A multi-objective optimization method and system for a super-high cutting height fully mechanized mining equipment withdrawal process

Abstract

The application relates to the technical field of mining engineering, and discloses a multi-target optimization method and system for a super-large mining height fully-mechanized mining equipment withdrawal process. The method comprises the following steps: obtaining equipment node information, disassembly dependency relationship, spatial adjacency relationship, transportation occupation information and support continuity relationship, and constructing a disturbance attribute graph; combining support equipment parameters and surrounding rock pressure parameters to generate a disturbance safety window matrix; encoding an initial withdrawal sequence into a multi-section type chromosome; decoding and calculating multi-target fitness; performing a perception crossover and mutation operation, identifying and repairing conflict nodes, and generating a candidate withdrawal sequence; performing rigid checking on the candidate withdrawal sequence, and outputting a target withdrawal scheme; obtaining actual time consumption, transportation waiting time and surrounding rock stability monitoring data, updating disturbance penalty coefficients and transportation congestion weights, and generating a degradation rollback sequence when the threshold is exceeded. The application solves the multi-target collaborative optimization problem under multi-constraint coupling, and improves the adaptability and safety level of the withdrawal process.

Description

Technical Field

[0001] This application relates to the field of mining engineering technology, and in particular to a multi-objective optimization method and system for the retreat process of ultra-high fully mechanized mining equipment. Background Technology

[0002] Ultra-high fully mechanized mining faces typically have a mining height of over 7 meters. After mining is completed, heavy equipment such as hydraulic supports, scraper conveyors, and coal mining machines must be safely removed. Due to the extremely high mining height, the surrounding rock disturbance range is wide, the risk of roof instability is high, and there are complex dismantling dependencies, spatial adjacencies, and transportation channel occupancy relationships between equipment. Optimizing the removal process is directly related to the continuity of mine production and operational safety.

[0003] Currently, most on-site methods employ sequential withdrawal based on empirical rules, such as dismantling supports sequentially by number or separating equipment dismantling, transportation, and support into independent sequences. These methods often overlook spatial adjacency constraints between equipment, leading to large-area overhangs and excessive roof subsidence when adjacent supports are dismantled consecutively. Furthermore, empirical methods struggle to dynamically respond to changes in surrounding rock pressure; continuing to dismantle in a fixed sequence during cyclical pressure increases or support resistance decreases can easily trigger roof collapses. Another approach uses linear programming or the critical path method, simplifying the withdrawal task into a single sequential optimization. However, this cannot simultaneously address transportation congestion issues when multiple pieces of equipment share a narrow roadway, often resulting in frequent waiting times for transport vehicles. In addition, some methods treat the withdrawal plan as a static plan, failing to dynamically adjust it based on real-time on-site monitoring data, such as actual time consumption, transportation waiting time, and surrounding rock pressure. Once a disturbance occurs, the entire sequence needs to be manually rearranged, resulting in delayed response and increasing the risk of safety hazards due to personnel fatigue or misjudgment. Existing technologies also suffer from multi-objective imbalances: either excessively compressing the construction period results in insufficient intervals between adjacent removals and excessive exposure time of the surrounding rock; or overemphasizing safety leads to the addition of a large number of temporary supports, causing a sharp increase in total cost. In summary, existing methods lack the ability to uniformly model dismantling dependencies, spatial adjacency, transportation occupation, and support continuity relationships, and are also difficult to achieve an adaptive balance among multiple objectives such as construction period, cost, and safety, thus failing to meet the requirements for efficient and safe withdrawal under ultra-high mining conditions.

[0004] To address the above shortcomings, this application constructs a disturbance attribute map and a disturbance safety window matrix, combined with multi-segment chromosome coding and perceptual crossover mutation optimization. This solves the problems of existing methods, such as difficulty in uniformly modeling and decomposing dependencies, spatial adjacency, transportation occupation and support continuity, inability to adaptively balance multiple objectives such as construction period, cost and surrounding rock safety, and inability to dynamically adjust according to on-site disturbances. This improves the adaptability and safety level of the retreat process. Summary of the Invention

[0005] This application provides a multi-objective optimization method and system for the retreat process of ultra-high fully mechanized mining equipment, which solves the multi-objective collaborative optimization problem under multiple constraint coupling and improves the adaptability and safety level of the retreat process.

[0006] Firstly, this application provides a multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment, including: S1. Obtain equipment node information, disassembly dependency relationship, spatial adjacency relationship, transportation occupation information and support continuity relationship of ultra-high fully mechanized mining face, and construct disturbance attribute map; S2. Based on the disturbance attribute map, combined with the support equipment parameters and surrounding rock pressure parameters, generate a disturbance safety window matrix; S3. Based on the disturbance attribute map, the initial retreat sequence is encoded into a multi-segment chromosome containing equipment node sequence bits, transportation path bits, work group bits, and support reinforcement bits. S4. Based on the perturbation safety window matrix, decode the multi-segment chromosome and calculate the multi-objective fitness according to the initial perturbation penalty coefficient and transportation congestion weight. S5. Perform a perceptual crossover mutation operation on the multi-segment chromosome according to the perturbation safety window matrix, select according to the multi-target fitness, identify and repair conflict nodes, and generate candidate retraction sequences; S6. Perform rigid verification on the candidate withdrawal sequence for surrounding rock exposure time, adjacent withdrawal interval, transportation bottleneck and personnel exposure peak, and output the target withdrawal plan; S7. Obtain the actual on-site time, transportation waiting time, and surrounding rock stability monitoring data; update the disturbance penalty coefficient and the transportation congestion weight; when the surrounding rock stability monitoring data exceeds the preset stability range, generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight.

[0007] Secondly, this application provides a multi-objective optimization system for the retreat process of ultra-high fully mechanized mining equipment, used to implement the aforementioned multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment, including: The disturbance mapping module is used to obtain equipment node information, disassembly dependencies, spatial adjacency relationships, transportation occupancy information, and support continuity relationships of ultra-high fully mechanized mining faces, and to construct a disturbance attribute map. The matrix generation module is used to generate a disturbance safety window matrix based on the disturbance attribute map, combined with the support equipment parameters and the surrounding rock pressure parameters; The sequence encoding module is used to encode the initial retreat sequence into a multi-segment chromosome containing equipment node sequence bits, transportation path bits, work group bits, and support reinforcement bits based on the disturbance attribute map. The fitness calculation module is used to decode the multi-segment chromosome based on the perturbation safety window matrix and calculate the multi-objective fitness according to the initial perturbation penalty coefficient and transportation congestion weight. The sequence generation module is used to perform a perceptual crossover mutation operation on the multi-segment chromosome according to the perturbation safety window matrix, select according to the multi-target fitness, identify and repair conflict nodes, and generate candidate retraction sequences. The scheme output module is used to rigidly verify the surrounding rock exposure time, adjacent removal interval, transportation bottleneck and personnel exposure peak of the candidate withdrawal sequence, and output the target withdrawal scheme. The parameter update module is used to acquire the actual on-site time, transportation waiting time and surrounding rock stability monitoring data, update the disturbance penalty coefficient and the transportation congestion weight, and generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight when the surrounding rock stability monitoring data exceeds the preset stability range.

[0008] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the aforementioned multi-objective optimization method for the retreat process of an ultra-high fully mechanized mining equipment.

[0009] In summary, this application proposes a multi-objective optimization method and system for the retreat process of ultra-high fully mechanized mining equipment. It solves the problems of existing methods, such as the difficulty in uniformly modeling and decomposing dependencies, spatial adjacency, transportation occupation, and support continuity; the inability to adaptively balance multiple objectives including schedule, cost, and surrounding rock safety; and the inability to dynamically adjust according to on-site disturbances. This improves the adaptability and safety level of the retreat process. Compared with existing technologies, this application integrates multiple constraints into a structured graph model through a disturbance attribute graph, quantifies the spatiotemporal safety boundary to the encoding stage using a disturbance safety window matrix, and encodes equipment sequence, transportation path, shift allocation, and support reinforcement into multi-segment chromosomes. It proactively avoids safety risks through perceptual cross-variation and conflict repair, rigidly verifies and screens feasible solutions, and dynamically updates the penalty coefficient and congestion weight based on on-site measured data. When the surrounding rock exceeds a threshold, it automatically generates a downgrade retreat sequence. Specifically, the perturbation attribute map and perturbation safety window matrix provide a unified constraint expression and safety boundary for subsequent optimization; multi-segment chromosome encoding couples the four decision variables within the same framework, enabling perceptual cross-mutation and conflict repair to directly utilize safety window information to generate candidate sequences that satisfy the constraints, reducing the generation and repair costs of infeasible solutions; rigid verification further eliminates schemes with varying surrounding rock exposure durations, adjacent removal intervals, transportation bottlenecks, and excessive personnel peaks, ensuring that the output target withdrawal scheme is feasible in actual working conditions; field data-driven parameter updates and degradation rollback mechanisms enable offline optimization results to adapt to dynamic perturbations such as changes in surrounding rock pressure and transportation congestion, thus forming a closed loop from constraint modeling and joint optimization to dynamic response. Through the combined effect of these steps, the total withdrawal period is shortened and the overall cost is reduced while ensuring surrounding rock stability and personnel safety, and the adaptive capability of the withdrawal scheme to complex downhole environments is improved. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating a multi-objective optimization method for the retreat process of an ultra-high fully mechanized mining equipment according to this application. Figure 2 This is the iterative convergence plot of the Pareto front in this application; Figure 3 This is a graph showing the cost increase when security priority is increased in this application. Figure 4 This is a schematic diagram of the structure of a multi-objective optimization system for the retreat process of an ultra-high fully mechanized mining equipment according to this application. Detailed Implementation

[0012] This application provides a multi-objective optimization method and system for the retreat process of ultra-high fully mechanized mining equipment. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 An embodiment of a multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment in this application includes: S1. Obtain equipment node information, disassembly dependencies, spatial adjacency relationships, transportation occupancy information, and support continuity relationships of the ultra-high fully mechanized mining face, and construct a disturbance attribute map.

[0014] In one specific embodiment, performing S1 includes the following steps: Obtain equipment node information, disassembly dependencies, and spatial adjacency relationships for hydraulic supports, scraper conveyors, and coal mining machines in ultra-high fully mechanized mining faces; Obtain transportation occupancy information for intake airway, return airway and haulage roadway, as well as the support continuity relationship of support equipment; Extract key feature data from equipment node information, disassembly dependencies, spatial adjacency relationships, transportation occupancy information, and support continuity relationships; Key feature data are mapped to nodes and edges in a graph structure to generate a perturbation attribute graph.

[0015] Specifically, data is collected from ultra-high fully mechanized mining faces, which have a mining height exceeding 7 meters. After mining is completed, hydraulic supports, scraper conveyors, and coal mining machines must be removed. Equipment node information is collected through the underground positioning system and equipment ledger, including equipment number, type, weight, dismantling time, and location coordinates. The equipment number serves as an identifier; the type distinguishes between hydraulic supports, scraper conveyors, and coal mining machines; the weight determines the allocation of hoisting and transportation resources; the dismantling time is obtained from historical operation records; and the location coordinates are represented using three-dimensional coordinates in the working face coordinate system, with the horizontal axis along the working face advancement direction, the vertical axis along the working face inclination direction, and the vertical axis along the coal seam thickness direction. Dismantling dependencies describe the sequential constraints of equipment dismantling, including the preceding equipment number, the following equipment number, and the dependency type. Dependency types include hard dependencies and soft dependencies. Hard dependencies indicate that the following equipment can only be dismantled after the preceding equipment has been dismantled, while soft dependencies indicate that dismantling the following equipment is more convenient after the preceding equipment has been dismantled, but not mandatory. Spatial adjacency reflects the physical proximity of equipment, including adjacent equipment numbers, adjacent distance, and the roof influence range. The adjacent distance is calculated using coordinates. The roof influence range refers to the radius of the area above the equipment where deformation may occur after demolition; this radius is determined by both the equipment span and the mining height. The equipment span... For hydraulic supports, the support spacing is used; for scraper conveyors, the length of a single section of the central trough is used; for coal mining machines, the machine width is used. The formula for calculating the radius of influence is as follows: ,in The mining height is (meters). The benchmark height is 5 meters. Adjustment coefficients for equipment type (1.2 for hydraulic supports, 0.6 for scraper conveyors, and 0.9 for coal mining machines). Example: When the mining height is 7 meters, the influence range of the hydraulic support (center distance 1.8 meters) is approximately 3.0 meters, the scraper conveyor (single section length 1.5 meters) is approximately 1.3 meters, and the coal mining machine (body width 2.0 meters) is approximately 2.5 meters.

[0016] Transportation occupancy information is derived from the scheduling systems of the intake airway, return airway, and transportation roadways. It includes the roadway number, length, width, available time window, and maximum carrying capacity. The roadway number identifies the transportation route; length and width determine passage size restrictions; the available time window indicates the period during which the roadway can be occupied during withdrawal operations, formatted as start to end time; and the maximum carrying capacity refers to the maximum number or total weight of equipment allowed to pass through per unit time. Support continuity is provided by support monitoring sensors and support design data, including the minimum dismantling interval and temporary support requirements. The minimum dismantling interval refers to the time interval that must be maintained between the dismantling of adjacent supports to maintain roof stability; and temporary support requirements refer to the type and quantity of temporary supports that need to be added after dismantling.

[0017] Key feature data is extracted from equipment node information, dependency breakdown, spatial adjacency, transportation occupancy information, and support continuity. Key feature data refers to a subset of fields selected from the data fields for graph structure modeling. Equipment entities are mapped to equipment nodes in the graph structure, channel entities to channel nodes, and support requirement points to support nodes. Node attribute vectors are filled with the feature data of the corresponding entities and are numerical arrays describing node attributes, with values ​​arranged sequentially. Dependency breakdown is mapped to dependency edges, with the edge direction pointing from the preceding device node to the following device node. Edge weights are quantified by dependency type: hard dependency edges have a weight of 1.0, soft dependency edges have a weight of 0.5, and edge weights are numerical values ​​describing the importance of the edge. Spatial adjacency relationships are mapped to adjacent edges. Edges are formed by pairs of adjacent equipment nodes, with no distinction in connection direction. Edge attributes include adjacent distance and roof influence range. When the adjacent distance is less than or equal to 10 meters and the roof influence range overlaps, the adjacent edge is marked as a strongly associated edge; otherwise, it is marked as a weakly associated edge. The condition for overlapping roof influence ranges is that the adjacent distance is less than or equal to the sum of the radii of the roof influence ranges of the two devices. Transportation occupancy information is mapped to occupancy edges. Equipment nodes and channel nodes are connected through occupancy edges. Edge attributes include available time windows and maximum load-bearing capacity. When the equipment weight exceeds 80% of the channel's maximum load-bearing capacity, the occupancy edge is marked as an overload risk edge. Support continuity relationships are mapped to support edges. Support edges connect adjacent equipment nodes. Edge attributes include minimum dismantling interval time and temporary support requirements. The minimum dismantling interval time is in minutes, and the temporary support requirement is in the quantity of support materials. The graph structure uses an adjacency list for storage. An adjacency list is a graph storage structure that records the list of adjacent nodes for each node. The total number of nodes is the sum of the number of devices, the number of channels, and the number of support requirement points. The total number of edges is the sum of the number of dependencies, the number of adjacencies, the number of occupancy relationships, and the number of support relationships. The node feature matrix dimension is the number of nodes multiplied by the number of feature dimensions, with the number of feature dimensions set to 10. These feature dimensions correspond to the device number code, type code, weight, dismantling time, horizontal axis component of location coordinates, vertical axis component of location coordinates, vertical axis component of location coordinates, channel length, channel width, and minimum dismantling interval time, respectively. Missing features are filled with 0.

[0018] A perturbation attribute graph is a graph containing more than one type of node and more than one type of edge; such graphs are called heterogeneous graphs. The graph construction algorithm uses a rule-based manual construction method. Construction parameters include an adjacency distance threshold of 10 meters, an overload risk threshold of 80%, a strong-association edge weight of 1.2, a weak-association edge weight of 0.8, a hard-dependency edge weight of 1.0, a soft-dependency edge weight of 0.5, a support edge time weight obtained by dividing the minimum dismantling interval by the baseline time of 60 minutes, and a support edge material weight obtained by dividing the temporary support requirement quantity by the baseline quantity of 2. The node feature matrix is ​​normalized using max-min. Weight is in tons divided by the maximum weight of 200 tons; dismantling time is in hours divided by the maximum working time of 48 hours; coordinate components are in meters divided by the maximum working face length of 300 meters; passage length is in meters divided by 500 meters; passage width is in meters divided by 5 meters; and the minimum dismantling interval is in minutes divided by 1440 minutes. The normalized node feature matrix and edge attribute matrix together constitute the numerical representation of the perturbation attribute graph. This graph unifies four types of constraints—decomposition dependency, spatial adjacency, transportation occupancy, and support continuity—into a structured topological expression. This structured topological expression represents the relationships in the form of a graph structure, enabling subsequent process optimization to directly read the constraint relationships based on the graph structure, thus avoiding the shortcomings of scattered constraints and difficulty in joint modeling in empirical methods.

[0019] S2. Based on the disturbance attribute map, combined with the support equipment parameters and surrounding rock pressure parameters, generate a disturbance safety window matrix.

[0020] In one specific embodiment, performing S2 includes the following steps: Extract the support equipment's support spacing, mining height, and channel support parameters from the disturbance attribute map, and obtain the periodic pressure step distance and support resistance attenuation curve from the surrounding rock pressure parameters; The spatial constraints are determined based on the support spacing, mining height and passage support parameters. The spatial constraints include the overhead area constraint, the adjacent demolition spacing constraint and the passage support strength constraint. The time constraints are determined based on the periodic pressure step distance and the support resistance attenuation curve. The time constraints include periodic pressure avoidance constraints and support resistance attenuation constraints. By combining spatial and temporal constraints, the safety threshold of each device node in different time periods is calculated, and a disturbance safety window matrix is ​​generated.

[0021] Specifically, the support spacing, mining height, and channel support parameters of the support equipment are extracted from the disturbance attribute map. Simultaneously, the periodic pressure step distance and support resistance decay curve are obtained from the surrounding rock pressure monitoring system. Support spacing refers to the horizontal distance between the centerlines of adjacent hydraulic supports. Mining height refers to the coal seam mining height. Channel support parameters include the initial support force of a single hydraulic prop, working resistance, and support density. Periodic pressure step distance refers to the distance the working face advances before the basic top periodic fracture. The support resistance decay curve describes the functional relationship between the resistance of the support equipment and time, with the horizontal axis representing time and the vertical axis representing the support resistance value. This curve is obtained by fitting continuously collected data from hydraulic support pressure sensors. The fitting function adopts an exponential decay model, where the support resistance equals the initial resistance multiplied by the negative decay coefficient of the natural exponential function multiplied by time raised to a power. The decay coefficient is determined by regression fitting of field measured data.

[0022] Spatial constraints are determined based on the support spacing, mining height, and passage support parameters. These constraints include overhang area constraints, adjacent demolition spacing constraints, and passage support strength constraints. The overhang area constraint is calculated by multiplying the cumulative demolition length by the control distance. Let the support spacing be... The number of consecutive dismantling frames is N, and the control distance is [missing information]. Suspended ceiling The unit is square meters. Control distance. Determined by support design parameters or on-site measurements, in meters. When the suspended ceiling area... When the suspended roof area exceeds the threshold of 15 square meters, the area is marked as a high-risk suspended roof zone. The 15 square meter suspended roof area threshold is determined by the working face mine pressure observation data and corresponds to the maximum exposed area before the initial fracture of the roof. The spacing constraint between adjacent demolition sections is jointly determined by the support spacing and the allowable exposed span of the roof. The channel support strength constraint is obtained by multiplying the support density in the channel support parameters by the working resistance of the individual hydraulic prop. The support density is obtained by statistically analyzing the number of props per unit area through support design drawings or on-site measurements, with the unit being props per square meter. The working resistance of the individual hydraulic prop is obtained by using the rated value on the prop's nameplate or by on-site pressure sensor measurements, with the unit being Newtons (N). When the channel support strength is less than the peak value of the surrounding rock pressure, it is marked as an insufficiently supported channel. The peak value of the surrounding rock pressure is obtained from geological exploration data, with the unit being Newtons per square meter (N).

[0023] The time constraints are determined based on the periodic pressure step distance and the support resistance attenuation curve. The process for determining the periodic pressure avoidance constraint is as follows: Divide the working face advance speed by the periodic pressure step distance to obtain the periodic pressure cycle duration. The periodic pressure step distance is in meters, the working face advance speed is in meters per hour, and the periodic pressure cycle duration is in hours. The equipment dismantling time window is moduloed with the periodic pressure cycle duration to obtain the remainder. When the remainder is less than the periodic pressure duration of 4 hours, this time window is marked as a high-risk period for pressure, and dismantling operations are prohibited. The 4-hour periodic pressure duration is obtained from microseismic monitoring data and corresponds to the concentrated period of energy release from the basic roof fracture. The process for determining the support resistance attenuation constraint is as follows: Read the support resistance attenuation curve and find the point in the curve where the support resistance drops to 80% of the initial resistance. The time period between this point and the next pressure replenishment point is marked as the support attenuation danger period. Dismantling operations must not be started during the support attenuation danger period. The 80% initial resistance threshold is set based on the safe lower limit ratio of the rated working resistance of the hydraulic support.

[0024] Combining spatial and temporal constraints, the safety thresholds for each equipment node in different time periods are calculated, generating a disturbance safety window matrix. For each equipment node, the withdrawal operation period is divided into 1-hour time granularities, establishing a two-dimensional mapping relationship between nodes and time periods. If the suspended roof area of ​​a certain equipment node in a specific time period is less than 15 square meters, the adjacent demolition spacing is greater than the allowable exposed span of the roof, the channel support strength is greater than the peak value of the surrounding rock pressure, and the time period is not in a high-risk period of incoming pressure or in a critical period of support attenuation, then the safety threshold for that node in that time period is assigned a value of 1, indicating that demolition is allowed; if any spatial or temporal constraint is violated, the safety threshold is assigned a value of 0, indicating that demolition is prohibited. The safety thresholds of each equipment node in each time period are arranged by node number as the row index and operation time period as the column index to form a disturbance safety window matrix. For example, the hydraulic support numbered 5 satisfies all constraints from 8:00 to 9:00, so the element in the 5th row and 9th column of the matrix is ​​1; the same equipment is in a critical period of support attenuation from 9:00 to 10:00, so the element in the 5th row and 10th column of the matrix is ​​0.

[0025] S3. Based on the disturbance attribute map, the initial pullback sequence is encoded into a multi-segment chromosome containing equipment node sequence bits, transportation path bits, work group bits, and support reinforcement bits.

[0026] In one specific embodiment, performing S3 includes the following steps: Based on the dismantling dependencies and spatial adjacency relationships in the perturbation attribute graph, the order of device dismantling in the initial pullback sequence is mapped to the device node sequence position; Based on the transportation occupancy information in the disturbance attribute map, the feasible transportation routes of each device in the underground passage are mapped to transportation path locations; Obtain data on the qualification type and operational capabilities of work teams, and map the personnel performing demolition and transportation tasks to work team positions; Obtain support requirement data for the disturbed surrounding rock area, and map additional support operations for the disturbed surrounding rock area as support reinforcement positions; The equipment node sequence position, transportation path position, work team position, and support reinforcement position are sequentially spliced ​​into a multi-segment chromosome.

[0027] Specifically, the dismantling dependencies are read from the perturbation attribute graph. Hard dependency edges are weighted at 1.0, and soft dependency edges are weighted at 0.5. The direction of the dependency edges is from the preceding device node to the following device node. Hard dependency means that the following device can only be dismantled after the preceding device is dismantled, while soft dependency means that dismantling the following device is more convenient after the preceding device is dismantled, but not mandatory. The hard in-degree value of each device node is calculated, which is equal to the number of hard dependency edges pointing to that node. Soft dependency edges are not included in the hard in-degree calculation, but the soft dependency correlation of each node is recorded, which is equal to the number of soft dependency edges pointing to that node. During the traversal, device nodes with a hard in-degree value of 0 are added to the candidate queue. In the candidate queue, nodes with low soft dependency correlation are prioritized to enter the starting position of the initial pullback sequence. The node and its outgoing edges are removed, and the hard in-degree value of the remaining nodes is updated. This process is repeated until all device nodes are sorted, forming the baseline dismantling order. Spatial adjacency relationships are applied to local adjustments after the baseline demolition sequence is generated: Adjacent edges are read from the disturbance attribute map. When the adjacent distance is less than or equal to 10 meters and the influence range of the roof overlaps, the adjacent edges are marked as strongly correlated edges, with a weight of 1.2 for strongly correlated edges and a weight of 0.8 for weakly correlated edges. For adjacent equipment nodes in the baseline demolition sequence connected by strongly correlated edges, the minimum demolition interval time in the support continuity relationship is extracted, with the minimum demolition interval time in minutes. The minimum demolition interval time is divided by the baseline operation period of 60 minutes to obtain the minimum sequence interval number. If the difference in the arrangement positions of two strongly correlated devices in the baseline demolition sequence is less than the minimum sequence interval number, other equipment nodes without dependency conflicts are inserted between them to maintain a safe distance between strongly correlated devices in the sequence position. The adjusted equipment demolition sequence is mapped to the equipment node sequence position, with each gene position storing the topological sorting number corresponding to the equipment number.

[0028] Transportation occupancy information is read from the disturbance attribute map. This information includes the channel number, channel length, channel width, available time window, and maximum carrying capacity. The intake airway, return airway, and transportation roadway constitute an underground channel network, with each channel node connected to equipment nodes via an occupancy edge. When the equipment weight exceeds 80% of the channel's maximum carrying capacity, the occupancy edge is marked as an overload risk edge. For each piece of equipment, the Euclidean distance is calculated based on its location coordinates and the channel entrance coordinates. Combined with the channel's available time window, feasible transportation routes are selected. Feasible transportation routes must satisfy the condition that the equipment weight is less than the channel's maximum carrying capacity and that the available time window covers the equipment's transportation period. The selected feasible transportation routes are then encoded by integers according to the channel number to generate transportation path bits. For example, if a hydraulic support weighs 45 tons, the maximum load capacity of the intake airway is 60 tons, and the maximum load capacity of the return airway is 40 tons, then the return airway is excluded because 45 tons is greater than 40 tons. If the available time window for the intake airway is from 8:00 to 16:00 and covers the transportation period of the support, then the transportation path is coded as the channel number corresponding to the intake airway.

[0029] The system acquires data on the qualification types and operational capabilities of work teams. Qualification types include hydraulic support dismantling qualifications, scraper conveyor hoisting qualifications, coal mining machine electrical dismantling qualifications, and transportation driving qualifications. Operational capability data includes the number of team members, maximum lifting weight, time quota, and special equipment operation certificate number. Personnel performing dismantling and transportation tasks are grouped and mapped to work team positions based on the matching relationship between qualification type and equipment type. During the mapping process, it verifies whether the team's maximum lifting weight is greater than or equal to the equipment weight and whether the number of team members meets the dismantling time requirements. The time requirement equals the equipment dismantling time divided by the team's rated number of workers, which is set to 6. Teams that meet the verification conditions are coded as work team positions using integers, with each integer corresponding to a team number. For example, the maximum lifting weight of the hydraulic support dismantling team is 50 tons, the number of personnel is 8, and the equipment dismantling time is 12 hours. Then the working time requirement is 12 hours divided by 6 people, which equals 2 hours per person. This team meets the requirements of hydraulic support weight of 45 tons and working time. The work team code is the team number.

[0030] Obtain support requirement data for the disturbed surrounding rock area. This data includes the number of temporary individual hydraulic props, the number of articulated roof beams, support density, and support installation time. The disturbed surrounding rock area is determined by the local open roof area after equipment removal. This local open roof area is calculated based on the equipment span corresponding to the equipment type and the control roof dimension perpendicular to the working face: the local open roof area for hydraulic supports equals the support spacing multiplied by the control roof distance; the local open roof area for scraper conveyors equals the length of a single section of the central trough multiplied by the width of the central trough; the local open roof area for coal mining machines equals the machine width multiplied by the machine length. The number of temporary individual hydraulic props equals the local open roof area multiplied by the support density, with the support density unit being props per square meter. The number of articulated roof beams equals the number of temporary individual hydraulic props divided by 2 and rounded up. The support installation time equals the number of temporary individual hydraulic props multiplied by the installation time per prop, with the installation time per prop set to 0.5 hours. The number of temporary individual hydraulic props, the number of articulated roof beams, and the support installation time are coded as support reinforcement positions according to a fixed length. For example, the mining height is 7 meters, the hydraulic support spacing is 1.5 meters, the roof control distance is 7 meters, and the local open roof area is equal to If the area is 1 square meter and the support density is 2 supports per square meter, then the number of temporary single hydraulic props is 21, the number of hinged top beams is 11, and the support installation time is 10.5 hours. The above data is coded as support reinforcement positions.

[0031] The equipment node sequence position, transportation path position, work team position, and support reinforcement position are sequentially assembled into a multi-segment chromosome. The length of the equipment node sequence position is equal to the total number of equipment nodes, and each gene position stores the topological sorting number corresponding to an equipment number. The length of the transportation path position is equal to the total number of equipment nodes, and each gene position stores the channel number of the corresponding equipment. The length of the work team position is equal to the total number of equipment nodes, and each gene position stores the work team number of the corresponding equipment. The length of the support reinforcement position is three times the total number of equipment nodes, with each equipment occupying three gene positions, storing the number of temporary single hydraulic props, the number of articulated top beams, and the support installation time, respectively. The four gene sequences are connected end-to-end in the order of equipment node sequence position, transportation path position, work team position, and support reinforcement position to form a multi-segment chromosome. The total length of the chromosome is six times the total number of equipment nodes. Gene coding adopts a mixed coding method: the equipment node sequence position and transportation path position use integer permutation coding, the work team position uses integer coding, and the support reinforcement position uses real number coding. For example, if the working face contains 20 pieces of equipment, then the length of the equipment node sequence position is 20, the length of the transportation path position is 20, the length of the work group position is 20, the length of the support and reinforcement position is 60, and the total length of the multi-segment chromosome is 120.

[0032] S4. Based on the perturbation safety window matrix, decode the multi-segment chromosome and calculate the multi-objective fitness according to the initial perturbation penalty coefficient and transportation congestion weight.

[0033] In one specific embodiment, performing S4 includes the following steps: The multi-segment chromosome is decoded based on the perturbation safety window matrix to obtain decoding results including the dismantling sequence, dismantling time, transportation route, shift allocation and support reinforcement scheme of each device; Based on the decoding results, the total construction period and comprehensive cost are calculated. The total construction period includes equipment dismantling time, transportation time and temporary support installation time. The comprehensive cost includes dismantling cost, transportation cost, temporary support cost and personnel cost. Based on the demolition sequence and support reinforcement plan, and combined with the disturbance safety window matrix, calculate the penalty for prolonged exposure of surrounding rock, the penalty for continuous removal of adjacent supports, and the penalty for cross-occupancy of passageways. The penalties for prolonged exposure of surrounding rock and continuous removal of adjacent supports are weighted according to the initial disturbance penalty coefficient. The penalties for cross-occupancy of passages are weighted according to the initial transportation congestion weight. The multi-objective fitness is calculated by combining the total project duration and comprehensive cost.

[0034] Specifically, the multi-segment chromosome is composed of equipment node sequence positions, transportation path positions, work team positions, and support reinforcement positions spliced ​​together at fixed lengths. During decoding, it is divided into four segments according to a preset length. The length of the equipment node sequence position is equal to the total number of equipment nodes. Each gene position stores a topological sorting number. During decoding, the sequence numbers are arranged in ascending order to obtain the equipment dismantling order. The length of the transportation path position is equal to the total number of equipment nodes. Each gene position stores a channel number. During decoding, the corresponding channel number is extracted according to the equipment index in the equipment dismantling order to obtain the transportation route. The length of the work team position is equal to the total number of equipment nodes. Each gene position stores a team number. During decoding, the team number is extracted according to the equipment index to obtain the team allocation. The length of the support reinforcement position is three times the total number of equipment nodes. Every three gene positions correspond to one piece of equipment. They are decoded sequentially to obtain the number of temporary single hydraulic props, the number of articulated top beams, and the support installation time to obtain the support reinforcement scheme. The dismantling time is determined by the disturbance safety window matrix and the equipment dismantling sequence: Each piece of equipment is assigned a dismantling time slot according to the dismantling sequence. The safety threshold corresponding to the equipment node in the disturbance safety window matrix is ​​queried. If the safety threshold is 1, the time slot is assigned; if the safety threshold is 0, the time slot is postponed to the first time slot with a safety threshold of 1. The dismantling end time equals the dismantling start time plus the equipment dismantling time. For example, the working face contains 20 pieces of equipment. After decoding the equipment node sequence bit, the dismantling sequence from hydraulic support 1 to hydraulic support 10 to the coal mining machine is obtained. The transportation path bit is decoded as the channel numbers such as the intake airway and return airway. The work group bit is decoded as the numbers such as group 3 and group 25. The support reinforcement bit is decoded as 21 temporary single hydraulic props, 11 articulated top beams, and 10.5 hours of support installation time corresponding to hydraulic support 1.

[0035] The total construction period is determined by the equipment dismantling time. Transportation time and temporary support installation time Summation yields, Equipment dismantling time Read directly from the dismantling time field in the equipment node information, in hours. Transportation time. It equals the channel length divided by the transport speed, which is 3600 meters per hour. The channel length is extracted from the transport occupancy information and is in meters. The unit is hours. Temporary support installation time. Read from the decoding results of the support reinforcement position, in hours. The overall cost consists of the dismantling cost. Transportation costs Temporary support costs and personnel costs Summation yields, Disassembly costs This equals the equipment dismantling time multiplied by the dismantling labor hourly rate, which is 200 yuan per hour. Transportation costs. The cost is calculated by multiplying the channel length by the freight rate per ton-kilometer, then multiplying by the equipment weight. The freight rate per ton-kilometer is 5 yuan. Equipment weight is in tons, and channel length is in kilometers. (Temporary support cost) The cost is calculated as follows: (Number of temporary single hydraulic supports multiplied by the unit price of the supports) + (Number of hinged top beams multiplied by the unit price of the top beams). The unit price of each support is 150 yuan, and the unit price of each top beam is 80 yuan. (Personnel costs are also included.) The cost is obtained by summing the dismantling team cost and the transportation team cost. The dismantling team cost equals the number of personnel in the dismantling team multiplied by the equipment dismantling time. Multiply by the dismantling labor unit price. The number of dismantling team members is read from the work team's decoding results. The dismantling labor unit price is 50 yuan per person per hour. The transportation team cost equals the number of transportation team members multiplied by the transportation time. Multiply by the unit price of transportation labor, the number of personnel in the transportation team is read from the team file corresponding to the transportation task, and the unit price of transportation labor is 50 yuan per person per hour.

[0036] The penalty for prolonged exposure of surrounding rock is calculated based on the time interval between the removal of adjacent equipment in the removal sequence. (Surrounding rock exposure duration) It equals the start time of the subsequent equipment removal minus the end time of the previous equipment removal, in hours. Safety threshold. Read from the minimum removal interval time in the support continuity relationship, in hours. Greater than Penalty for prolonged exposure of surrounding rock Equals the difference between the surrounding rock exposure time and the safety threshold, multiplied by the surrounding rock exposure penalty coefficient. , The fee is 600 yuan per hour. The penalty for consecutive removal of adjacent supports is generated by checking the strong association edge constraint: traverse the pairs of adjacent equipment nodes connected by strong association edges in the removal sequence, calculate their position difference in the removal sequence. The position difference is equal to the absolute value of the difference between the two equipment sequence numbers. If the position difference is less than the minimum sequence interval, the penalty for consecutive removal of adjacent supports is triggered. , Equals the number of violations multiplied by the consecutive removal penalty coefficient. , The penalty is 5000 yuan per instance. The minimum sequence interval is obtained by dividing the minimum dismantling interval time by 60 minutes. The channel cross-occupancy penalty is calculated jointly by the transport path bit decoding result and the disturbance safety window matrix: The number of devices in the same channel at the same time period is counted; if this number exceeds the channel's maximum carrying capacity, the number of overloaded devices is determined. Equals the actual number of devices minus the maximum load capacity, and the overload duration. This is equivalent to a time period of 1 hour, with a penalty for cross-occupancy of the passageway. Equals the number of overloaded devices multiplied by the duration of overload multiplied by the congestion penalty coefficient. , It costs 300 yuan per hour per unit.

[0037] The multi-objective fitness is a vector form consisting of schedule, cost, and safety objectives. Project timeline Equal to the total construction period Cost Target equals comprehensive cost Add perturbation penalty coefficient Multiply by the sum of the penalty for prolonged exposure of surrounding rock and the penalty for continuous removal of adjacent supports, and add the transportation congestion weight. Multiply by the channel cross-occupancy penalty, i.e. Perturbation penalty coefficient The weight of transportation congestion is 0.8. The target value is 0.6. This equals the sum of the number of times the surrounding rock exposure timed out, the number of times adjacent supports were removed in violation of regulations, and the number of times the passage was overloaded. These three objectives are simultaneously expressed as vectors, and their individual performance is compared in the subsequent non-dominated ranking.

[0038] S5. Perform perceptual crossover mutation operation on multi-segment chromosomes according to the perturbation safety window matrix, select according to multi-target fitness, identify and repair conflict nodes, and generate candidate reversal sequences.

[0039] In one specific embodiment, performing S5 includes the following steps: Based on the perturbation safety window matrix, a perceptual crossover mutation operation is performed on the multi-segment chromosome to obtain the mutated chromosome; Decode the mutated chromosome, recalculate the multi-objective fitness, and obtain the multi-objective fitness after mutation; Compare the mutated multi-objective fitness with the original multi-objective fitness, and select non-dominant individuals; Identify conflict nodes in non-dominated individuals that violate the constraints of the disturbance safety window matrix and disturbance attribute graph, including equipment in non-safe dismantling periods, equipment whose adjacent dismantling intervals do not meet the minimum interval constraint, and equipment that causes channel overload; For conflict nodes, time adjustment, path adjustment, shift adjustment and support adjustment are performed sequentially to generate candidate pullback sequences that satisfy the constraints of the disturbance safety window matrix and disturbance attribute map.

[0040] Specifically, multiple chromosome segments are extracted from the current population, with a population size of 100. Each chromosome is composed of equipment node sequence positions, transportation path positions, work group positions, and support reinforcement positions spliced ​​together at a fixed length. The sensing crossover and mutation operation includes a crossover phase, a mutation phase, and a sensing verification phase. During the crossover phase, the equipment node sequence position uses partial mapping crossover. Two cut points are randomly generated on the chromosome, and the segments between the cut points of the two parent individuals are exchanged. A mapping table is established to record the correspondence of genes within the exchanged segments. Duplicate genes generated after the exchange are replaced according to the mapping table until all genes are free of duplication. The crossover probability is set to 0.85. The transportation path position uses single-point crossover. One crossover point is randomly selected, and all genes to the right of that point are exchanged. The crossover probability is set to 0.85. The work group position uses uniform crossover. A binary mask of the same length as the chromosome is generated. When the mask bit is 1, the corresponding gene is exchanged. When the mask bit is 0, the original gene is retained. The crossover probability is set to 0.85. The support reinforcement position uses arithmetic crossover. The support reinforcement genes of the two parent individuals are linearly combined with a weight of 0.5. The combination result is truncated to the non-negative interval. The crossover probability is set to 0.85. During the mutation phase, the sequence position of the equipment node undergoes an exchange mutation with a mutation probability of 0.15, randomly selecting two gene positions to exchange values; the transportation path position undergoes a reset mutation with a mutation probability of 0.15, randomly selecting gene positions and replacing them with feasible channel numbers; the work team position undergoes a reset mutation with a mutation probability of 0.15, randomly selecting gene positions and replacing them with standby team numbers; the support reinforcement position undergoes a Gaussian mutation with a mutation probability of 0.15, superimposing Gaussian random numbers with a mean of 0 and a standard deviation of 0.1 onto the gene values, and truncating the superimposed values ​​to the non-negative interval. The perception verification phase is embedded after the crossover and mutation: for the sequence position of the equipment node, the equipment number corresponding to the mutation position is extracted, and the safety threshold of the equipment in the corresponding time period is queried in the disturbance safety window matrix. If the safety threshold is 0, the mutation is rejected and the original parent gene is restored; for the transportation path position, the relationship between the maximum carrying capacity corresponding to the new channel number and the equipment weight is verified. If the equipment weight is greater than the maximum carrying capacity, the mutation is rejected and the original gene is restored. For example, the safety threshold of hydraulic support 5 is 0 during the period from 9:00 to 10:00. If the dismantling order is adjusted to this period during cross-operation, the perception verification triggers a rejection command, and the gene position is restored to the original topological sorting number.

[0041] The mutated chromosome is decoded and segmented according to the preset lengths of equipment node sequence, transportation path, work team, and support reinforcement positions to obtain the dismantling sequence, transportation route, work team allocation, and support reinforcement scheme. Based on the disturbance safety window matrix, a dismantling time period is allocated to each piece of equipment. The total project duration and comprehensive cost are calculated, as are penalties for excessive exposure of surrounding rock, continuous removal of adjacent supports, and cross-occupancy of passageways. The mutated multi-objective fitness is calculated, including three dimensions: project duration, cost, and safety. This calculation process is consistent with S4 above and will not be repeated.

[0042] The parent population is merged with the mutated offspring population, resulting in a population size of 200. NSGA-II fast non-dominated sorting is performed: for each individual in the merged population, the dominance relationships of their multi-objective fitness vectors are compared. If individual A is not inferior to individual B in all objectives and is superior to individual B in at least one objective, then individual A dominates individual B. The number of times each individual is dominated is counted. Individuals dominated by 0 individuals are placed in the first non-dominated layer. After removing individuals from the first layer, the remaining individuals dominated by 0 individuals are placed in the second non-dominated layer. This process is repeated until all individuals are stratified. Crowding distance is calculated for each layer: individuals are sorted in ascending order of objective function values. The crowding distance for boundary individuals is set to infinity. The crowding distance for intermediate individuals is equal to the difference in function values ​​of adjacent individuals on that objective divided by the difference between the maximum and minimum values ​​of that objective. The total crowding distance is obtained by summing the crowding distances for each objective. Individuals are selected in ascending order of non-dominated layer number and in descending order of crowding distance within the same layer until the population size of 100 is filled, forming a new generation population.

[0043] For each individual in the new generation population, decode the data to identify whether it violates the constraints of the perturbation safety window matrix and perturbation attribute graph. Non-safe dismantling time period conflict identification: Traverse each device in the dismantling sequence, query the safety threshold of that device in the allocated time period in the perturbation safety window matrix; if the safety threshold is 0, mark it as a time conflict node. Adjacent dismantling interval conflict identification: Traverse pairs of device nodes connected by strong correlation edges, calculate their position difference in the dismantling sequence. The position difference is equal to the absolute value of the difference between the sequence numbers of the two devices. If the position difference is less than the minimum sequence interval number, mark it as an adjacent conflict node. The minimum sequence interval number is obtained by dividing the minimum dismantling interval time by 60 minutes. Channel overload conflict identification: Count the number of devices in each channel by time period. If the number of devices exceeds the maximum carrying capacity of the channel, all devices in that channel during that time period are marked as transportation conflict nodes. For example, hydraulic support number 8 is assigned to the time period from 9:00 to 10:00, during which the safety threshold is 0, and is marked as a time conflict node; hydraulic support 3 and hydraulic support 4 are connected by a strongly associated edge and have a position difference of 1 in the dismantling sequence, with a minimum sequence interval of 3, and are marked as adjacent conflict nodes; there are 3 pieces of equipment in the air intake roadway during the time period from 8:00 to 9:00, with a maximum load capacity of 2 pieces, and all 3 pieces of equipment are marked as transportation conflict nodes.

[0044] For time-conflicting nodes, time adjustment is performed by postponing the start time of equipment dismantling by one hour. The disturbance safety window matrix is ​​re-queried; if the safety threshold is still 0, the postponement continues until the safety threshold is 1. The dismantling end time equals the adjusted start time plus the equipment dismantling time. For adjacent conflicting nodes, shift or time adjustment is performed. If two conflicting devices are handled by the same shift, the device is assigned to a spare shift number, which is extracted from the auxiliary memory array. If time conflicts and adjacent conflicts overlap, time adjustment takes precedence. For transportation conflicting nodes, path adjustment is performed by replacing the channel number in the transportation path of the device with a spare channel number with a load rate below 80%. Spare channels are obtained by filtering the occupied edges connected to the device node in the disturbance attribute graph. The filtering criteria are that the device weight is less than the maximum carrying capacity of the spare channel and the available time window covers the transportation period. For nodes with insufficient support, support adjustments are performed, increasing the number of temporary individual hydraulic props in the support reinforcement positions by 20%, and simultaneously increasing the number of articulated top beams by 20%. The support installation time is recalculated based on the increased number of props, with the installation time for a single prop set to 0.5 hours. The repaired individual props are then re-decoded and verified. If conflicts still exist, the above adjustments are repeated until all constraints are satisfied, generating a candidate withdrawal sequence.

[0045] S6. Perform rigid verification on the candidate withdrawal sequence for surrounding rock exposure time, adjacent withdrawal interval, transportation bottleneck and peak personnel exposure, and output the target withdrawal plan.

[0046] In one specific embodiment, performing S6 includes the following steps: Acquire channel occupancy data, transportation equipment load data, team qualification data, and support and reinforcement resource data, and dynamically route and allocate dismantling tasks, transportation tasks, and reinforcement tasks in the candidate withdrawal sequence. Rigid verification is performed on the candidate pullback sequence after dynamic routing allocation, including verifying whether the exposure time of the surrounding rock exceeds the safety threshold, verifying whether the removal interval of adjacent equipment is less than the minimum removal interval, verifying whether the number of equipment in the channel exceeds the maximum carrying capacity, and verifying whether the number of workers on the working face at the same time exceeds the safety limit. For candidate pullback sequences that pass the rigid verification, the optimal non-dominated solution is identified based on the corresponding multi-objective fitness and output as the target pullback scheme.

[0047] Specifically, dismantling tasks, transportation tasks, and reinforcement tasks are extracted from the candidate withdrawal sequence. Dismantling tasks include equipment number, dismantling start time, dismantling end time, and work team number. Transportation tasks include equipment number, transportation route number, transportation start time, and transportation end time. Reinforcement tasks include equipment number, support installation start time, support installation end time, number of temporary single hydraulic props, and number of articulated top beams.

[0048] Acquire channel occupancy data, which includes channel number, current occupancy period, and remaining available capacity. Remaining available capacity equals the channel's maximum carrying capacity minus the number of currently assigned tasks. Acquire transport equipment load data, which includes equipment number, rated lifting weight, current load status, and location coordinates. Rated lifting weight refers to the maximum weight the equipment is allowed to lift. A current load status of 0 indicates no load, and 1 indicates full load. Acquire work team qualification data, which includes work team number, current work status, location coordinates, and qualification type code. A current work status of 0 indicates idle, and 1 indicates busy. Acquire support and reinforcement resource data, which includes material inventory point number, temporary single hydraulic prop inventory quantity, articulated top beam inventory quantity, and inventory point location coordinates. The inventory point location coordinates are the three-dimensional coordinates of the material inventory point in the working face coordinate system.

[0049] During dynamic routing allocation, for dismantling tasks, the Euclidean distance is calculated based on the team's location coordinates and the equipment's location coordinates. The dismantling task is assigned to the team with the smallest Euclidean distance and a current operation status of 0. After allocation, the current operation status of that team is updated from 0 to 1. For transportation tasks, the Euclidean distance is calculated based on the equipment's location coordinates and the equipment's location coordinates. The transportation task is assigned to the equipment with the smallest Euclidean distance and a current load status of 0. The rated lifting weight is verified to be greater than or equal to the equipment weight. After verification, the current load status of that equipment is updated from 0 to 1. For reinforcement tasks, the Euclidean distance is calculated based on the coordinates of the inventory point and the location to be reinforced. The coordinates of the location to be reinforced are the three-dimensional coordinates of the center of the surrounding rock disturbance area after equipment dismantling. The inventory point with the smallest Euclidean distance is selected as the material supply source. The inventory of temporary single hydraulic props and the inventory of articulated top beams are verified to be greater than or equal to the task requirement. If both are satisfied, the allocation is confirmed. After allocation, the inventory of temporary single hydraulic props is reduced by the task requirement, and the inventory of articulated top beams is reduced by the task requirement. For example, the dismantling task of hydraulic support 5 is assigned to team 3. The Euclidean distance between the position coordinates of team 3 and the position coordinates of hydraulic support 5 is 15 meters, and the current working status of team 3 is updated from 0 to 1. The transportation task of hydraulic support 5 is assigned to equipment 2. The Euclidean distance between the position coordinates of equipment 2 and the position coordinates of hydraulic support 5 is 20 meters. The rated lifting weight of equipment 2 is 50 tons, and the equipment weight is 45 tons. The current load status is updated from 0 to 1. The reinforcement task of hydraulic support 5 is to extract 21 temporary single hydraulic props and 11 articulated top beams from storage point 1. The Euclidean distance between the position coordinates of storage point 1 and the position coordinates to be reinforced is 8 meters. The inventory of temporary single hydraulic props at storage point 1 is updated from 100 to 79, and the inventory of articulated top beams is updated from 50 to 39.

[0050] Rigid checks are performed on candidate pullback sequences after dynamic routing allocation, including: surrounding rock exposure duration check, adjacent equipment dismantling interval check, number of equipment in the passageway check, and number of workers on the working face at the same time. The surrounding rock exposure duration check iterates through adjacent equipment node pairs in the dismantling sequence, calculating the surrounding rock exposure duration, which is equal to the start time of the subsequent equipment dismantling minus the end time of the previous equipment dismantling, in hours. The minimum dismantling interval time is extracted as a safety threshold, in hours. If the surrounding rock exposure duration is greater than the safety threshold, the check fails, and the adjacent equipment node pair is marked as having an exposure violation. The adjacent equipment dismantling interval check iterates through equipment node pairs connected by strongly correlated edges, calculating their position difference in the dismantling sequence, which is equal to the absolute value of the difference between the two equipment sequence numbers. The minimum dismantling interval time is divided by the baseline working period of 60 minutes to obtain the minimum sequence interval number. If the position difference is less than the minimum sequence interval number, the check fails, and the equipment node pair is marked as having an interval violation. The equipment quantity verification for each transport channel is performed by counting the number of devices in each transport channel over a time period. The number of devices equals the sum of the number of transport tasks and the number of dismantling tasks within that time period. The maximum carrying capacity of the channel is extracted from the transport occupancy information, in units of units per hour. If the number of devices exceeds the maximum carrying capacity of the channel, the verification fails, and the channel is marked as an overloaded channel. The number of workers on the workface at the same time is verified by counting the sum of the number of workers in each work group corresponding to all dismantling tasks, transport tasks, and reinforcement tasks within the workface over a time period to obtain the total number of workers. The safety upper limit Lim is calculated based on the effective ventilation cross-sectional area of ​​the retreat channel, the allowable wind speed, and the standard air supply per person. The calculation formula is as follows: ,in Allowable wind speed for the retreat channel, For effective ventilation cross-sectional area, 4 represents the minimum air supply per person per minute. This is the air volume reserve coefficient. This indicates rounding down to the nearest integer. For example, when the cross-sectional area of ​​the pullback channel... Permissible wind speed Reserve coefficient The safe limit is 30 people. If the total number of workers exceeds the safe limit, the verification will fail and the period will be marked as a period of excessive personnel.

[0051] For candidate pullback sequences that pass the rigid validation, the optimal non-dominated solution is identified based on the multi-objective fitness vector. The multi-objective fitness vector includes three dimensions: schedule objective, cost objective, and safety objective. The schedule objective is the total schedule, the cost objective is the sum of comprehensive cost and weighted penalty, and the safety objective is the sum of the number of violations. All candidate pullback sequences that pass the validation are traversed, and the dominance relationships of their multi-objective fitness vectors are compared: if sequence A is not inferior to sequence B in all objectives and is superior to sequence B in at least one objective, then sequence A dominates sequence B. The number of times each sequence is dominated is counted, and sequences dominated by 0 sequences constitute the non-dominated solution set. In the non-dominated solution set, the sequence with the smallest schedule objective is selected as the target pullback scheme output. If multiple sequences have the same schedule objective, the sequence with the smallest cost objective is selected; if the cost objectives are still the same, the sequence with the smallest safety objective is selected. For example, candidate sequence A has a project duration target of 22.71 hours, a cost target of 15,547.5 yuan, and a safety target of 2; candidate sequence B has a project duration target of 23.5 hours, a cost target of 14,800 yuan, and a safety target of 1; and candidate sequence C has a project duration target of 22.71 hours, a cost target of 15,000 yuan, and a safety target of 2. Sequence C dominates sequence A, while sequences B and C do not dominate each other. The non-dominated solution set includes sequences B and C. Sequence C, which has the smallest project duration target, is selected as the target withdrawal scheme output.

[0052] S7. Obtain actual on-site time, transportation waiting time, and surrounding rock stability monitoring data; update the disturbance penalty coefficient and transportation congestion weight; and generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight when the surrounding rock stability monitoring data exceeds the preset stability range.

[0053] In one specific embodiment, performing S7 includes the following steps: Acquire actual on-site time, transportation waiting time, and surrounding rock stability monitoring data. The surrounding rock stability monitoring data includes support resistance, roof subsidence, and internal squeezing of the two sides. The disturbance penalty coefficient is updated based on the actual on-site time and surrounding rock stability monitoring data, and the transportation congestion weight is updated based on the transportation waiting time. If the surrounding rock stability monitoring data exceeds the preset stability range, then based on the updated disturbance penalty coefficient and transportation congestion weight, the multi-objective fitness is recalculated according to the logic from S4 to S6, and the perception cross-variation and rigid verification are performed to generate a downgraded rollback sequence that prioritizes the stability of the surrounding rock.

[0054] Specifically, actual operating time, transportation waiting time, and surrounding rock stability monitoring data are collected from the underground work site. Actual operating time is recorded via the work team's handheld terminal or underground positioning system, including the actual start and end times of dismantling for each piece of equipment. Actual operating time equals the actual end time minus the actual start time, measured in hours. Transportation waiting time is recorded through the transportation dispatch system, including vehicle arrival and departure times. Transportation waiting time equals the vehicle departure time minus the vehicle arrival time, measured in hours. Surrounding rock stability monitoring data is collected using hydraulic support pressure sensors, roof delamination meters, and roadway convergence meters, including support resistance, roof subsidence, and internal squeezing of the sidewalls. Support resistance is collected in real-time by hydraulic support column pressure sensors, with a collection frequency of once every 10 minutes, measured in Pascals (Pa). Roof subsidence is measured using a roof delamination meter, with the measurement base fixed within the roof strata and the reading head installed on the roadway roof surface, measured in millimeters. The internal displacement of the two sides of the roadway is measured by a roadway convergence gauge, with the measuring points located in the middle of the two sides of the roadway. The unit is millimeters.

[0055] The disturbance penalty coefficient is updated based on the actual on-site time consumption and surrounding rock stability monitoring data. The actual time consumption deviation ratio is calculated, which equals the difference between the actual on-site time consumption and the planned dismantling time, divided by the planned dismantling time. The planned dismantling time is read from the equipment node information and is in hours. The surrounding rock stability coefficient is calculated, which equals the measured support resistance divided by the rated support resistance multiplied by 0.5, plus the roof settlement threshold divided by the measured roof settlement multiplied by 0.3, plus the internal squeezing threshold divided by the measured internal squeezing of the two sides multiplied by 0.2. If the measured roof settlement or the measured internal squeezing of the two sides is 0, it is replaced with a minimum value of 0.001 mm. The rated support resistance is the rated working resistance of the hydraulic support, in Pascals. The roof settlement threshold is set to 50 mm, and the internal squeezing threshold is set to 30 mm. The old disturbance penalty coefficient is recorded as... The new disturbance penalty coefficient is denoted as , equal The sum of the values ​​multiplied by 1 and the actual time deviation ratio is then divided by the surrounding rock stability coefficient. When the surrounding rock stability deteriorates, the measured support resistance decreases, leading to a decrease in the surrounding rock stability coefficient; the measured roof subsidence increases, leading to a decrease in the second term; and the measured internal squeezing of the two sides increases, leading to a decrease in the third term. The overall surrounding rock stability coefficient is less than 1. After dividing by this coefficient, the penalty coefficient for new disturbances is greater than that for old disturbances. The weight of the penalty for prolonged exposure of surrounding rock and the penalty for continuous removal of adjacent supports in the cost target increases.

[0056] The transportation congestion weight is updated based on transportation wait time. The average transportation wait time is calculated as the sum of the waiting times of all equipment divided by the total number of equipment nodes, in hours. The baseline wait time is set to 0.5 hours. The old transportation congestion weight is recorded as follows. The new transportation congestion weight is recorded as , equal The sum of the values ​​after multiplying the average waiting time by 1 and then dividing the difference between the average waiting time and the baseline waiting time by the baseline waiting time is calculated. When the average waiting time is greater than the baseline waiting time (the value in parentheses is greater than 1), the new congestion weight is greater than the old congestion weight. When the average waiting time is equal to the baseline waiting time, the new congestion weight is equal to the old congestion weight. When the average waiting time is less than the baseline waiting time, the value in parentheses is less than 1, and the new congestion weight is less than the old congestion weight.

[0057] The system determines whether the surrounding rock stability monitoring data exceeds the preset stability range. The preset stability range includes the lower limit of the support resistance, the upper limit of the roof settlement, and the upper limit of the squeezing within the sidewalls. The lower limit of the support resistance is set at 70% of the rated support resistance. If the measured support resistance is less than 70% of the rated support resistance, the surrounding rock stability monitoring data is considered to exceed the preset stability range. The upper limit of the roof settlement is set at 50 mm. If the measured roof settlement is greater than 50 mm, the preset stability range is exceeded. The upper limit of the squeezing within the sidewalls is set at 30 mm. If the measured squeezing within the sidewalls is greater than 30 mm, the preset stability range is exceeded. Meeting any one of these three conditions triggers the downgrade / revert mechanism.

[0058] When the surrounding rock stability monitoring data exceeds the preset stability range, a downgrade retreat sequence is generated based on the updated disturbance penalty coefficient and transportation congestion weight. The disturbance safety window matrix is ​​updated: the overhang area threshold in S2 is lowered from 15 square meters to 12 square meters; the starting condition for the critical period of support attenuation in the support resistance attenuation constraint is increased from 80% to 85% of the initial support resistance; and the duration of the high-risk period of pressure in the periodic pressure avoidance constraint is extended from 4 hours to 5 hours. Based on the updated constraints, the safety thresholds for each equipment node in each time period are recalculated, and the safety thresholds for more time periods are marked as 0. The updated disturbance penalty coefficient and transportation congestion weight are substituted into the multi-objective fitness calculation to recalculate the cost objective of the candidate retreat sequence. The cost objective equals the comprehensive cost plus the sum of the new disturbance penalty coefficient multiplied by the surrounding rock exposure timeout penalty and the adjacent support continuous removal penalty, plus the new transportation congestion weight multiplied by the channel cross-occupancy penalty. A perceptual crossover and mutation operation is performed, with a population size of 100. The crossover probabilities for equipment node sequence positions are 0.85, transportation path positions, work team positions, and support reinforcement positions, respectively. The mutation probabilities for equipment node sequence positions are 0.15, transportation path positions, work team positions, and support reinforcement positions. The perceptual verification phase is based on the updated disturbance safety window matrix. A rigid verification is performed, maintaining the minimum demolition interval time as the safety threshold for surrounding rock exposure time, maintaining the minimum sequence interval number of adjacent equipment demolition intervals divided by 60 minutes, maintaining the original set value for the maximum carrying capacity of the passage, and maintaining a personnel safety limit of 30 people. After passing the rigid verification, the sequence with the smallest safety objective is selected from the non-dominated solution set as the degraded rollback sequence output. This sequence prioritizes the stability of the surrounding rock, while the schedule and cost objectives are secondary optimization directions. When downgrading and reverting, the safety target is adjusted to the sum of the number of times the surrounding rock exposure exceeds the time limit, the number of times the adjacent supports are removed in violation of regulations, the number of times the passage is overloaded, and the number of times the surrounding rock stability exceeds the limit. The number of times the surrounding rock stability exceeds the limit is equal to the sum of the number of times the measured support resistance is less than 70% of the rated support resistance, the number of times the measured roof subsidence is greater than 50 mm, and the number of times the measured internal squeezing of the two sides is greater than 30 mm.

[0059] Please see Figure 2 , Figure 2This is an iterative convergence plot of the Pareto front, with the number of algorithm iterations on the horizontal axis and the comprehensive cost of equipment withdrawal on the vertical axis. The plot shows the distribution characteristics of the Pareto front, the convergence speed, and the distribution range of non-dominated solutions during the iteration process. In the early stages of the iteration, the overall comprehensive cost is relatively high, and the solution distribution is also relatively scattered. As the number of iterations continues to increase, the Pareto front continuously converges towards regions with lower comprehensive costs. This plot illustrates that the combined strategy of multi-segment chromosome encoding and perceptual crossover mutation possesses excellent iterative optimization capabilities. The algorithm can continuously optimize the solution under multiple operational constraints such as project schedule and surrounding rock safety, continuously reduce the comprehensive cost, quickly select a non-dominated solution set that balances economy and on-site operational constraints, and simultaneously preserve the diversity of solutions throughout the process.

[0060] Please see Figure 3 , Figure 3 The graph shows the cost increase curve when safety priority is increased. The horizontal axis represents the safety priority coefficient, and the vertical axis represents the comprehensive cost of equipment withdrawal. It illustrates the changing trend, growth rate, and slope difference of the comprehensive cost for the two methods as safety control levels are progressively increased. The traditional method shows a significantly steeper curve slope; a slight increase in safety priority results in a sharp increase in comprehensive cost. In contrast, the curve for the method described in this application shows a relatively gentle trend. Even with continuously increasing surrounding rock safety control standards, the increase in comprehensive cost remains within a reasonable range, and the difference between the two curves continues to widen as safety priority increases. This indicates that the method described in this application, relying on disturbance attribute maps and disturbance safety window matrices to achieve multi-constraint quantitative control, combined with multi-objective optimization algorithms to coordinate and allocate various operational resources, can control cost increases while strengthening safety protection.

[0061] Please see Figure 4 The following describes a multi-objective optimization system for the withdrawal process of ultra-high fully mechanized mining equipment in an embodiment of this application. This multi-objective optimization system for the withdrawal process of ultra-high fully mechanized mining equipment includes: The disturbance mapping module is used to obtain equipment node information, disassembly dependencies, spatial adjacency relationships, transportation occupancy information, and support continuity relationships of ultra-high fully mechanized mining faces, and to construct a disturbance attribute map. The matrix generation module is used to generate a disturbance safety window matrix based on the disturbance attribute map, combined with the support equipment parameters and surrounding rock pressure parameters. The sequence encoding module is used to encode the initial pullback sequence into a multi-segment chromosome containing equipment node sequence bits, transportation path bits, work group bits, and support reinforcement bits based on the perturbation attribute map. The fitness calculation module is used to decode multi-segment chromosomes based on the perturbation safety window matrix and calculate multi-objective fitness according to the initial perturbation penalty coefficient and transportation congestion weight. The sequence generation module is used to perform perceptual crossover mutation operation on multi-segment chromosomes according to the perturbation safety window matrix, select based on multi-target fitness, identify and repair conflict nodes, and generate candidate retraction sequences. The scheme output module is used to rigidly verify the surrounding rock exposure time, adjacent removal interval, transportation bottleneck and personnel exposure peak of the candidate withdrawal sequence, and output the target withdrawal scheme. The parameter update module is used to acquire the actual on-site time, transportation waiting time and surrounding rock stability monitoring data, update the disturbance penalty coefficient and transportation congestion weight, and generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight when the surrounding rock stability monitoring data exceeds the preset stability range.

[0062] Through the collaborative efforts of the aforementioned components, this system constructs a multi-objective collaborative optimization mechanism for the withdrawal operation of ultra-high fully mechanized mining equipment, considering factors such as schedule, cost, and surrounding rock safety under multiple constraints. Simultaneously, it leverages on-site monitoring data to dynamically adjust the plan, overcoming the shortcomings of traditional process arrangement methods, such as weak constraint integration capabilities, objective imbalance, and poor adaptability to working conditions. Specifically: The disturbance mapping module, as the front-end data processing and constraint modeling unit of the system, centrally collects all related data on working face equipment, transport roadways, and support systems. It maps complex constraints such as disassembly dependencies, spatial adjacency, transport occupancy, and support continuity into a heterogeneous disturbance attribute map, transforming scattered on-site operation rules into standardized structured data. This provides a unified constraint benchmark and data support for all subsequent functional modules, serving as a fundamental prerequisite for the entire system's operation. The matrix generation module directly calls the node and edge attribute data of the disturbance attribute map, combines support equipment parameters and surrounding rock pressure parameters to divide spatial and temporal constraints, quantifies the safety thresholds of each device at different operating times, and generates a disturbance safety window matrix. This transforms the underground surrounding rock safety boundary into a numerical model that the algorithm can recognize, realizing the transformation of safety requirements from engineering experience to digital indicators, and setting hard safety red lines for subsequent coding and optimization stages. The sequence encoding module completes the process encoding conversion based on the topological logic of the perturbation attribute graph. It maps four types of work content—equipment dismantling sequence, transportation route, personnel grouping, and reinforcement support—to corresponding coding bits and combines them into multi-segment chromosomes. This transforms the decision variables in actual engineering into gene sequences that can be processed by intelligent algorithms, completing the data docking between engineering operation logic and intelligent optimization algorithms. The fitness calculation module simultaneously associates the multi-segment chromosomes with the perturbation safety window matrix. By decoding and restoring the complete withdrawal plan, it sequentially calculates the total project duration, comprehensive cost, and various violation penalties. Combined with preset perturbation penalty coefficients and transportation congestion weights, it calculates multi-objective fitness vectors, providing quantitative evaluation basis for algorithm iteration and individual selection. The sequence generation module uses multi-objective fitness and the perturbation safety window matrix as judgment criteria. It performs perceptual crossover and mutation operations to complete population iteration, selects high-quality individuals through non-dominated sorting, identifies and repairs various constraint conflict nodes, and continuously outputs compliant candidate withdrawal sequences. It is the core computing unit for the system to achieve multi-objective intelligent optimization. The solution output module dynamically routes and allocates tasks for candidate pullback sequences, verifies four rigid indicators based on on-site resources, selects the optimal non-dominated solution from compliant solutions, and outputs a formally feasible target pullback solution, completing the final transformation of algorithm optimization results into on-site operation plans. The parameter update module, as the core of the system's dynamic feedback, collects real-time data on on-site operation time, transportation status, and surrounding rock monitoring, dynamically corrects the disturbance penalty coefficient and transportation congestion weight. When the surrounding rock status deviates from the safe range, it links all upstream calculation modules to re-iterate and generate a priority safety-ensuring downgrade pullback sequence, enabling the static optimization solution to cope with dynamic disturbances underground. The seven modules are sequentially connected and interconnected according to data flow logic. The output of the preceding module serves as the calculation input for the following module, and the on-site feedback data from the following module in turn corrects the overall calculation parameters, forming a complete closed-loop operation logic. This allows the entire system to possess both offline precise optimization and online dynamic adjustment capabilities, fully adapting to the complex pullback operation environment of ultra-high fully mechanized mining faces.

[0063] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the multi-objective optimization method for the retreat process of an ultra-high mining fully mechanized mining equipment.

[0064] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the methods and systems described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0065] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment, characterized in that, include: S1. Obtain equipment node information, disassembly dependency relationship, spatial adjacency relationship, transportation occupation information and support continuity relationship of ultra-high fully mechanized mining face, and construct disturbance attribute map; S2. Extract the support spacing, mining height, and passage support parameters from the disturbance attribute map, and obtain the periodic pressure step distance and support resistance attenuation curve from the surrounding rock pressure parameters; determine the spatial constraints based on the support spacing, mining height, and passage support parameters, including the overhang area constraint, adjacent demolition spacing constraint, and passage support strength constraint; determine the time constraints based on the periodic pressure step distance and support resistance attenuation curve, including the periodic pressure avoidance constraint and support resistance attenuation constraint; combine the spatial and time constraints to calculate the safety threshold of each equipment node in different time periods, and generate a disturbance safety window matrix. S3. Based on the dismantling dependency and spatial adjacency relationship in the disturbance attribute graph, the order of device dismantling in the initial pullback sequence is mapped to the device node sequence position; Based on the transportation occupancy information in the disturbance attribute map, the feasible transportation routes of each piece of equipment in the underground passage are mapped to transportation path positions; the qualification type and operational capability data of the work teams are obtained, and the personnel grouping for demolition and transportation tasks is mapped to work team positions; the support requirement data of the surrounding rock disturbance area is obtained, and the additional support operations for the surrounding rock disturbance area are mapped to support reinforcement positions; the equipment node sequence positions, transportation path positions, work team positions, and support reinforcement positions are sequentially spliced ​​into a multi-segment chromosome; S4. Decode the multi-segment chromosome based on the perturbation safety window matrix to obtain the decoding results, which include the dismantling sequence, dismantling time, transportation route, shift allocation, and support reinforcement scheme of each piece of equipment. Based on the decoding results, calculate the total project duration and comprehensive cost. The total project duration includes equipment dismantling time, transportation time, and temporary support installation time. The comprehensive cost includes dismantling cost, transportation cost, temporary support cost, and personnel cost. Based on the dismantling sequence and support reinforcement scheme, and combined with the perturbation safety window matrix, calculate the penalty for excessive rock exposure, the penalty for continuous removal of adjacent supports, and the penalty for cross-occupancy of passageways. Weight the penalty for excessive rock exposure and the penalty for continuous removal of adjacent supports according to the initial perturbation penalty coefficient, and weight the penalty for cross-occupancy of passageways according to the initial transportation congestion weight. Combine these with the total project duration and comprehensive cost to calculate the multi-objective fitness. S5. Perform perceptual crossover mutation operation on multi-segment chromosomes according to the perturbation safety window matrix, select according to multi-target fitness, identify and repair conflict nodes, and generate candidate retraction sequences; S6. Perform rigid verification on the candidate withdrawal sequence for surrounding rock exposure time, adjacent withdrawal interval, transportation bottleneck and peak personnel exposure, and output the target withdrawal plan; S7. Obtain actual on-site time, transportation waiting time, and surrounding rock stability monitoring data; update the disturbance penalty coefficient and transportation congestion weight; and generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight when the surrounding rock stability monitoring data exceeds the preset stability range.

2. The method according to claim 1, characterized in that, S1 includes: Obtain equipment node information, disassembly dependencies, and spatial adjacency relationships for hydraulic supports, scraper conveyors, and coal mining machines in ultra-high fully mechanized mining faces; Obtain transportation occupancy information for intake airway, return airway and haulage roadway, as well as the support continuity relationship of support equipment; Extract key feature data from the equipment node information, the disassembly dependency relationship, the spatial adjacency relationship, the transportation occupancy information, and the support continuity relationship; The key feature data is mapped to nodes and edges in a graph structure to generate the perturbation attribute graph.

3. The method according to claim 1, characterized in that, S5 includes: Based on the perturbation safety window matrix, a perceptual crossover mutation operation is performed on the multi-segment chromosome to obtain the mutated chromosome; Decode the mutated chromosome, recalculate the multi-target fitness, and obtain the mutated multi-target fitness. The mutated multi-objective fitness is compared with the stated multi-objective fitness, and non-dominant individuals are selected. Identify conflict nodes in non-dominated individuals that violate the constraints of the disturbance safety window matrix and disturbance attribute graph, including equipment in non-safe dismantling periods, equipment whose adjacent dismantling intervals do not meet the minimum interval constraint, and equipment that causes channel overload; The conflict nodes are sequentially adjusted in terms of time, path, shift, and support to generate a candidate pullback sequence that satisfies the constraints of the disturbance safety window matrix and the disturbance attribute graph.

4. The method according to claim 1, characterized in that, S6 includes: Acquire channel occupancy data, transportation equipment load data, team qualification data, and support and reinforcement resource data, and dynamically route and allocate dismantling tasks, transportation tasks, and reinforcement tasks in the candidate withdrawal sequence. Rigid verification is performed on the candidate pullback sequence after dynamic routing allocation, including verifying whether the exposure time of the surrounding rock exceeds the safety threshold, verifying whether the removal interval of adjacent equipment is less than the minimum removal interval, verifying whether the number of equipment in the channel exceeds the maximum carrying capacity, and verifying whether the number of workers on the working face at the same time exceeds the safety limit. For the candidate pullback sequences that pass the rigid verification, the optimal non-dominated solution is identified as the target pullback scheme based on the corresponding multi-objective fitness.

5. The method according to claim 1, characterized in that, S7 includes: Acquire actual on-site time, transportation waiting time, and surrounding rock stability monitoring data, including support resistance, roof subsidence, and internal squeezing of the two sides. The disturbance penalty coefficient is updated based on the actual on-site time and the surrounding rock stability monitoring data, and the transportation congestion weight is updated based on the transportation waiting time. If the surrounding rock stability monitoring data exceeds the preset stability range, then based on the updated disturbance penalty coefficient and transportation congestion weight, the multi-objective fitness is recalculated according to the logic of S4 to S6, and the perception cross-mutation and rigid verification are performed to generate a downgrade backoff sequence that prioritizes the stability of the surrounding rock.

6. A multi-objective optimization system for the retreat process of ultra-high fully mechanized mining equipment, used to implement the multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment as described in any one of claims 1 to 5, characterized in that, include: The disturbance mapping module is used to obtain equipment node information, disassembly dependencies, spatial adjacency relationships, transportation occupancy information, and support continuity relationships of ultra-high fully mechanized mining faces, and to construct a disturbance attribute map. The matrix generation module is used to generate a disturbance safety window matrix based on the disturbance attribute map, combined with the support equipment parameters and the surrounding rock pressure parameters; The sequence encoding module is used to encode the initial retreat sequence into a multi-segment chromosome containing equipment node sequence bits, transportation path bits, work group bits, and support reinforcement bits based on the disturbance attribute map. The fitness calculation module is used to decode the multi-segment chromosome based on the perturbation safety window matrix and calculate the multi-objective fitness according to the initial perturbation penalty coefficient and transportation congestion weight. The sequence generation module is used to perform a perceptual crossover mutation operation on the multi-segment chromosome according to the perturbation safety window matrix, select according to the multi-target fitness, identify and repair conflict nodes, and generate candidate retraction sequences. The scheme output module is used to rigidly verify the surrounding rock exposure time, adjacent removal interval, transportation bottleneck and personnel exposure peak of the candidate withdrawal sequence, and output the target withdrawal scheme. The parameter update module is used to acquire the actual on-site time, transportation waiting time and surrounding rock stability monitoring data, update the disturbance penalty coefficient and the transportation congestion weight, and generate a downgrade rollback sequence based on the updated disturbance penalty coefficient and transportation congestion weight when the surrounding rock stability monitoring data exceeds the preset stability range.

7. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements a multi-objective optimization method for the retreat process of ultra-high fully mechanized mining equipment as described in any one of claims 1-5.

Images