for local equipment-level planning organization agents in different scenarios; The four-dimensional objective function Represented as ; in, This is a risk indicator function built based on safety procedures. This is an efficiency evaluation model constructed with man-hours per ton of coal as the core indicator. This is an energy consumption evaluation model constructed using electricity consumption per ton of coal and material consumption per ton of coal as core indicators. It is a mathematical function constructed with the automatic operation time of each coal mining system as the core indicator; The reward function V of the four-dimensional objective function F is expressed as: ; Where w1 to w4 are the weights of each reward function. Represent reward functions in different dimensions.

[0006] Preferably, the feature is that a risk index function is constructed based on safety regulations. ,include: First, the key safety provisions of the safety procedures are formalized into a set of indicators, and the actual value of each indicator in the set of indicators constitutes a risk indicator set. ,in This represents the true situation of the j-th risk indicator; Then, according to The properties construct binary judgment propositions or continuous risk level ; For binary decision propositions ,Sure The proposition is true under safe conditions. And expressed as: ; For continuous risk Continuous risk is measured based on the thresholds set in the safety regulations. The corresponding threshold is Then continuous risk level Represented as: ; in, and It is an OR relationship, meaning that the same risk indicator cannot be counted twice; Next, determine the weights of the risk indicators. , , ; Next, obtain the confidence level of the risk indicators. ; Finally, the set of risk indicators Aggregate all risk indicators in the dataset, calculate the weighted risk score, and obtain... , is represented as: .

[0007] Preferably, the efficiency evaluation model constructed using man-hours per ton of coal as the core indicator... The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate; the output is man-hours per ton of coal. The energy consumption evaluation model constructed with electricity consumption per ton of coal and material consumption per ton of coal as core indicators. The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate. The outputs are the power consumption per ton of coal and the material consumption per ton of coal.

[0008] Preferably, a mathematical function is constructed using the automatic operation time of each coal mining system as the core indicator. ,include: First, determine the set of time proportions for automatic operation of each coal mining system. ,in, This represents the percentage of automatic running time for the i-th system. Next, determine the weight of the indicator for the percentage of automatic running time of the i-th system. , , ; Then, determine the confidence level of the indicator for the percentage of automatic running time of the i-th system. ; Finally, for the time percentage set The weighted automation rate is calculated by aggregating the percentage of all automated running times. , is represented as: .

[0009] Preferably, when the local device-level planning arbitration agent detects whether there is a conflict in any multi-objective connection task strategy, it includes: S41, Local device-level planning arbitration agent predefines preset constraints for resource conflicts, timing conflicts, and security conflicts; S42, the local device-level planning arbitration agent determines whether any multi-objective connection task strategy has resource conflicts, timing conflicts or security conflicts based on preset constraints. If any of the conflicts exist, it is determined that the multi-objective connection task strategy has a conflict.

[0010] Preferably, the preset constraints for resource conflicts include hydraulic system flow constraints, expressed as: ; in, These represent the flow rates required for the hydraulic support cluster to lower, move, raise, and push the support column, respectively. Indicates the flow threshold of the hydraulic system; Indicates the resource conflict detection results; This indicates no conflict. This indicates a conflict; The preset constraints for the timing conflict include performing time planning calculations on the point logic time relationship matrix M of the hydraulic support actions to obtain several R_timetables that satisfy all time constraints of the action event set. Each R_timetable is then verified. If the verification passes, it indicates no conflict; otherwise, it indicates a conflict, expressed as follows: ; in, This indicates that the verification failed. This indicates that the verification has passed. This indicates the timing conflict detection result. This indicates no conflict. This indicates a conflict; The preset constraints for the safety conflict include a constraint on whether the gas level exceeds the limit, expressed as: ; in, Real-time gas content, The gas exceedance threshold; Indicates the results of security conflict detection. This indicates no conflict. This indicates a conflict.

[0011] Preferably, the local device-level planning organization agent re-plans the multi-objective connection task strategy using the MAPPO algorithm every preset time interval.

[0012] Preferably, the global system-level planning organization agent plans the target output for the current production cycle based on the relationship between the historical output and the target output plan of the previous production cycle, including: S11, the global system-level planning organization agent determines the target output plan set for each shift, day, week, month, and year of the previous production cycle. And determine the set of actual completed quantities for each shift, day, week, month, and year of the previous production cycle. ; S12, Global System-Level Planning Organization Agent Calculation Deviation Vector Set ; S13, the global system-level planning organization agent modifies the target output plan for the current production cycle based on the deviation vector set, and obtains the target output for the current production cycle. , is represented as: ; in, To adjust the coefficient matrix; , which represents the correction rate for the corresponding level.

[0013] Preferably, in step S4, the local device-level planning organization agent issues the optimal multi-objective connection task strategy to the agent alliance, including: Local device-level planning-type organizational agents will issue the most valuable multi-objective interconnected task strategy to reasoning-type model agents; The reasoning model agent performs dynamic decision-making reasoning on the value-optimal multi-objective connection task strategy, transforms the value-optimal multi-objective connection task strategy into specific and implementable device execution instructions, and then sends the device execution instructions to the reflective program agent. The reflective program agent controls the coal mining equipment to execute the equipment execution instructions.

[0014] All of the above-mentioned optional technical solutions can be combined arbitrarily, and the present invention will not provide a detailed description of the structure after each combination.

[0015] By means of the above solution, the beneficial effects of the present invention are as follows: By using a global system-level planning organization agent to plan the target output for the current production cycle, and local equipment-level planning organization agents to obtain the four-dimensional objective optimization function and reward function for their respective scenario tasks, multiple multi-objective linked task strategies are planned using the MAPPO algorithm. Local equipment-level planning arbitration agents detect conflicts among these strategies until a conflict-free strategy is obtained. Finally, based on the corresponding reward function and a cooperative game-based arbitration strategy, the local equipment-level planning organization agents calculate the optimal multi-objective linked task strategy and distribute it to the agent alliance. This provides a method for intelligent decision-making on multi-dimensional objectives and multi-faceted constraints using various types of agents. This method, by combining objective optimization and multi-constraint resolution in coal mining scenarios, improves the intelligence of decision-making, reduces labor costs, increases decision-making efficiency, and minimizes safety hazards.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a flowchart of the multi-objective, multi-constraint hybrid decision-making intelligent agent planning method for coal mining processes provided in this embodiment of the invention.

[0018] Figure 2 This is a schematic diagram of the centralized distributed hierarchical multi-timescale dynamic organization strategy system architecture provided in the embodiments of the present invention.

[0019] Figure 3 This is a schematic diagram of the overall framework of the coal mining face following the machine scenario in an embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] To facilitate understanding of the multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes provided in this embodiment of the invention, the various types of agents involved in this embodiment of the invention will be explained below.

[0022] Reflective Program Agent: An automation program (existing mature technology) for the microcontrollers of fully mechanized mining equipment such as coal mining machines, hydraulic supports, and scraper conveyors. It enables the automatic execution of simple tasks by the mining equipment. Located at the equipment control layer, its response time is in milliseconds or seconds. For example, it can be deployed as a coal mining machine controller or a hydraulic support controller.

[0023] The reasoning-based model agent is a control decision-making algorithm for scenarios such as hydraulic support and fluid supply coordination control, mine pressure support, machine-following regulation, and straightness control. It can realize fast-response dynamic decision-making for coal mining process scenario task identification and coal mining machine control strategies. It is set in the distributed decision layer (process control layer) with a response time of seconds or minutes.

[0024] Planning-oriented organization agent: Organizes multiple production stages in the coal mining process, achieving dynamic optimal decision-making for human-machine collaborative processes by determining dynamically optimal target planning schemes. For example, it uses optimal planning algorithms to sequence manual and system behaviors under dynamic targets and conditions such as gas content, geological conditions, coal extraction volume, advance speed, execution cost, and efficiency, enabling dynamic decision-making for slow-response human-machine collaborative processes in the coal mining process. It is set at the collaborative decision-making layer (process control layer), with a response time of minutes or hours. This embodiment of the invention sets up a global system-level planning-oriented organization agent and local equipment-level planning-oriented organization agents.

[0025] Planning-based arbitration agent: Utilizes resource, timing, and security conflict detection and adjustment algorithms to plan task sequence, ensuring the rationality and feasibility of the proposed task flow. Deployed at the collaborative decision-making layer (process control layer), with a response time of minutes or hours.

[0026] The planning organization agent includes a global system-level planning organization agent and multiple local device-level planning organization agents. Their functions will be described in detail in the following examples.

[0027] like Figure 1 As shown in the embodiment of the present invention, the multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes includes the following steps S1 to S4: S1, the global system-level planning organization agent plans the target output of the current production cycle based on the relationship between the historical output and the target output plan of the previous production cycle, and distributes the target output of the current cycle to multiple local equipment-level planning organization agents of the coal mining task process; among them, one scenario task corresponds to one local equipment-level planning organization agent and one local equipment-level planning arbitration agent.

[0028] Specifically, this embodiment of the invention employs a hierarchical reinforcement learning and goal planning method to construct a local device-level organizational planning agent and a global system-level organizational planning agent. The global system-level organizational planning agent is responsible for planning the goal tasks for each shift, day, week, month, and year, and distributes the planned goal tasks to the local device-level organizational planning agents.

[0029] In one specific embodiment, the global system-level planning organization agent plans the target output for the current production cycle based on the relationship between historical output and target output plan from the previous production cycle, including: S11, the global system-level planning organization agent determines the target output plan set for each shift, day, week, month, and year of the previous production cycle. And determine the set of actual completed quantities for each shift, day, week, month, and year of the previous production cycle. S12, Global system-level planning-type organizational agent calculation deviation vector set S13, The global system-level planning-type organizational agent modifies the target output plan for the current production cycle based on the deviation vector set to obtain the target output for the current production cycle. , is represented as: ; in, To adjust the coefficient matrix; This represents the correction rate for the corresponding level, and its value is an empirical value set by experts at each level, for example... The range is commonly [0.1-0.3], and can be flexibly adjusted according to the actual situation. Similarly.

[0030] It should be noted that if This indicates that the previous production cycle exceeded the target, so the target output will be automatically reduced in the next production cycle. If the actual output of the previous production cycle is lower than the planned target output, the target output will be automatically increased in the next production cycle. If the actual production volume of the previous production cycle is consistent with the planned target, then the target output plan for the next production cycle will remain unchanged.

[0031] The global system-level organizational planning agent summarizes the target output plan and actual execution results at each time level (shift, day, week, month, year) in each production cycle, and calculates the deviation vector. This deviation serves as feedback input, used to revise the target output plan set for the next production cycle, forming a closed-loop self-adjustment mechanism, thereby achieving dynamic output planning and adaptive optimization.

[0032] S2, each local device-level planning organization agent obtains its corresponding scenario task's four-dimensional objective optimization function and its reward function.

[0033] It should be noted that, before S2, this embodiment of the invention requires the construction of a four-dimensional objective optimization function and its reward function for each scenario task. Specifically, this embodiment of the invention constructs an Agent community organizational structure and designs a centralized-distributed hierarchical multi-timescale dynamic organizational strategy system architecture based on the different response time requirements of equipment control (milliseconds / seconds), process control (seconds / minutes), and process control (minutes / hours), such as... Figure 2 As shown, based on the principles of safety → efficiency → energy consumption → automation rate, execution priorities and resource weights are assigned to different scenario tasks (follow-up, straightening, support, etc.), and a four-dimensional objective optimization function and its reward function for each scenario task are constructed. The specific steps include: S0, based on the principles of safety, efficiency improvement, cost reduction, and intelligent coal mining with fewer personnel, predefines a four-dimensional objective function and its reward function for local equipment-level planning organization agents in different scenarios; The four-dimensional objective function Represented as ; in, This is a risk indicator function built based on safety procedures. This is an efficiency evaluation model constructed with man-hours per ton of coal as the core indicator. This is an energy consumption evaluation model constructed using electricity consumption per ton of coal and material consumption per ton of coal as core indicators. It is a mathematical function constructed with the automatic operation time of each coal mining system as the core indicator; The reward function V of the four-dimensional objective function F is expressed as: ; Here, w1 to w4 represent the weights of each reward function, which are dynamically adjusted using an attention gating mechanism based on various uncertainties in different task scenarios. Specifically, when... When the threshold is exceeded, w 4取极大值，其余为0 ; Represent reward functions in different dimensions.

[0034] Specifically, in a coal mining face scenario, there is equipment such as a coal mining machine, scraper conveyor, n hydraulic supports, a transfer crusher, and a belt conveyor. Based on the entire coal mining process of "crushing, loading, transporting, supporting, and processing," the following variables are set: Speed-related variables: coal mining machine speed, scraper conveyor operating speed, scraper conveyor advancing speed, hydraulic support following speed, and belt conveyor speed. The speed is constrained by the following values ​​due to process or mechanical limitations. Regarding the action strategies: Each hydraulic support has six main action strategies: lowering the column, moving the support, raising the column, pushing the slide, retracting the side guard plate, and extending the side guard plate. The action strategy of the nth hydraulic support is as follows: Regarding flow rate: Let the flow rate of the hydraulic system and the flow rate of the transfer crusher be respectively... The constraints are respectively and Regarding the safety assurance system: Based on safety regulations, let the actual situation of the following variables or propositions be a set, containing risk indicators such as gas content, wind speed, dust concentration, and whether workers are in safe areas. Let this set be denoted as . .

[0035] Based on the above, a risk indicator function is constructed according to safety regulations. ,include: First, the key safety provisions of the safety procedures are formalized into a set of indicators, and the actual value of each indicator in the set of indicators constitutes a risk indicator set. ,in This represents the true situation of the j-th risk indicator; Then, according to The properties construct binary judgment propositions or continuous risk level ; For binary decision propositions ,Sure The proposition is true under safe conditions. And expressed as: ; Taking whether the workers are in a safe area as an example, This indicates that the workers are in a safe area. This indicates that the workers are in a dangerous area; For continuous risk Continuous risk is measured based on the thresholds set in the safety regulations. The corresponding threshold is Then continuous risk level Represented as: ; by Taking gas content as an example, it belongs to the continuous risk level. If the threshold value is the gas content, then the continuous risk level is... Represented as: ; in, and It is an OR relationship, meaning that the same risk indicator cannot be counted twice; Next, determine the weights of the risk indicators. , , It can be determined through expert scoring or data fitting; Next, obtain the confidence level of the risk indicators. When data is missing or unreliable, reduce the weight contribution of the risk indicator; the confidence level of the risk indicator can also be determined by expert scoring or data fitting. Finally, the set of risk indicators Aggregate all risk indicators in the dataset, calculate the weighted risk score, and obtain... , is represented as: .

[0036] In one specific embodiment, the efficiency evaluation model constructed with man-hours per ton of coal (the man-hours required to produce one ton of coal) as the core indicator... The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate. The output is man-hours per ton of coal.

[0037] For example, let L be the working time per ton of coal, which is related to variables such as the speed of the coal mining machine, the operating speed of the scraper conveyor, the forward speed of the scraper conveyor, the following speed of the hydraulic support, the belt speed, the flow rate of the hydraulic system, and the flow rate of the transfer crusher. That is, the input to the efficiency evaluation model is... The output is Various data are collected from the coal mining face, cleaned, and then used to construct a neural network (such as an FNN or MLP) that can map the nonlinear relationship between the output and input. .

[0038] In one specific embodiment, the energy consumption evaluation model is constructed using electricity consumption per ton of coal and material consumption per ton of coal (the electricity and materials required to produce one ton of coal) as core indicators. The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate. The outputs are the power consumption per ton of coal and the material consumption per ton of coal.

[0039] For example, let E and G be the electricity and material consumption per ton of coal, respectively. These are related to variables such as the speed of the coal mining machine, the operating speed of the scraper conveyor, the forward speed of the scraper conveyor, the following speed of the hydraulic support, the belt speed, the flow rate of the hydraulic system, and the flow rate of the transfer crusher. In other words, the input to the efficiency evaluation model is... The output is Various data are collected from the coal mining face, cleaned, and then used to construct a neural network (such as an FNN or MLP) that can map the nonlinear relationship between the output and input. .

[0040] In one specific embodiment, a mathematical function is constructed using the automatic operation time of each coal mining system as the core indicator. ,include: First, determine the set of time proportions for automatic operation of each coal mining system. ,in, This represents the percentage of automatic running time for the i-th system. Next, determine the weight of the indicator for the percentage of automatic running time of the i-th system. , , ; It can be determined through expert scoring or data fitting; Then, determine the confidence level of the indicator for the percentage of automatic running time of the i-th system. When data is missing or unreliable, reduce the weight contribution of this indicator. Finally, for the time percentage set The weighted automation rate is calculated by aggregating the percentage of all automated running times. , represented as .

[0041] Because the target output is the top-level constraint for different task scenarios, the four-dimensional objective function decomposes the target output into four dimensions, planning the optimal task-connecting strategy. To achieve the target output within the expected timeframe, the four-dimensional objective function... Each element is indispensable because, regardless of the scenario, the task must be performed in a risk-free manner, ensuring both efficiency and automation, while also minimizing energy consumption, as excessive energy consumption would compromise economic benefits. By linking the target output with the four-dimensional objective function, it is possible to ensure that local device-level planning agents, when determining multiple multi-objective task strategies, can optimize the strategy while also considering the target output.

[0042] S3, each local device-level planning organization agent plans multiple multi-objective connection task strategies based on the four-dimensional objective optimization function and reward function of its corresponding scenario task using the MAPPO algorithm (Multi-Agent Proximal Policy Optimization), and then passes the multiple multi-objective connection task strategies to the corresponding local device-level planning arbitration agent.

[0043] In one specific embodiment, the local device-level planning organization agent re-plans the multi-objective connection task strategy every preset time using the MAPPO algorithm.

[0044] Specifically, the local equipment-level organization and planning agent is responsible for performing task organization and planning at preset intervals to obtain multi-objective task linkage strategies. These preset intervals are significantly longer than the time required for equipment actions and control. For example, if a hydraulic support requires 2-3 seconds to lower or raise its column, and 1-5 seconds to move it, then the local equipment-level organization and planning agent will plan at preset intervals. These preset intervals must be significantly longer than the planned time for equipment actions and control (in this example, longer than the sum of the two) to avoid starting the next planning cycle before the current multi-objective task linkage strategy is completed.

[0045] When planning multi-objective association task strategies, the local device-level organizational planning agent uses reinforcement learning algorithms. Based on the examples above, the reinforcement learning actions of the local device-level organizational planning agent during strategy planning can be represented as: multi-objective association task strategy. .

[0046] When planning strategies for local equipment-level organizational planning agents, the reinforcement learning state can be characterized using multi-sensor equipment on the longwall mining face. In addition to the pre-set parameters, each hydraulic support is equipped with a hydraulic support stroke sensor, pressure sensor, attitude angle sensor, and support height sensor. The variables for the nth hydraulic support are as follows: The state can then be characterized as: ; In state After actions (multi-objective linked task strategy) After feedback, a new state will be reached. , is represented as: ; in, The target output for the current cycle is planned by the global system-level organizational planning agent, and the target output received in a production cycle is the same.

[0047] At this point, the parameters for the reinforcement learning portion of the local device-level organizational planning agent have been set, and the MAPPO algorithm can be used to formulate multi-agent connection action strategies. The plan.

[0048] S4, the local device-level planning arbitration agent detects whether there are conflicts among the multiple multi-objective connection task strategies. If each multi-objective connection task strategy has a conflict, the detection result is fed back to the corresponding local device-level planning organization agent. The local device-level planning organization agent modifies some constraint parameters of the four-dimensional objective function under the corresponding scenario task and re-plans the multi-objective connection task strategy until a conflict-free multi-objective connection task strategy is obtained. Then, the local device-level planning organization agent calculates the optimal multi-objective connection task strategy based on the arbitration strategy of cooperative game based on the corresponding reward function and issues the optimal multi-objective connection task strategy to the agent alliance.

[0049] Step S4 primarily involves constructing a multi-constraint conflict resolution model for a local device-level planning arbitration agent. Specifically, considering the potential resource conflicts, timing conflicts, and security conflicts that may arise during the actual execution of multi-objective connection task strategies (examples of these three types of conflicts are shown in Table 1), this embodiment designs different constraint conditions (symbolic rules or data models) to determine whether a conflict has occurred. A constraint reasoning algorithm is used to automatically detect conflict points in multi-objective connection task strategies, and a rule-based arbitration strategy is designed. If all multi-objective connection task strategies have conflicts, feedback is sent to the local device-level planning organization agent, modifying some constraints of the four-dimensional objective function (e.g., adjusting the threshold) to force a replanning decision. Furthermore, this embodiment also designs an arbitration strategy based on cooperative game theory, that is, based on the reward function of the task in this scenario, selecting the strategy that maximizes the benefit for policy distribution. Forced replanning decision means that once a conflict is identified, the local device-level planning organization agent is forced to replan its decision, while simultaneously modifying some constraints in its four-dimensional objective function.

[0050]

[0051] In one specific embodiment, when the local device-level planning arbitration agent detects whether there is a conflict in any multi-objective connection task strategy, it includes: S41, the local device-level planning arbitration agent predefines preset constraints for resource conflicts, timing conflicts, and security conflicts; S42, the local device-level planning arbitration agent determines whether there is a resource conflict, timing conflict, or security conflict in any multi-objective connection task strategy according to the preset constraints. If any one of the conflicts exists (any one of resource conflict, timing conflict, or security conflict), then it is determined that the multi-objective connection task strategy has a conflict.

[0052] Specifically, the preset constraints for resource conflicts include hydraulic system flow constraints. If the total flow consumed by the hydraulic support cluster in performing its actions exceeds the hydraulic system flow threshold, i.e. If the total flow consumed is less than or equal to the hydraulic system's flow threshold, then the hydraulic system's power is considered to be exceeding the limit. If the total flow consumed is less than or equal to the hydraulic system's flow threshold, then the hydraulic system's power is considered to satisfy the multi-objective linkage task strategy. Let the presence or absence of conflict in the hydraulic system be... Then it can be expressed as: ; in, , , and These represent the flow rates required for the hydraulic support cluster to lower, move, raise, and push the support column, respectively. Indicates the flow threshold of the hydraulic system; Indicates the resource conflict detection results; This indicates no conflict. This indicates a conflict.

[0053] Furthermore, regarding timing conflicts, the existence of a conflict is determined based on its preset constraints. Taking the action of hydraulic supports near the coal mining machine as an example, from a technological perspective, it is unreasonable for the hydraulic supports in front of the coal mining machine to move. This embodiment of the invention uses the point logic time relationship matrix defined in the patent "Method for Timing Planning of Hydraulic Support Actions Following the Coal Mining Process" to restrict the preset constraints of timing conflicts. Specifically, the preset constraints of timing conflicts include performing time planning calculations on the point logic time relationship matrix M of the hydraulic support actions to obtain several R_timetables that satisfy all time constraints of the action event set. Each R_timetable is then verified. If the verification passes, it indicates no conflict; otherwise, it indicates a conflict, expressed as: ; in, This indicates that the verification failed. This indicates that the verification has passed. This indicates the timing conflict detection result. This indicates no conflict. This indicates a conflict.

[0054] The preset constraints for the safety conflict include a constraint on whether the gas level exceeds the limit, expressed as: ; in, Real-time gas content, The gas exceedance threshold; Indicates the results of security conflict detection. This indicates no conflict. This indicates a conflict.

[0055] Of course, the preset constraints for resource conflicts and security conflicts mentioned above may also include other contents, and the embodiments of the present invention do not further limit them.

[0056] This invention designs an arbitration strategy based on cooperative game theory. Specifically, based on the reward function of the task scenario, a cooperative game payoff model is constructed for the agent alliance to maximize its payoff. In other words, under the premise of no conflict, the payoff of the agent alliance composed of multiple agents is maximized based on the reward function of the current task scenario. Assume that the local device-level organizational agents can simultaneously generate N multi-agent connection action strategies. The value of each multi-agent connection action strategy is recorded. Under the premise of no conflict, the multi-agent connection action strategy with the highest value is selected for distribution. Mathematically, this can be expressed as: Specifically, for details on the implementation of the multi-objective linkage task strategy that determines the optimal value, please refer to the relevant content in the patent "Multi-Agent Collaborative Control Decision Method for Hydraulic Supports".

[0057] In one specific embodiment, in step S4, the local equipment-level planning organization agent issues the optimal multi-objective connection task strategy to the agent alliance, including: the local equipment-level planning organization agent issues the optimal multi-objective connection task strategy to the reasoning model agent; the reasoning model agent performs dynamic decision reasoning on the optimal multi-objective connection task strategy, transforming the optimal multi-objective connection task strategy into specific and implementable equipment execution instructions, and then issues the equipment execution instructions to the reflective program agent; the reflective program agent controls the coal mining entity equipment to execute the equipment execution instructions.

[0058] It should be noted that both the inference model agent and the reflection program agent are local device-level. One inference model agent and one reflection program agent correspond to one scenario task.

[0059] Taking the scenario task "straightening all hydraulic supports of the 31001 longwall mining face" as an example, the local equipment-level planning agent, based on the four-dimensional objective optimization function and its reward function, uses the MAPPO algorithm to plan multiple multi-objective connection task strategies. After the local equipment-level planning arbitration agent arbitrates and determines that there are no conflicts among the multi-objective connection task strategies, it determines the multi-objective connection task strategy with the optimal value and issues it to the reasoning model agent. The reasoning model agent performs dynamic decision-making reasoning on the multi-objective connection task strategy with the optimal value, deciding on the equipment execution command for all hydraulic supports of the 31001 longwall mining face to execute the scenario task, such as moving the first hydraulic support 1m. Then, the equipment execution command is sent to the reflective program agent of each hydraulic support, which executes the equipment execution command to complete the scenario task of straightening all hydraulic supports of the 31001 longwall mining face.

[0060] Taking the specific process of a coal mining face following machine operation as an example, the overall framework of this embodiment of the invention is as follows: Figure 3 As shown.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes, characterized in that, The method includes: S1, the global system-level planning organization agent plans the target output of the current production cycle based on the relationship between the historical output and the target output plan of the previous production cycle, and distributes the target output of the current cycle to multiple local equipment-level planning organization agents of the coal mining task process; among them, one scenario task corresponds to one local equipment-level planning organization agent and one local equipment-level planning arbitration agent. S2, each local device-level planning organization agent obtains its corresponding scenario task's four-dimensional objective optimization function and its reward function; S3, each local device-level planning organization agent uses the MAPPO algorithm to plan multiple multi-objective connection task strategies based on the four-dimensional objective optimization function and reward function of its corresponding scenario task, and then passes the multiple multi-objective connection task strategies to the corresponding local device-level planning arbitration agent; S4, the local device-level planning arbitration agent detects whether there is a conflict among the multiple multi-objective connection task strategies. If there is a conflict among the multiple multi-objective connection task strategies, the detection result is fed back to the corresponding local device-level planning organization agent. The local device-level planning organization agent modifies some constraint parameters of the four-dimensional objective function under the corresponding scenario task and re-plans the multi-objective connection task strategy until a conflict-free multi-objective connection task strategy is obtained. Then, the local device-level planning organization agent calculates the optimal multi-objective connection task strategy based on the arbitration strategy of cooperative game based on the corresponding reward function and issues the optimal multi-objective connection task strategy to the agent alliance. The local device-level planning arbitration agent, when detecting whether any multi-objective connection task strategy has a conflict, includes: S41, the local device-level planning arbitration agent predefines preset constraints for resource conflicts, timing conflicts, and security conflicts; S42, the local device-level planning arbitration agent determines whether any multi-objective connection task strategy has a resource conflict, timing conflict, or security conflict based on the preset constraints. If any conflict exists, it is determined that the multi-objective connection task strategy has a conflict. The preset constraints for resource conflicts include hydraulic system flow constraints, expressed as: ； in, and These represent the flow rates required for the hydraulic support cluster to lower, move, raise, and push the support column, respectively. Indicates the flow threshold of the hydraulic system; Indicates the resource conflict detection results; This indicates no conflict. This indicates a conflict; The preset constraints for the timing conflict include performing time planning calculations on the point logic time relationship matrix M of the hydraulic support actions to obtain several R_timetables that satisfy all time constraints of the action event set. Each R_timetable is then verified. If the verification passes, it indicates no conflict; otherwise, it indicates a conflict, expressed as follows: ； in, This indicates that the verification failed. This indicates that the verification has passed. This indicates the timing conflict detection result. This indicates no conflict. This indicates a conflict; The preset constraints for the safety conflict include a constraint on whether the gas level exceeds the limit, expressed as: ； in, Real-time gas content, The gas exceedance threshold; Indicates the results of security conflict detection. This indicates no conflict. This indicates a conflict.

2. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 1, characterized in that, Before S2, it also includes: S0, based on the principles of safety, efficiency improvement, cost reduction, and intelligent coal mining with fewer personnel, predefines a four-dimensional objective function and its reward function for local equipment-level planning organization agents in different scenarios; The four-dimensional objective function Represented as ; in, This is a risk indicator function built based on safety procedures. This is an efficiency evaluation model constructed with man-hours per ton of coal as the core indicator. This is an energy consumption evaluation model constructed using electricity consumption per ton of coal and material consumption per ton of coal as core indicators. It is a mathematical function constructed with the automatic operation time of each coal mining system as the core indicator; The reward function V of the four-dimensional objective function F is expressed as: ； Where w1 to w4 are the weights of each reward function. Represent reward functions in different dimensions.

3. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 2, characterized in that, Construct risk indicator functions based on safety regulations ,include: First, the key safety provisions of the safety procedures are formalized into a set of indicators, and the actual value of each indicator in the set of indicators constitutes a risk indicator set. ,in This represents the true situation of the j-th risk indicator; Then, according to The properties construct binary judgment propositions or continuous risk level ; For binary decision propositions ,Sure The proposition is true under safe conditions. And expressed as: ； For continuous risk Continuous risk is measured based on the thresholds set in the safety regulations. The corresponding threshold is Then continuous risk level Represented as: ； in, and It is an OR relationship, meaning that the same risk indicator cannot be counted twice; Next, determine the weights of the risk indicators. , , ; Next, obtain the confidence level of the risk indicators. ; Finally, the set of risk indicators Aggregate all risk indicators in the dataset, calculate the weighted risk score, and obtain... , is represented as: 。 4. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 2, characterized in that, The efficiency evaluation model constructed with man-hours per ton of coal as the core indicator The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate; the output is man-hours per ton of coal. The energy consumption evaluation model constructed with electricity consumption per ton of coal and material consumption per ton of coal as core indicators. The inputs include the coal mining machine speed, scraper conveyor operating speed, scraper conveyor advance speed, hydraulic support following speed, belt speed, hydraulic system flow rate, and transfer crusher flow rate. The outputs are the power consumption per ton of coal and the material consumption per ton of coal.

5. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 2, characterized in that, Mathematical functions are constructed using the automatic operation time of various coal mining systems as the core indicator. ,include: First, determine the set of time proportions for automatic operation of each coal mining system. ,in, This represents the percentage of automatic running time for the i-th system. Next, determine the weight of the indicator for the percentage of automatic running time of the i-th system. , , ; Then, determine the confidence level of the indicator for the percentage of automatic running time of the i-th system. ; Finally, for the time percentage set The weighted automation rate is calculated by aggregating the percentage of all automated running times. , represented as 。 6. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 1, characterized in that, The local device-level planning organization agent re-plans the multi-objective connection task strategy every preset time using the MAPPO algorithm.

7. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 1, characterized in that, The global system-level planning organization agent plans the target output for the current production cycle based on the relationship between historical output and target output plans from the previous production cycle, including: S11, the global system-level planning organization agent determines the target output plan set for each shift, day, week, month, and year of the previous production cycle. And determine the set of actual completed quantities for each shift, day, week, month, and year of the previous production cycle. ; S12, Global System-Level Planning Organization Agent Calculation Deviation Vector Set ; S13, the global system-level planning organization agent modifies the target output plan for the current production cycle based on the deviation vector set, and obtains the target output for the current production cycle. , is represented as: ； in, To adjust the coefficient matrix; , which represents the correction rate for the corresponding level.

8. The multi-objective, multi-constraint hybrid decision-making agent planning method for coal mining processes according to claim 1, characterized in that, In step S4, the local device-level planning organization agent issues the optimal multi-objective connection task strategy to the agent alliance, including: Local device-level planning-type organizational agents will issue the most valuable multi-objective interconnected task strategy to reasoning-type model agents; The reasoning model agent performs dynamic decision-making reasoning on the value-optimal multi-objective connection task strategy, transforms the value-optimal multi-objective connection task strategy into specific and implementable device execution instructions, and then sends the device execution instructions to the reflective program agent. The reflective program agent controls the coal mining equipment to execute the equipment execution instructions.

Images