Slurry leakage blocking strategy generation method and system based on coal mine separation layer grouting

Abstract

The invention relates to the technical field of coal mine intelligent grouting control and scheduling, and discloses a method and system for generating a slurry leakage blocking strategy based on coal mine separation layer grouting, and the method comprises the steps: collecting grouting state parameters through a multi-source sensing device, constructing a state representation vector, and building a weighted Petri net model; based on the historical state representation vector, extracting a long-term trend component and a short-term residual component, and fusing to generate a prediction state representation vector; the system fuses the current state information and the predicted state information, calculates the scheduling score of each transfer position, and selects the transfer position with the highest score to execute plugging operation; and dynamically correcting a state vector and updating a weighting function based on grouting feedback information to realize adaptive evolution of scheduling logic. The accuracy and execution efficiency of plugging decision making are improved, and the problems of resource coordination and dynamic control under the complex grouting working condition are effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent grouting control and scheduling in coal mines, in particular to a running grouting plugging strategy generation method and system based on coal mine separation layer grouting. BACKGROUND

[0002] In the process of coal mining, due to the complex structure of strata and frequent mining disturbance, hidden danger areas such as rock separation layer and fracture channel are easily formed. If grouting plugging is not timely or the method is not reasonable, the grout may run out of the predetermined plugging area along the fracture, forming the "running grouting" phenomenon, which not only causes waste of materials and rise of treatment cost, but also affects subsequent roadway support and safety production, and even induces water inrush, roof fall and other disaster accidents. At present, coal mine separation layer grouting mainly relies on manual experience to determine grouting parameters and plugging path planning, and lacks effective modeling and prediction means for complex geological changes and dynamic scheduling state, so it is difficult to cope with the plugging scheduling problem under the interaction of multiple factors, and shows limitations such as low intelligence level, slow response and rigid scheduling.

[0003] Existing research attempts to introduce rule base, expert system or simple logic model to assist in optimizing grouting scheduling, but there are problems such as rule lag and inability to adapt to site changes, and lack of dynamic modeling ability for state evolution of grouting process. In recent years, Petri net has been gradually applied to manufacturing execution, network control and other fields due to its good visual modeling and scheduling logic expression ability. However, when facing complex evolution of plugging state and dynamic changes of resource constraints, the static structure and fixed scheduling strategy of traditional Petri net model are difficult to support flexible plugging scheduling under complex working conditions.

[0004] Therefore, it is urgent to build a new plugging decision method integrating state perception, trend prediction and self-adaptive evolution mechanism to realize accurate identification of running grouting risk and intelligent dynamic scheduling of plugging path, so as to improve the safety, reliability and resource allocation efficiency of coal mine grouting treatment. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the technical problem solved by the present application is that the existing running grouting risk identification in the process of coal mine separation layer grouting is not accurate, the plugging scheduling response is not timely, and there is a lack of dynamic self-adaptive mechanism.

[0007] To solve the above technical problems, the present application provides the following technical scheme: a running grouting plugging strategy generation method based on coal mine separation layer grouting, comprising: collecting a set of grouting state parameters through a multi-source perception device, and constructing a state representation vector based on the set of grouting state parameters;

[0008] construct a weighted Petri net model based on the state representation vector;

[0009] extract a long-term trend component and a short-term residual component based on the historical state representation vector, and fuse to generate a predicted state representation vector;

[0010] input the predicted state representation vector and the state representation vector into the scheduling score calculation of each transition position;

[0011] select the transition position with the highest scheduling score, and execute the plugging scheduling operation;

[0012] Collect the execution feedback information of the plugging task, update the state representation vector and the weighted Petri net model, and realize the adaptive evolution of the scheduling logic.

[0013] As a preferred scheme of the coal mine off-layer grouting run-out plugging strategy generation method, wherein: the grouting state parameter set includes grouting process parameters, geological structure parameters, grouting material inventory information, grouting equipment running state, operation personnel scheduling state and historical plugging data;

[0014] The grouting state parameter set is preprocessed, and after preprocessing, the state representation vector of the node is constructed according to the grouting state parameter set;

[0015] The state representation vector of the node includes: based on the grouting state parameter set, establishing a parameter combination judgment condition, and according to the assignment rule, the corresponding features are divided into different levels; the discrete characteristic value of the run-out risk, the discrete characteristic value of the plugging complexity, the discrete characteristic value of the resource capacity and the discrete characteristic value of the operation availability are combined to form the state representation vector of the node;

[0016] The assignment rule includes: when all conditions are met at the same time, the assignment is 1; when part of the conditions are met, the assignment is 0.5; when no condition is met, the assignment is 0;

[0017] The assignment of the discrete characteristic value includes: executing the judgment based on the combination threshold condition set by the grouting state parameter set, and assigning the judgment result according to the assignment rule.

[0018] As a preferred scheme of the coal mine off-layer grouting run-out plugging strategy generation method, wherein: the weighted Petri net model is based on the state representation vector of the node, and constructs the dynamic mapping relationship G=(S,T,E,Token,W) between the task state and the behavior transition, wherein S represents the state bit set; T represents the transition bit set; E represents the directed edge set; Token represents the dynamic marking information corresponding to the state bit; W represents the weight function set.

[0019] As a preferred solution of the method for generating a slurry leakage blocking strategy based on coal mine separation grouting according to the present invention, wherein: during the operation of the weighted Petri net model, the dynamic tag information Token (S i ), combined with the weight function set W, calculate the transfer bit T corresponding to the state bit j The scheduling score DS(T j ), the formula is:

[0020]

[0021] Among them, W ij Indicates status bit S i For the transfer bit T j The influence strength of , and j represents the index; F(M(P i )) is a correction function, which represents the weighted fusion of the state representation vector including the slurry leakage risk value, the complexity of the plugging operation, the resource capability score and the operation availability score;

[0022] According to the scheduling score results of each transfer bit, in the current scheduling cycle, the transfer bit T with the highest score is selected from the transfer bit set T. * =argmax(DS(T j )), triggering the corresponding blocking action; where T * Indicates the optimal transfer selected for execution in the current scheduling cycle; argmax indicates finding the transfer bit with the highest score from all transfer bits; DS(T j ) represents the shift bit T j The scheduling score value of

[0023] When the status bit S i When the corresponding node is insufficient in resources or the scheduling score is lower than the set execution threshold for m consecutive scheduling cycles and no connected transfer bit is triggered, a bridge transfer bit T is automatically generated. bridge , connection status bit S m With status bit S i ; and add a bridge path S in the weighted Petri net model m →T bridge →S i ; Collect feedback information during the plugging execution process to correct the state representation vector of the corresponding node; update the weight function set W, increase the weight of the successful path, and reduce the weight of the failed path; the feedback information includes the deviation between the actual grouting volume and the predicted value, the regional pressure recovery rate, the operation response delay and the equipment stability;

[0024] After every N scheduling cycles, the average trigger success rate of each transfer bit in the last L scheduling cycles is counted; when the activity of each transfer bit is lower than the activity threshold, or the average trigger success rate of each transfer bit is lower than the set success rate threshold in c consecutive scheduling cycles, the transfer bit is removed from the transfer set.

[0025] As a preferred solution of the method for generating a slurry leakage blocking strategy based on coal mine separation grouting according to the present invention, the extraction of the long-term trend component and the short-term residual component includes: constructing a state vector sequence of fixed length based on the historical state representation vector of the node in each scheduling cycle; performing a sliding weighted average process on the state vector sequence to obtain the long-term trend component of each feature dimension; defining the difference between the current state vector and the long-term trend component as the short-term residual component;

[0026] The fusion generation of the prediction state representation vector includes: calculating the fusion weight coefficient based on the absolute value of the long-term trend component and the short-term residual component in each feature dimension; using the fusion weight coefficient, performing a linear combination of the long-term trend component and the short-term residual component of each feature dimension to obtain a fusion prediction value; splicing all fusion prediction values ​​into a fusion prediction vector in the order of feature dimensions, and inputting the fusion prediction vector into a preset fully connected layer to generate a prediction state representation vector for the next scheduling period.

[0027] As a preferred solution of the method for generating a slurry leakage blocking strategy based on coal mine separation grouting according to the present invention, the calculation of the scheduling score of each transfer position includes obtaining the current state representation vector and the corresponding predicted state representation vector of each node in each scheduling cycle, and constructing a fused state representation vector based on a weighted fusion method;

[0028] The scheduling score value of each transfer bit is calculated by using the weight function relationship between each state bit and each transfer bit, where the score value of each transfer bit is calculated jointly by the fused state representation vector corresponding to the state bit connected to the transfer bit and the weight function; the scheduling score values ​​of the transfer bits are sorted, and the transfer bit with the highest score value is selected as the blocking behavior action to be executed in the current scheduling cycle.

[0029] As a preferred scheme of the coal mine off-bed grouting based on the running slurry plugging strategy generation method, wherein: the update state representation vector and the weighted Petri net model include collecting feedback information after completing the plugging operation, correcting the state representation vector of the node, and updating the weight function set in the weighted Petri net model, increasing the weight of the successful scheduling path and reducing the weight of the failed scheduling path; when the state bit is not scheduled in the continuous preset scheduling period, and the scheduling score of all the transition bits connected by the state bit is lower than the score threshold, the bridging mechanism is triggered to generate a bridging transition bit and a bridging edge to connect the state bit with rich resources and the unscheduled state bit, and a temporary bridging path is constructed.

[0030] A coal mine off-bed grouting based on the running slurry plugging strategy generation system, wherein: a data module acquires a set of grouting state parameters through a multi-source sensing device, and constructs a state representation vector based on the set of grouting state parameters;

[0031] A model module constructs a weighted Petri net model based on the state representation vector;

[0032] A prediction module extracts a long-term trend component and a short-term residual component based on a historical state representation vector, and fuses to generate a predicted state representation vector;

[0033] A scoring module calculates the scheduling score of each transition bit by taking the predicted state representation vector and the state representation vector as input;

[0034] An execution module selects the transition bit with the highest scheduling score to perform the plugging scheduling operation;

[0035] A feedback module collects the execution feedback information of the plugging task, updates the state representation vector and the weighted Petri net model, and realizes the adaptive evolution of the scheduling logic.

[0036] A computer device, comprising: a memory and a processor; the memory stores a computer program, characterized in that: the processor executes the computer program to realize the steps of the method of any one of the present application.

[0037] A computer readable storage medium having a computer program stored thereon, characterized in that: the computer program is executed by a processor to realize the steps of the method of any one of the present application.

[0038] The coal mine off-layer grouting-based running slurry plugging strategy generation method provided by the application fuses multi-source sensing data to construct a state representation vector, dynamically maps nodes and behavior paths by combining weighted Petri net modeling, predicts the state representation vector based on the historical state representation vector, and calculates the transition score based on the fused state and optimizes the execution path within the scheduling period, effectively improving the real-time performance and accuracy of plugging scheduling. After scheduling, task feedback information is collected, the state vector and the weight function are dynamically corrected, the structure is continuously evolved and adaptively optimized, and the plugging response capability and resource regulation efficiency of the system under complex geological conditions are significantly enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 The overall flowchart of the coal mine off-layer grouting-based running slurry plugging strategy generation method provided by the first embodiment of the application. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned purposes, features and advantages of the application more apparent and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0042] Embodiment 1, refer to Figure 1 For an embodiment of the application, a coal mine off-layer grouting-based running slurry plugging strategy generation method is provided, which comprises:

[0043] S1: Collect the grouting state parameter set through the multi-source sensing device, and construct a state representation vector based on the grouting state parameter set.

[0044] In the system running process, first, based on the mine construction drawing, the geological structure model, the historical grouting record and the real-time monitoring data, the plugging scheduling object node set is identified, denoted as:

[0045] P={P1,P2,...,P n}

[0046] Each node P iCorresponding to a specific space grouting unit, it can be identified from the following sources: (1) the grouting hole coordinates laid down in the mine and the roadway distribution information in the design drawings; (2) the fracture zone, fault intersection area and hydraulic anomaly area identified in the geological model; (3) the typical slurry running area appearing multiple times in the historical plugging data; (4) the area aggregation point where the pressure suddenly changes and the flow rate rises in the current monitoring and scheduling period. The system will aggregate the above sources to form a node set P as the basic target object for scheduling and modeling.

[0047] In the system operation, a variety of sensing devices laid in the mine and the ground control terminal are used to collect grouting state parameter sets, including grouting process parameters, geological structure parameters, grouting material inventory information, grouting equipment operating status, operation personnel scheduling status and historical plugging data.

[0048] The grouting process parameters include grouting pressure and grouting flow rate, which are collected by the pressure sensor and flow sensor connected to the grouting pump in real time; the geological structure parameters include the fracture rate and porosity of the rock stratum, which are obtained from geological exploration data or acoustic logging devices; the grouting material inventory information is obtained by real-time updating through the Internet of Things storage module; the grouting equipment operating status includes the start-stop state of the grouting pump, the current load and the equipment fault code, which are obtained from the PLC or DCS control interface; the operation personnel scheduling status is determined by the linkage of the underground positioning system and the task assignment system to determine the availability and geographical distribution of the current operation personnel; the historical plugging data include the success rate of plugging, the execution record in similar scenarios, the plugging time and effect evaluation index, which are obtained by reading from the grouting construction database.

[0049] After the preprocessing of the grouting state parameter set is completed, the system constructs four types of state feature values of the node based on the normalized parameter information of each type, including slurry running risk, plugging complexity, resource capacity and operation availability, all of which are quantized in a {0, 0.5, 1} three-value discrete manner. They are used to represent the state level of different plugging nodes. The determination method of each feature value is as follows:

[0050] Among them, the determination of slurry running risk is based on three types of indicators: grouting process parameters, geological structure parameters and historical plugging data. The system determines whether the following three conditions are met at the same time: the change rate of grouting pressure is lower than the set threshold, the real-time flow rate is higher than the threshold corresponding to the historical average value, and the historical grouting average value exceeds the reference benchmark; or the sum of fracture rate and porosity exceeds the preset threshold and the number of historical plugging failures exceeds the limited number. If any complete combination condition is established, the slurry running risk is assigned a value of 1; if only one of the above conditions is met, the value is 0.5; otherwise, the value is 0.

[0051] The determination of the plugging complexity is based on the geological structure parameters and historical plugging data. The system determines whether the following three conditions are met simultaneously: the number of lithology types reaches the set complexity standard, the number of small-angle fractures in the fracture intersection angle distribution exceeds the set number, and the average plugging time exceeds the limited time length or the plugging success rate is lower than the set level. If any complete combination condition is established, the plugging complexity is assigned a value of 1; if only one of the conditions is met, the value is 0.5; otherwise, the value is 0.

[0052] The determination of the resource capacity relies on the grouting material inventory information and the operation state of the grouting equipment. The system determines whether the following two conditions are met simultaneously: the ratio of material inventory to demand is higher than the set proportion, and the equipment load is in the set range and there is no fault information at present. If both conditions are met, the resource capacity is assigned a value of 1; if only one of the conditions is met, the value is 0.5; otherwise, the value is 0.

[0053] The determination of the operation availability is composed of the operation personnel scheduling state and the response index in the historical plugging data. The system determines whether the following two conditions are met simultaneously: the ratio of the number of currently callable operation personnel to the scheduling demand is greater than the set standard, and the historical average scheduling response time delay is lower than the limited time threshold. If both conditions are met, the operation availability is assigned a value of 1; if only one of the conditions is met, the value is 0.5; otherwise, the value is 0.

[0054] According to the above four types of state characteristic values, a state representation vector for describing the state of the node is constructed, denoted as:

[0055] M(P i )=[R i ,D i ,C i ,A i ]

[0056] Wherein, R i represents the discrete characteristic value of the grouting risk of node P i , which is calculated by weighting and fusing the pressure fluctuation coefficient calculated based on pressure sensor data, the fracture rate of the position, and the historical plugging failure records corresponding to the node; D i represents the discrete characteristic value of the plugging complexity of node P i , which is calculated by weighting according to the construction environment parameters and structure arrangement features of the region it is located in, combined with the structure difficulty level marked in the structure design drawing; C i represents the discrete characteristic value of the resource capacity of node P i , which is obtained by normalizing and superimposing the remaining grout volume, the number of schedulable grouting pumps, and the pipe network unobstructed state; A i represents the discrete characteristic value of the operation availability of node P idiscrete eigenvalues of the job availability, is generated by combining the number of on-site workers in real time and the online state of the grouting equipment with the scheduling shift plan. M(P i ) represents the state vector of node P i ; P i represents the i-th grouting node, and i represents the node index.

[0057] By introducing multi-source sensing devices, the system can comprehensively collect core parameters in the grouting process, including grouting pressure, flow rate, rock structure characteristics, material inventory and operation state, etc., greatly enhancing the comprehensive perception ability of the actual working conditions of the construction site. On this basis, the system constructs a standardized grouting state parameter set, and realizes high consistency processing of data through time alignment, normalization and outlier correction, significantly improving the data quality and real-time performance. The finally generated state representation vector covers multi-dimensional indicators such as grouting risk, plugging complexity, resource capacity and job availability, providing structured and high-precision foundation support for subsequent modeling and scheduling.

[0058] Most of the existing grouting scheduling schemes rely on manual experience or single sensing parameters, which are difficult to effectively identify key nodes under complex geological conditions, and have low state perception and data update lag. In comparison, by integrating design drawings, real-time monitoring and historical data, the plugging scheduling target is comprehensively identified from the aspects of spatial distribution and risk trend, a dynamic node set is formed, and a high-dimensional state vector is constructed based on the standardized processing results, realizing comprehensive quantification and accurate modeling of the state of the grouting unit, effectively breaking through the traditional bottlenecks of "ambiguous node target identification" and "discrete parameters cannot be modeled".

[0059] By constructing a multi-dimensional state representation vector of the node, the system systematically integrates key indicators such as pressure fluctuation coefficient, fracture rate, structural complexity, resource supply capacity and operation accessibility, which significantly improves the completeness, interpretability and computability of node state modeling compared to traditional schemes that rely only on local parameters or static models. At the same time, the state vector is directly input to the subsequent Petri net modeling and scheduling decision-making, effectively establishing a closed-loop control link from "perception - modeling - decision", promoting the fundamental transformation of the grouting scheduling system from experience-driven to data-driven.

[0060] S2: Construct a weighted Petri net model based on the state representation vector.

[0061] After completing the node state representation vector M(P i )=[R i ,D i ,C i ,A iAfter the construction of the state representation vector, a weighted Petri net model for the plugging task modeling and scheduling control is further constructed based on the state representation vector. The weighted Petri net model is denoted as G=(S, T, E, Token, W), wherein S represents a state bit set, used to describe the current state of the node; T represents a transition bit set, describing the behavior event in the plugging scheduling process; E represents a directed edge set, defining the logical connection and resource dependence between the state bit and the transition bit; Token represents dynamic marking information corresponding to the state bit; and W represents a weight function set, used to define the transition trigger condition, resource consumption relationship and scheduling priority calculation method.

[0062] The system is based on the node set P, and for each node P i , a corresponding state bit S i is created, and the state representation vector is taken as the state bit Token:

[0063] Token(S i )=M(P i )=[R i ,D i ,C i ,A i ]

[0064] Wherein S i represents the i-th state bit corresponding to the node P i ; M(P i ) represents the state representation vector of the node P i ; and Token(S i ) represents that the state representation vector M(P i ) is taken as the marking of the state bit S i .

[0065] In order to depict the dynamic influence of the node state on the scheduling behavior, based on the relationship between the aforementioned state representation vector M(P i ) and the transition behavior unit T j , a global state weight matrix W is constructed, wherein the matrix element W i,j represents the influence intensity of the current state of the node P i on the scheduling behavior T j ; and the formula of W i,j is represented as:

[0066] W i,j =f(M(P i ),T j )

[0067] Wherein f(·) represents a scheduling weight calculation function, which maps the influence intensity according to the state and behavior; and T j represents the j-th transition behavior unit, such as "starting grouting" and "calling resources".

[0068] The global state weight matrix dynamically represents the scheduling driving force of each node for a specific transition action, i.e. the priority tension of scheduling.

[0069] The system constructs a corresponding transition set T according to the possible behaviors of the plugging operation (such as starting plugging, resource allocation, plugging interruption) j . But unlike the static judgment of transition triggering in the traditional method, this method defines a scheduling score function DS(T j ) for each transition, which is dynamically calculated based on the weight matrix:

[0070]

[0071] DS(T j ) represents the scheduling score value of the transition T j .

[0072] In the scoring calculation process, in order to map the multi-dimensional state representation vector M(P i ) to a single value result that can be used for transition scoring, a correction function F(M(P i )) is introduced.

[0073] F(M(P i )) performs weighted fusion on the running risk value, plugging operation complexity, resource capacity score and operation availability score in the state representation vector, and the formula is expressed as:

[0074] F(M(P i )) = α1·R i + α2·D i + α3·C i + α4·A i

[0075] Wherein α1, α2, α3, α4 represent weight coefficients; the correction function is used to compress multi-dimensional state information into scheduling score input, ensuring that the scoring mechanism has implementability and parameter controllability.

[0076] All transition sets are sorted according to the competition score, and the system executes the transition set with the highest score in each scheduling period:

[0077] T * = argmax DS(T j )

[0078] Wherein T * represents the optimal transition to be executed in the current scheduling period; argmax represents finding the one with the highest score from all transition sets; DS(T j ) represents the scheduling score value of the transition T j .

[0079] To establish the connection structure between state bits and transition bits, the system associates state Token with transition action logic through a directed edge set E, and each edge is assigned a weight value representing the corresponding resource consumption or state threshold. Further, the system adds an adaptive bridging mechanism. When a node state bit S i When the long-term low scheduling priority is in resource shortage or state degradation, the system automatically triggers the bridging rule to dynamically generate a temporary transition bit T bridge connecting the node and the surrounding resource sufficient node. bridge The weight function set W dynamically adjusts the bridging resource call cost and priority, so that the system structure evolves dynamically with state feedback, and the formula is as follows:

[0080] S m →T bridge →S i

[0081] Among them, S m represents a resource sufficient or state good auxiliary node; T bridge represents the dynamically generated bridging transition of the system, which is used to temporarily support the blocked resources; S i represents the node currently in resource shortage or scheduling block.

[0082] After the above transition operation, the system enters the state feedback phase. At this time, the feedback information in the blocking execution process is collected, including: the deviation between the actual grouting amount and the estimated value; the grouting area pressure recovery rate; the actual response time of the operation personnel and the equipment operation stability.

[0083] The system uses this feedback information to correct the state representation vector of the corresponding node:

[0084] M(P i )←M(P i )+ΔM(P i )

[0085] Among them, M(P i ) represents the current state representation vector of node P i ; ΔM(P i ) represents the state correction amount calculated based on the feedback information.

[0086] Synchronize the weight function set W, and increase the weight of the corresponding item in the scoring function for the stable scheduling path, and implement scheduling punishment for the frequently failed path, to improve the robustness and structure evolution ability of the scheduling strategy.

[0087] To ensure the long-term self-evolution ability of the structure, the system performs a structure stability evaluation operation every N scheduling periods. At the evaluation node, the system calculates the average trigger success rate of each transition site T j Backtracking the running performance of the last k consecutive scheduling periods, calculate its average trigger success rate:

[0088]

[0089] Wherein, represents the actual trigger success rate of transition site T j in the tth scheduling period. j The average trigger success rate in the last L scheduling periods; L represents the number of scheduling periods for calculating the average value, i.e. the width of the sliding window; t0 represents the starting scheduling period number for backtracking statistics; j The actual trigger success rate of transition site T active in the tth scheduling period.

[0090] For transition sites with activity count (T j ) lower than the activity threshold θ active or transition success rate lower than the success rate threshold θ for c consecutive scheduling periods, determine them as inefficient paths and automatically exclude them from the transition site succ

[0091] Set T←T\{T j}, improve the overall scheduling efficiency and structural compactness of the network.

[0092] By constructing a weighted Petri net model based on the state representation vector, a dynamic mapping relationship between node state and scheduling behavior is established, realizing the precise modeling of the grouting plugging task in a multi-node and multi-constraint environment. Using dynamic weight functions and state marking mechanisms, the system can calculate the scoring priority of each scheduling path in real time, ensuring the optimal transition operation in each scheduling period, effectively improving the plugging response efficiency, resource allocation rationality, and the intelligent level of scheduling decisions. At the same time, the introduction of structure evolution and bridging mechanism enables the model to adapt to dynamic geological conditions and resource environments, significantly enhancing the robustness and long-term availability of the scheduling system in complex coal mine environments.

[0093] Unlike the method of using static state description and fixed scheduling rules in traditional technology, by introducing state vector driving mechanism and scoring function dynamic calculation logic, the state bit has a quantifiable and updateable dynamic influence expression on the scheduling transition behavior. At the same time, unlike the fixed process and unchangeable structure execution mechanism in existing methods, the scheme realizes structural bypass of scheduling bottleneck state bits through the bridging mechanism, dynamically generates auxiliary transitions and connection edges, effectively solving the problem of plugging task failure caused by resource bottleneck nodes unable to schedule, and enhancing the topological adaptability and response flexibility of the model.

[0094] S3: Extracting long-term trend component and short-term residual component based on historical state representation vectors, and fusing to generate predicted state representation vectors.

[0095] The system first extracts the state bit set S in each state bit S i The corresponding historical state representation vector is sampled to construct the historical state representation vector sequence within the time window k:

[0096] X i = [M (t-k+1) (P i ), M (t-k+2) (P i ), …, M (t) (P i )]

[0097] Wherein, represents the state representation vector of node P i in the tth scheduling period; k is the window length, indicating the number of time steps participating in prediction. X i represents the historical state representation vector sequence of node P i ; M (t) (P i ) represents the state representation vector of node P i at time t in the scheduling period; t represents the current scheduling period, and k represents the time window length, i.e. the number of historical scheduling periods participating in prediction.

[0098] In order to fully exploit the time evolution information in the historical state sequence, the original state sequence is disassembled into long-term trend component and short-term residual component.

[0099] Long-term trend component: represents the overall trend of each feature dimension in time series, which is used to describe the continuous change trend of node state.

[0100] Short-term residual component: represents the rapid fluctuation change relative to the trend component, reflecting the abnormal response or sudden disturbance of grouting behavior.

[0101] Based on the weighted moving average method, the trend of the state sequence is extracted. For each state vector dimension p, the weighted long-term trend component is defined as:

[0102]

[0103] Wherein, represents the state representation vector of node P iThe weighted long-term trend component in the pth dimension within the scheduling period τ; τ represents the scheduling period index; i represents the number of nodes, ranging from 1 to N, N being the total number of nodes. p represents the pth dimension in the state representation vector, ranging from 1 to d, d being the number of dimensions of each state vector; k represents the length of the time window; j represents the historical offset step number within the time window. t represents the time index of the current scheduling period. represents node P i The actual state value in the pth dimension at time t; w j represents the weighting coefficient corresponding to time step t-j in the sliding window.

[0104] The short-term residual component is:

[0105] wherein, represents node P i The short-term residual component in the pth dimension at time t.

[0106] The long-term trend component and the short-term residual component will respectively constitute two types of feature inputs, which are sent to the prediction model for state evolution trend reasoning.

[0107] To avoid the representation deviation caused by directly splicing the trend and the residual, based on the long-term trend component and the short-term residual component the relative amplitude in each feature dimension is used to construct an adaptive fusion coefficient, and linear weighting is used to realize dimension-by-dimension fusion, thereby improving the stability and response accuracy of the prediction vector.

[0108] First, the fusion coefficient α p in each dimension p is defined, which is used to measure the dominance of the trend in the pth dimension, and its calculation method is:

[0109]

[0110] wherein, ε is a small constant to prevent the denominator from being zero. p represents the fusion coefficient in the pth dimension.

[0111] Subsequently, the system performs fusion operation in each feature dimension to obtain each dimension value of the prediction vector:

[0112]

[0113] wherein, represents node P i The fusion state prediction value of the pth dimension at the next scheduling period t+1.

[0114] The above operation is equivalent to adjusting the degree of introduction of the residual error on the basis of the trend in each dimension, achieving a dynamic balance between the trend and the anti-dynamic.

[0115] Finally, the fused dimension values constitute the prediction representation vector of the node Pi:

[0116]

[0117] The fused prediction vector is input into the fully connected mapping layer, and the prediction state representation vector is output:

[0118]

[0119] wherein, represents the prediction representation vector of the node P i The fused prediction representation vector in the scheduling period t+1; (P i ) represents the prediction representation vector of the node P i The prediction state representation vector in the scheduling period t+1; t represents the time step of the current scheduling period; W f represents the weight matrix of the fully connected mapping layer; b f represents the bias vector of the fully connected mapping layer.

[0120] The fully connected mapping layer adopts a single-layer perceptron structure, including an input dimension d and an output dimension d ′ , wherein d is the dimension number of the fused prediction vector, and d ′ is the dimension number of the prediction state representation vector, and d' is usually set to d to maintain consistency. The mapping layer adopts a linear transformation form containing a bias term, and superimposes a nonlinear activation function ReLU (Rectified Linear Unit) for enhancing the fitting ability and the ability to express nonlinear trends of the model. The mapping function form is:

[0121]

[0122] wherein, W ∈ R represents the weight matrix of the fully connected mapping layer, and b ∈ R d′ represents the bias vector; is the fused prediction vector of the node P i in the scheduling period t+1, is the prediction state representation vector output by the mapping. The weight matrix and the bias vector are trained through a supervised learning mode driven by the historical state sequence and the feedback result, and the training target is to minimize the mean square error (MSE) between the prediction state and the actual state, so as to enhance the prediction accuracy and the model generalization ability.

[0123] During the state prediction process, the node's historical state vector sequence is decomposed into a long-term trend component and a short-term residual component. This effectively characterizes the dual factors of "slowly changing characteristics" and "sudden disturbances" in the scheduling evolution process, overcoming the fuzzy representation problem caused by existing methods that only use the original state sequence or simple sliding window prediction. The long-term trend component extracts the stable trends of various dimensional indicators through weighted sliding average, accurately reflecting the main evolution of the plugging state, such as grouting risk, resource utilization, and operational capacity. The short-term residual component, as a high-frequency response to trend deviations, characterizes abnormal events and short-term fluctuations, making the system sensitive to nonlinear disturbances, thereby effectively improving prediction accuracy and scheduling robustness under dynamic conditions.

[0124] In order to further overcome the "scale imbalance" problem that occurs in the fusion of trends and residuals, the prediction mechanism introduces an adaptive fusion coefficient design based on relative amplitude. This coefficient is dynamically generated according to the absolute value ratio of the trend component to the residual component, reflecting the balance between the trend dominance and the disturbance intensity in each dimension within the current scheduling cycle. Unlike the existing splicing method or fixed weighted fusion method, this structure can increase the sensitive weight of the residual component in the state mutation stage and strengthen the trend dominance in the state stable stage, thereby maintaining the predicted value with both anti-interference ability and responsiveness, realizing the "steady state-disturbance self-regulation" mechanism of the prediction representation.

[0125] The fused vectors are fed into a fully connected mapping layer, further enhancing the coupled modeling capabilities between multidimensional prediction states. This allows the system's output prediction state representation to not only preserve the dynamic evolution of each feature dimension but also construct a nonlinear interaction structure between dimensions. This approach effectively overcomes the structural limitations of existing prediction methods based on direct fitting and fixed thresholds, addressing the lag and insensitivity of scheduling path selection under complex operating conditions. It also significantly enhances the a priori accuracy of the subsequent scheduling scoring module and the adaptive capabilities of Petri net structure updates.

[0126] S4: taking the predicted state representation vector and the state representation vector as input, and calculating the scheduling score of each transfer position.

[0127] In each scheduling period t, the system uses all nodes P in the node set P i The current state representation vector M (t) (P i ) and the corresponding predicted state representation vector As input, calculate each transfer bit T in the transfer bit set T j The scheduling score value of is used to evaluate its triggering priority in the current scheduling cycle. First, define the trend correction function Used to fuse the predicted state representation vector with the state representation vector:

[0128]

[0129] Among them, M (t) (P i ) represents the state representation vector in the current scheduling period t; α∈[0,1] represents the state fusion coefficient, which is initially set to α=0.6.

[0130] Based on the above fusion results, the system combines the weight relationship between the state bit and the transfer bit to calculate the transfer bit T j Scheduling scoring function:

[0131]

[0132] Among them, DS (t) (T j ) represents the shift bit T j The rating value in the current scheduling cycle; S i Represents the node P i The corresponding status bit.

[0133] In each scheduling cycle, a weighted fusion mechanism of the predicted state representation vector and the current state representation vector is introduced to construct a fused state representation vector, which is used as input to participate in the scheduling score calculation of each transfer position. This can achieve a feedforward response to the future state trend of the node, enhance the scheduling strategy's comprehensive perception of the complexity of blocking tasks, resource volatility and operation timeliness, and thus improve the foresight and accuracy of scheduling decisions.

[0134] Compared with traditional scoring methods that rely on static rules or historical averages, this method integrates the current status and predicted trends to construct a trend correction function, dynamically adjusting the status input of each node in the scoring mechanism, so that the scheduling score not only reflects the current resource status, but also reflects the direction of trend changes, significantly improving the adaptability and response efficiency of scheduling behavior in dynamic environments.

[0135] This mechanism breaks through the technical bottlenecks of "inability to integrate trend changes", "delayed scheduling response" and "rigidly fixed scoring model" in existing blocking scheduling, realizes the trend adaptive adjustment of the state representation vector and the dynamic evolution of the scheduling scoring mechanism, and provides an intelligent and controllable scheduling strategy basis for high-risk, multi-path blocking tasks under complex working conditions.

[0136] S5: Select the transfer bit with the highest scheduling score and execute the blocking scheduling operation; collect the execution feedback information of the blocking task, update the state representation vector and weighted Petri net model, and realize the adaptive evolution of the scheduling logic.

[0137] According to the sorting of all scoring results, the transfer bit with the highest score is selected as the execution action in the current scheduling cycle:

[0138]

[0139] wherein T *(t) represents the target transition bit with the highest score in the tth scheduling period; argmax represents the variable that makes the function value behind it the largest; T j represents the jth candidate transition bit in the transition bit set; DS (t) (T j ) represents the scheduling score of the transition bit T j in the tth scheduling period.

[0140] To improve the scheduling response ability of low active nodes, if a state bit S k corresponding node is not scheduled for m consecutive scheduling periods and the connected transition bit scores are all lower than the set score threshold, the bridging mechanism is automatically triggered to dynamically generate a temporary bridging path:

[0141]

[0142] By introducing temporary transition bits T and bridging edges E bridge , connecting state bits S r and S k , and based on the weight function set W, the calling cost and priority of the bridging path are dynamically calculated, so as to realize the dynamic decoupling and adaptive evolution of the network topology structure. After the scheduling behavior is executed, the system synchronously collects the feedback information of the plugging task, including the deviation between the grouting amount and the target value, the grouting area pressure recovery rate, the operation personnel response time and the equipment operation stability, etc.

[0143] Based on the feedback information, the system corrects the state representation vector of the corresponding node:

[0144] M (t+1) (P k )←M (t) (P k )+ΔM(P k )

[0145] wherein S r represents a resource sufficient and state active auxiliary state bit; represents a temporary bridging transition bit dynamically introduced in the tth scheduling period; S k represents a target state bit that is not scheduled continuously or whose score is continuously low; M (t) (P k ) represents the state representation vector of node P k in the current tth scheduling period; ΔM(P k ) represents the state correction increment of node P k calculated according to the feedback information; M (t+1)(P k ) denotes the new state vector of the updated t+1 scheduling period node P k .

[0146] Meanwhile, the weight function set W is updated to increase the weight of the scheduling performance of the good path in the scoring function and reduce the scheduling priority of the repeated failure path, so as to enhance the stability and adaptive ability of the overall scheduling strategy. In order to further support the long-term optimization of the network structure, the system performs a structure evaluation every N scheduling periods, including: node activity statistics; transition trigger success rate analysis; bridge mechanism call frequency analysis.

[0147] Node activity statistics: evaluate the frequency of each state bit participating in scheduling in the last k periods, and mark it as a candidate for deletion when it is below the activity threshold;

[0148] Transition trigger success rate analysis: count the ratio of the actual execution success times of the transition bit to the total trigger times, and if it is below the set success rate threshold for consecutive periods, reduce its scheduling priority or remove it;

[0149] Bridge mechanism call frequency analysis: when a transition path is frequently triggered by the bridge operation, it indicates that the original structure is invalid and needs to be replaced or reconstructed;

[0150] Based on the scheduling execution feedback results, the state representation vector is dynamically corrected and the weighted Petri net model structure is updated, realizing the continuous optimization and evolution of the sealing scheduling logic, and enhancing the system scheduling stability and resource control intelligence in complex environments.

[0151] After completing the state representation vector correction and weighted Petri net weight update based on the scoring feedback, further record the "state bit-transition bit-feedback score" trajectory information in each round of scheduling process, and construct a historical behavior trajectory atlas. The trajectory atlas adopts a directed graph storage method based on a hash index structure, taking the state bit-transition bit pair as the directed edge index key in the graph, associating the scoring results, execution feedback and timestamp information of each round of execution, forming a path behavior record pool. Every T eval scheduling periods, the system performs effect clustering analysis on the path data in the trajectory atlas, and the clustering features include: the time series fluctuation degree of the score value, the path trigger frequency, the feedback bias mean and stability, etc. Through K-Means clustering and local outlier factor (LOF) algorithm joint evaluation, the following two types of behavior paths are extracted:

[0152] Efficient trajectory: the score value is stable for a long time, the trigger frequency is high, and the feedback effect is good.

[0153] Failed trajectory: the score fluctuates sharply, continuously deviates from the expected target, and the success rate is low.

[0154] For the transition path corresponding to the failure trajectory, the following strategy is executed: reduce the base weight in its score function, and add it to the disabled candidate pool, and after N fail consecutive failures, perform temporary removal processing. For the efficient trajectory path, increase the additional incentive weight Aw eff in the score function, and adjust its priority in the candidate transition site ranking to form a strategic scheduling guide. The above processing further breaks the static constraints of the weighted Petri net path structure, enabling the system to evolve and optimize the structure based on historical effects, thereby improving the scheduling adaptability and efficiency in complex environments.

[0155] In the plugging strategy generation process, the score-driven mechanism selects the path with the highest score as the current scheduling action by calculating the scheduling score value of each transition path, which can significantly improve the matching degree of scheduling behavior to the current state. At the same time, a bridge mechanism is designed to automatically trigger when the score is low for a long time or the state bit node is not scheduled continuously, generating a temporary bridge path connecting the "resource-rich state bit" and the "low-active state bit", solving the problem of resource scheduling deadlock in the low-active area of traditional Petri net structure, and enhancing the path flow and structural flexibility in the plugging task execution process. This strategy effectively avoids the phenomenon of "scheduling starvation" of nodes and breaks through the problem of path reconstruction in existing static topology scheduling.

[0156] After the completion of the scheduling behavior, the system collects multi-dimensional feedback information of the grouting behavior, including grouting quantity deviation, pressure recovery speed, operation response time delay, and equipment stability, etc. Key parameters, based on which the relevant state representation vectors are corrected, and the path weights in the weighted Petri net are adjusted synchronously. Compared with the traditional plugging scheduling strategy based on static rule adjustment, this structure realizes continuous parameter self-correction based on real task execution feedback, can dynamically strengthen high-quality paths and suppress inefficient paths, effectively cope with the interference of uncertain factors such as changes in geological conditions and resource fluctuations, thereby enhancing the stability and precision of the overall plugging system.

[0157] The system records the execution trajectory of "state bit → transition bit → feedback score" after each round of scheduling, constructs the behavior trajectory atlas, and performs cluster analysis on the trajectory execution effect in the structure evaluation period, identifying "repeated failure trajectories" and "efficient trajectories". By disabling or reducing the weight of the failure path and adjusting the incentive score of the high-quality path, a self-evolution mechanism based on execution history is formed. This path evolution process does not rely on pre-set fixed graph structure, but actively adjusts the topology connection mode based on scheduling effect, significantly breaking through the static limitations of traditional Petri net in path structure, giving the model the ability to continuously adapt to dynamic task distribution, and embodying strong robustness and intelligent evolution ability.

[0158] Embodiment 2, which is an embodiment of the present application, provides a coal mine separation layer grouting based on the running grouting plugging strategy generation system, comprising:

[0159] A data module acquires a grouting state parameter set through a multi-source perception device, and constructs a state representation vector based on the grouting state parameter set.

[0160] A model module constructs a weighted Petri net model based on the state representation vector.

[0161] A prediction module extracts a long-term trend component and a short-term residual component based on a historical state representation vector, and fuses to generate a predicted state representation vector.

[0162] A scoring module calculates a scheduling score of each transition position by taking the predicted state representation vector and the state representation vector as input.

[0163] An execution module selects the transition position with the highest scheduling score and executes a plugging scheduling operation.

[0164] A feedback module acquires execution feedback information of the plugging task, updates the state representation vector and the weighted Petri net model, and realizes adaptive evolution of the scheduling logic.

[0165] Embodiment 3, which is an embodiment of the present application, is different from the previous two embodiments in that:

[0166] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0167] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination of the above. For the purposes of this specification, the term "computer-readable medium" covers all tangible, physical media and media that can be accessed by a

[0168] A "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, propagation medium, or computer medium.

[0169] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical apparatus), a portable computer diskette (magnetic apparatus), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical apparatus), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then employable by a computer.

[0170] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or their combinations can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0171] Embodiment 4, which is an embodiment of the present application, provides a coal mine separation layer grouting run grouting plugging strategy generation method and system. In order to verify the beneficial effects of the present application, a simulation experiment is carried out for scientific demonstration.

[0172] A certain mine area actual working face W1207 is selected as the experimental site. The working face has a typical separation layer development area, is prone to grouting run grouting, has the characteristics of complex geological structure and multi-point plugging demand, and is suitable for verifying the scheduling ability and reliability of the present application scheme.

[0173] Before the experiment, 6 groups of multi-source perception devices were deployed in the W1207 working face, including grouting flow meter, underground ground pressure sensor, grouting pressure sensor, material inventory detection unit, electric drive grouting machine state detection module and operation personnel positioning system, real-time collection of data including grouting pressure, rock stress, grouting rate, equipment load, personnel configuration and material remaining, etc. to construct the grouting state parameter set. After normalization processing and feature dimensionality reduction analysis of the parameter set, a four-dimensional state representation vector is generated, representing the running risk, sealing complexity, resource capacity and operation availability.

[0174] A weighted Petri net structure with nodes as the core is constructed, and the dynamic evolution relationship between state bits and transition bits is encoded into the weight function set. In units of scheduling periods, the system performs weighted moving average on the historical state representation vector sequence of the node, extracts the long-term trend component, and calculates the residual between the actual state value to obtain the short-term disturbance feature. By calculating the relative amplitude of the two components, the fusion coefficient of each dimension is obtained, and then the fusion prediction state vector of the next period is obtained.

[0175] The scheduling scoring function is based on the fusion state input composed of the current state vector and the predicted vector, and the highest scoring transition bit is selected to trigger the corresponding sealing behavior. After the scheduling is completed, the system collects feedback information such as actual grouting volume, pressure recovery rate, response delay and equipment stability, and adjusts the original state representation vector and the Petri net structure weight.

[0176] The entire experiment sets 20 scheduling periods to ensure that the system strategy is fully expressed under different working conditions.

[0177] During the experiment, 9 nodes were identified, involving 26 effective transition paths. In the initial stage (period 1 to period 5), some nodes failed to seal and had a score below 0.35. The third node had a score of only 0.28 and 0.31 in the first two periods and was not triggered. Through the trend and residual prediction mechanism, the score of the third node increased to 0.46 from the sixth period, and the actual grouting volume deviation was controlled within ±3.2%, and the regional pressure recovery time was shortened to 81.5% of the original.

[0178] In continuous scheduling, the system automatically identified and disabled 4 transition paths with high failure rate, with a score below 0.4 for 3 consecutive periods. After triggering the bridging mechanism, the sixth node successfully constructed a temporary bridging path with the second node, and completed the grouting scheduling through this path in the 13th period. The operation response time delay was reduced by 26.4% compared with the initial stage, and the equipment load stability was improved by 9.8%.

[0179] In the 15th to 20th cycle system enters the stable scheduling phase, the average per cycle trigger transfer path success rate is promoted to 92.3%, and there is a significant difference with 71.6% in the initial stage.

[0180] The experimental results show that the running slurry plugging strategy generation method constructed by the application is superior to the existing scheme in dynamic scheduling efficiency, self-adaptive adjustment capability and plugging behavior response accuracy.

[0181] Firstly, based on the decomposition mechanism of long-term trend and short-term residual, the state evolution trend of the node can be more accurately described, and the scheduling error caused by the traditional judgment based on the current state can be avoided. In the experiment, multiple nodes with low initial score are gradually identified and triggered under the guidance of the prediction trend, avoiding the occurrence of resource idle phenomenon.

[0182] Secondly, the prediction state vector and real-time state are fused to construct the scheduling score input, which significantly improves the accuracy of transfer selection. After adding the prediction logic to the score model, the actual grouting deviation of the plugging action is reduced from an average of ±7.8% to ±3.1%, and the scheduling response delay is reduced by an average of 23.7%, indicating that the scoring mechanism is more forward-looking and real-time.

[0183] Thirdly, by collecting scheduling feedback information to realize the adaptive evolution of the Petri net model, the system has the ability of dynamic learning and structure reconstruction. With the aid of the bridging mechanism and the failure path elimination strategy, the model can gradually optimize the path structure and form a more adaptive scheduling network topology to the current environment, effectively avoiding redundant calculation and inefficient scheduling caused by structure rigidity.

[0184] Overall, the traditional static plugging strategy cannot respond to changes in the field disturbance in a timely manner, and an intelligent scheduling mechanism is constructed, which integrates prediction, self-regulation and evolution. The technical path has innovation, and the actual effect has engineering promotion value, which can significantly improve the intelligent level and operation stability of the coal mine separation layer grouting plugging system.

[0185] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered by the claims of the present application.

Claims

1. A coal mine separation layer grouting based on the generation method of running slurry plugging strategy, characterized by, The method comprises the following steps: Collecting a set of grouting state parameters by a multi-source perception device, and constructing a state representation vector based on the set of grouting state parameters; Based on the state representation vector, a weighted Petri net model is constructed; Based on the historical state representation vector, a long-term trend component and a short-term residual component are extracted and fused to generate a predicted state representation vector; Taking the predicted state representation vector and the state representation vector as inputs, the scheduling score of each transition position is calculated; The transition position with the highest scheduling score is selected, and a plugging scheduling operation is performed; Collecting the execution feedback information of the plugging task, updating the state representation vector and the weighted Petri net model, and realizing the adaptive evolution of the scheduling logic.

2. The coal mine separation layer grouting-based run-off plugging strategy generation method according to claim 1, characterized in that: The set of grouting state parameters includes grouting process parameters, geological structure parameters, grouting material inventory information, grouting equipment operating status, operation personnel scheduling state and historical plugging data; The set of grouting state parameters is preprocessed, and after preprocessing, the state representation vector of the node is constructed according to the set of grouting state parameters; The construction of the state representation vector of the node includes: based on the set of grouting state parameters, establishing parameter combination judgment conditions, and dividing corresponding features into different levels according to assignment rules; the discrete characteristic values of the grouting risk, the discrete characteristic values of the plugging complexity, the discrete characteristic values of the resource capacity and the discrete characteristic values of the operation availability are combined to form the state representation vector of the node; The assignment rules include: when all conditions are met at the same time, the assignment is 1; when part of the conditions are met, the assignment is 0.5; when no condition is met, the assignment is 0; The assignment of the discrete characteristic value includes: based on the combination threshold condition set by the set of grouting state parameters, the judgment is performed, and the judgment result is assigned according to the assignment rule.

3. The coal mine separation layer grouting-based run-off plugging strategy generation method of claim 2, characterized in that: The weighted Petri net model is based on the state representation vector of the node, and constructs the dynamic mapping relationship between task state and behavior transition G=(S,T,E,Token,W), wherein S represents the state bit set; T represents the transition bit set; E represents the directed edge set; Token represents the dynamic marking information corresponding to the state bit; W represents the weight function set.

4. The coal mine separation layer grouting-based run-off plugging strategy generation method according to claim 3, characterized in that: During the running of the weighted Petri net model, based on the dynamic marking information Token(S i ) of each state bit in the state bit set S, in combination with the weight function set W, the scheduling score value of the transition bit T j corresponding to the state bit is calculated , which is expressed by the following formula: wherein W ij represents the state bit S i , i and j represent indexes; F(M(P i )) is a correction function, which represents the influence intensity of the transition bit T j on the state representation vector including the run-off risk value, the plugging operation complexity, the resource capacity score and the operation availability score. According to the dispatch score results of each transfer bit, in the current dispatch period, the transfer bit T with the highest score is selected from the transfer bit set T * = argmax(DS(T j )), and a corresponding plugging behavior operation is triggered; wherein T * represents the optimal transfer selected for execution in the current dispatch period; argmax represents finding the transfer bit with the highest score from all transfer bits; DS(T j ) represents the dispatch score value of the transfer bit T j . When the status bit S i When the corresponding node is insufficient in resources or the scheduling score is lower than the set execution threshold for m consecutive scheduling cycles and no connected transfer bit is triggered, a bridge transfer bit T is automatically generated. bridge , connection status bit S m With status bit S i ; and add a bridge path S in the weighted Petri net model m →T bridge →S i ; Collect feedback information during the plugging execution process to correct the state representation vector of the corresponding node; update the weight function set W, increase the weight of the successful path, and reduce the weight of the failed path; the feedback information includes the deviation between the actual grouting volume and the predicted value, the regional pressure recovery rate, the operation response delay and the equipment stability; Every N scheduling period, the average trigger success rate of each transition position in the last L scheduling period is counted; when the activity of each transition position is lower than the activity threshold, or the average trigger success rate of each transition position is lower than the set success rate threshold in continuous c scheduling periods, the transition position is removed from the transition set.

5. The coal mine separation layer grouting-based run-off plugging strategy generation method according to claim 4, characterized in that: The extraction of the long-term trend component and the short-term residual component includes: in each scheduling period, a fixed-length state vector sequence is constructed based on the historical state representation vector of the node; the sliding weighted average processing is performed on the state vector sequence to obtain the long-term trend component of each feature dimension; the difference between the current state vector and the long-term trend component is defined as the short-term residual component; The fusion generates a prediction state representation vector, which includes: calculating a fusion weight coefficient according to the absolute value size of the long-term trend component and the short-term residual component on each feature dimension; performing linear combination on the long-term trend component and the short-term residual component of each feature dimension by using the fusion weight coefficient to obtain a fusion prediction value; splicing all fusion prediction values in the order of feature dimensions into a fusion prediction vector, and inputting the fusion prediction vector into a preset full connection layer to generate a prediction state representation vector of a next scheduling period.

6. The coal mine separation layer grouting-based run-off plugging strategy generation method according to claim 5, characterized in that: The calculation of the scheduling score of each transition position includes, in each scheduling period, respectively acquiring the current state representation vector and the corresponding prediction state representation vector of each node, and constructing a fusion state representation vector based on a weighted fusion manner; The scheduling score value of each transition position is calculated by using the weight function relationship between each state position and each transition position, wherein the score value of each transition position is calculated by the fusion state representation vector corresponding to the state position connected with the transition position and the weight function; the scheduling score values of the transition positions are sorted, and the transition position with the highest score value is selected as the blocking behavior action executed in the current scheduling period.

7. The coal mine separation layer grouting-based run-off plugging strategy generation method according to claim 6, characterized in that: The updating of the state representation vector and the weighted Petri net model includes, after the completion of the blocking operation, collecting feedback information, correcting the state representation vector of the node, and updating the weight function set in the weighted Petri net model, increasing the weight of the successful scheduling path and reducing the weight of the failed scheduling path; when a state position is not scheduled in continuous preset scheduling periods and the scheduling scores of all transition positions connected with the state position are lower than a score threshold, a bridging mechanism is triggered to generate a bridging transition position and a bridging edge, connect the state position with rich resources and the unscheduled state position, and construct a temporary bridging path.

8. A coal mine separation layer grouting-based run-off plugging strategy generation system using the method of any one of claims 1-7, characterized in that: a data module acquires a set of grouting state parameters through a multi-source sensing device, and constructs a state representation vector based on the set of grouting state parameters; a model module constructs a weighted Petri net model based on the state representation vector; a prediction module extracts a long-term trend component and a short-term residual component based on a historical state representation vector, and fuses to generate a prediction state representation vector; a scoring module calculates the scheduling score of each transition position by taking the prediction state representation vector and the state representation vector as inputs; an execution module selects the transition position with the highest scheduling score to perform a plugging scheduling operation; a feedback module collects execution feedback information of the plugging task, updates the state representation vector and the weighted Petri net model, and realizes adaptive evolution of the scheduling logic. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the coal mine separation layer grouting-based run-off plugging strategy generation method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the coal mine separation layer grouting-based run-off plugging strategy generation method of any one of claims 1-7.

Images