An intelligent optimized room pillar type upward slicing waste rock filling mining method

Abstract

This application relates to the fields of mining engineering and mine backfilling technology, and discloses an intelligent optimized room-and-pillar type upward layered waste rock backfilling mining method, including: constructing a voxel model of the spatial state of the mining area, a waste rock batch and backfilling unit model, a time-varying operation network model, and a local mechanical proxy model; determining the goaf boundary, backfilling state probability field, candidate range of effective support interface, and local stability response quantity based on 3D point cloud, waste rock records, visible backfilling surface, and mechanical monitoring data; generating candidate mining and backfilling schemes, verifying pre-operation dependencies and spatiotemporal reachability conditions, and obtaining a set of feasible schemes; determining decision-sensitive areas based on changes in the screening results under state assumptions, generating directional data acquisition tasks and updating the state; comprehensively evaluating the set of feasible schemes, determining the mining and backfilling scheme, and passing the non-adjustable state and the updatable inherited state to the next layer. This invention helps to improve the matching degree between the mining and backfilling scheme and the state, and maintain the continuity of the state of adjacent layers.

Description

Technical Field

[0001] This invention relates to the field of mining engineering and mine backfilling technology, specifically to an intelligent and optimized room-and-pillar type upward layered waste rock backfilling mining method. Background Technology

[0002] Room-and-pillar upward layered mining typically involves extracting the stope in a layered sequence and using waste rock generated during the mining process to fill the goaf. The development of a mining plan requires comprehensive consideration of the ore body's spatial morphology, pillar condition, goaf boundaries, waste rock material conditions, backfill support conditions, underground access conditions, and local surrounding rock stability.

[0003] Existing mining plans are mostly determined based on geological models, stope design parameters, and historical production data. As mining and backfilling operations progress, the actual goaf boundary, waste rock accumulation morphology, roof contact state, and local rock mass mechanical parameters may change. Field measurement data, waste rock material records, and mechanical monitoring data are usually managed separately by different systems, making it difficult to establish timely correlations between spatial states, resulting in discrepancies between the design state and the actual field conditions.

[0004] Meanwhile, the existing scheme selection process still lacks sufficient consideration of the coordinated factors of backfill support conditions, operational sequence, equipment access conditions, and channel occupancy status. When some on-site conditions are uncertain, supplementary monitoring areas and tasks are usually determined manually, making it difficult to judge whether the relevant uncertainties will affect the feasibility of candidate schemes. Furthermore, the current layered pillars, backfills, channels, and operational status lack a unified inheritance method when moving to the next layer, potentially increasing the workload of subsequent scheme adjustments.

[0005] Therefore, this invention proposes an intelligent and optimized room-and-pillar type upward layered waste rock backfill mining method to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an intelligent and optimized room-and-pillar type upward layered waste rock backfilling mining method, which solves the problems that the mining and backfilling scheme is difficult to dynamically match with changes in the goaf, backfilling state, and local stability during room-and-pillar type upward layered waste rock backfilling mining process, and that the continuity of adjacent layer states is insufficient.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent and optimized room-and-pillar type upward layered waste rock backfilling mining method, comprising: Construct a voxel model of the spatial state of the mining area, a waste rock batch and backfilling unit model, a time-varying operation network model, and a local mechanical proxy model; The three-dimensional boundary of the actual goaf is determined based on the three-dimensional point cloud data. The probability field of the filling state is calculated based on the waste rock material record and the visible filling surface, and the candidate range of the effective support interface is determined. The local rock mass mechanical state parameters are updated based on the mechanical monitoring data, and the local mechanical proxy model or local three-dimensional finite difference numerical model is called according to the applicable domain of the local mechanical proxy model to obtain the local stability response. Based on the actual three-dimensional boundary of the goaf, the candidate range of the effective support interface and the local stability response, candidate mining and filling schemes are generated, and the pre-operation dependencies and spatiotemporal accessibility conditions are verified to obtain a set of feasible candidate schemes. Generate at least two state hypotheses for uncertain state variables, write the state hypotheses into the corresponding filling state probability field, local rock mass mechanical state parameters or time-varying operation network model and re-screen them to obtain a set of feasible candidate schemes corresponding to each state hypothesis. When the set of feasible candidate solutions corresponding to different state assumptions is not the same, the decision-sensitive area is determined and a targeted data collection task is generated. Supplementary acquisition data is obtained according to the directional data acquisition task, and the channel state in the filling state probability field, local rock mass mechanical state parameters, or time-varying operation network model is updated and re-filtered based on the supplementary acquisition data; A comprehensive evaluation is performed on the set of feasible candidate solutions obtained from the re-screening, and the selected adoption solution is determined and output. The current layered state that cannot be adjusted is defined as an unadjustable state and passed on as the initial constraint of the next layer. The state that can be updated based on field measurement data is defined as an updatable inherited state and passed on as the initial state of the next layer.

[0008] Preferably, a voxel model of the spatial state of the mining area and a waste rock batch and backfill unit model are constructed, including: The mining area is voxelized to obtain a regular three-dimensional voxel set; a multi-dimensional attribute state vector is configured for each voxel, which includes voxel spatial coordinates, ore and rock type, rock mass mechanical parameters, mining status identifier and filling status identifier; The waste rock is divided into at least two waste rock batches and batch attribute data is configured. The goaf area to be filled is divided into at least two filling units and filling demand status data is configured. The batch attribute data includes block size distribution parameters, fine particle content ratio and moisture content. The filling demand status data includes the demand target value and allowable deviation scale of the corresponding attribute indicators. When a batch of waste rock meets the lithological suitability conditions of the filling unit, the attribute suitability is calculated, and the batch of waste rock with an attribute suitability not less than the attribute suitability threshold is determined as a usable batch of waste rock.

[0009] Preferably, the attribute fit is calculated according to the following formula: ; In the formula, Indicates the first The waste rock batch and the first The attribute compatibility between individual filling units; This indicates the total number of attribute metrics involved in the matching; Indicates the first Item attribute indicators; Indicates the first The weighting coefficient of the item attribute indicator, with a value range of 100%. ,and ; Indicates the first The waste rock batch in the first The actual parameter values ​​under the item attribute indicator; Indicates the first The target value of the filling unit under the a-th attribute indicator; Indicates the first The filling unit for the first The allowable deviation scale for the attribute indicator. .

[0010] Preferably, constructing the local mechanical proxy model includes: The local mechanical sample set is divided into a training sample set and a validation sample set; a multilayer perceptron surrogate model is adopted. The input of the multilayer perceptron surrogate model is the goaf span, layer height, effective bearing width of the pillar, effective support contact area ratio, local rock mass uniaxial compressive strength and local rock mass elastic modulus, and the output is the maximum principal stress, the maximum displacement of the roof and the volume ratio of the plastic zone. The network weights are updated using the training sample set, and the prediction error is determined using the validation sample set; the applicable domain is determined based on the range of values ​​of the input parameters in the training sample set and the distance between the normalized local mechanical state vector and the nearest training sample.

[0011] Preferably, calculating the filling state probability field and determining the candidate range of effective support interfaces includes: The stackable volume of the waste rock batch is determined based on the actual mass and loose bulk density of the waste rock batch; the filling unit is discretized into at least two filling state voxels, and at least two candidate states of waste rock spatial distribution are generated based on the unloading location, unloading time sequence, stackable volume of the waste rock batch, angle of repose of the waste rock and the visible filling surface. Candidate states for waste rock spatial distribution that meet any of the following conditions are excluded: the absolute value of the difference between the total volume of waste rock and the stackable volume of the waste rock batch is greater than the volume allowable error; the candidate state for waste rock spatial distribution crosses the three-dimensional boundary of the actual goaf; the nearest point distance between the outer surface of the candidate state for waste rock spatial distribution and the visible filling surface within the scanning coverage area corresponding to the visible filling surface is greater than the measurement allowable error. The probability of waste rock occupation and the probability of roof contact are obtained by statistically analyzing the candidate states of the spatial distribution of remaining waste rock. The region where the probability of roof contact is not less than the support state probability threshold and meets the connectivity condition and the minimum support area condition is determined as the candidate range of the effective support interface.

[0012] Preferably, updating the local rock mass mechanical state parameters includes: Among the local stability response quantities, the response quantity that has the same type as the field-measured mechanical response is determined as the theoretical mechanical response; when at least one mechanical response index satisfies the following formula, a genetic algorithm is used for inversion: ; The inversion objective function is: ; In the formula, Indicates the first The inversion objective function value for a local region; This represents the total number of mechanical response indicators; Indicates the mechanical response index number; Indicates the first In the local region, the first Theoretical mechanical response; Indicates the first In the local region, the first The mechanical response was measured on-site. Indicates the weighting coefficient. ,and ; Represents the normalization scale, and ; After the inversion stops, the local rock mass mechanical state parameters are updated using the individual parameter with the smallest inversion objective function value.

[0013] Preferably, generating the candidate sampling and filling scheme includes: A hybrid coding chromosome is adopted, with the permutation coding segment representing the mining sequence of the ore to be mined, the integer coding segment representing the allocation relationship between waste rock batches and backfilling units and the transportation route number, and the real number coding segment representing the stope geometric parameters, the effective bearing width of the pillar and the layer height; Repair the crossed or mutated chromosomes so that the permutation coding segment includes each mining room number to be returned once, and make the coding values ​​in the integer coding segment and the real number coding segment fall within the value range determined by the voxel model of the mining area spatial state, the candidate range of the effective support interface, and the local stability response quantity.

[0014] Preferably, verifying the job's prerequisites and spatiotemporal reachability includes: For each candidate mining and filling scheme, a directed dependency graph of mining task support units is constructed; when a circular dependency is detected, the candidate mining and filling scheme is excluded; when no circular dependency is detected, a topological sort is performed to obtain a set of preceding tasks. Verify that the effective support contact area ratio is not lower than the allowable lower limit of the effective support contact area ratio, the maximum displacement of the roof plate is not greater than the allowable displacement value, and the planned start time of the task to be returned to the mining chamber is not earlier than the sum of the latest expected end time of the set of preceding tasks and the safety time interval. The time-varying operation network model is used to verify whether there are continuous available paths for the equipment and the batch of waste rock carried by the equipment that meet the requirements of channel status, allowed passage direction and equipment type.

[0015] Preferably, determining the decision-sensitive region and generating the targeted data acquisition task includes: Divide the allowable state range of continuous state variables into at least two discrete state intervals of equal width, and determine the predefined state categories of discrete state variables as discrete states; calculate the state variables according to the following formula. Normalized state information entropy: ; In the formula, This represents the total number of discrete states, and ; Indicates the discrete state number; Represents the state probability. ,and ; When the summation term is zero, the corresponding term is taken as 0. State variables whose normalized state information entropy is not less than the uncertainty judgment threshold are identified as the state variables to be measured. Write at least two state assumptions of the state variable to be tested into the corresponding filling state probability field, local rock mass mechanical state parameters or time-varying operation network model and re-select them; When the set of feasible candidate solutions corresponding to different state assumptions is different, the state variable to be tested is determined as the decision-sensitive state variable, and the region where the filling state voxel is located, the local mechanical calculation region, or the roadway region represented by the network edge is determined as the decision-sensitive region. Targeted data collection tasks are generated based on the decision-sensitive area, the type of state to be confirmed, the required data type, and the allowed collection time interval.

[0016] Preferably, determining the selected sampling scheme and transmitting the non-adjustable state and the updatable inherited state includes: Extract at least two production evaluation indicators from the set of feasible candidate solutions obtained from the re-screening: recovery rate, dilution rate, preparation work volume, total waste rock transportation distance, production capacity, backfilling work volume, and local stability response. The efficiency-based and cost-based production evaluation indicators are standardized. The comprehensive evaluation score is calculated based on the sum of the products of the standardized evaluation values ​​and the weighting coefficients. The candidate procurement scheme with the highest score is determined as the selected procurement scheme. The mining status of recovered elements, the records of completed backfilling operations, the status of permanently retained pillars, the status of channels that must be retained, and the status of irrevocable operations are identified as the unadjustable status and locked. The effective support interface candidate range, the state probability of the effective support interface candidate range, and the unexecuted channel plan state are determined as the updatable inheritable state; The non-adjustable state and the updatable inherited state are written into the cross-layer state dataset and passed to the next layer.

[0017] Preferably, the voxel model of the mining area spatial state is constructed by voxelizing the target mining area under a unified three-dimensional coordinate system of the mining area, and each voxel is configured with a multi-dimensional attribute state vector; the multi-dimensional attribute state vector includes voxel spatial coordinates, ore and rock type, rock mass mechanical parameters, mining status identifier, filling status identifier, and data source credibility label.

[0018] Preferably, the waste rock batch and backfilling unit model is constructed by dividing waste rock into multiple waste rock batches and establishing batch attribute data, and dividing the goaf area to be backfilled into multiple backfilling units and establishing backfilling demand status data; when the waste rock batch meets the waste rock lithology matching conditions of the backfilling unit, the attribute matching degree is calculated based on the batch attribute data and backfilling demand status data, and the available waste rock batch of the corresponding backfilling unit is determined based on the attribute matching degree.

[0019] Preferably, the time-varying operation network model includes network nodes and network edges; the network nodes include longwall faces, waste rock loading locations, waste rock temporary storage locations, backfilling units, equipment dwell locations, and monitoring equipment deployment locations; the network edges include transport roadways and connecting passages, and each network edge is configured with passage length, permitted passage direction, passage status, equipment occupancy time interval, permitted equipment type, and passage speed limit.

[0020] Preferably, the local mechanical surrogate model is trained based on a local mechanical sample set generated by a local three-dimensional finite difference numerical model; the inputs of the local mechanical surrogate model include the goaf span, layer height, effective bearing width of the pillar, effective support contact area ratio, local uniaxial compressive strength of the rock mass, and local elastic modulus of the rock mass, and the outputs include the maximum principal stress, the maximum displacement of the roof, and the volume ratio of the plastic zone; the applicable domain of the local mechanical surrogate model is determined according to the training value range of the input parameters and the training sample coverage conditions.

[0021] Preferably, multiple candidate spatial distribution states of waste rock are generated based on the waste rock unloading location, unloading time sequence, batch stackable volume of waste rock, angle of repose of waste rock, and visible filling surface; candidate spatial distribution states of waste rock that do not meet material balance constraints, cross the actual three-dimensional boundary of the goaf, or are inconsistent with the visible filling surface are excluded, and the waste rock occupation probability and roof contact probability are calculated based on the remaining candidate spatial distribution states of waste rock; the area where the roof contact probability meets the support state probability threshold, connectivity condition, and minimum support area condition is determined as the effective support interface candidate range.

[0022] Preferably, the response quantity in the local stability response quantity that is the same as the field measured mechanical response type is taken as the theoretical mechanical response, and the theoretical mechanical response is compared with the field measured mechanical response; when the difference in mechanical response exceeds the corresponding allowable range, the local rock mass uniaxial compressive strength and local rock mass elastic modulus are inverted using a genetic algorithm, and the local rock mass mechanical state parameters are updated using the inversion results.

[0023] Preferably, the candidate mining and filling scheme is generated using a hybrid coding chromosome; the hybrid coding chromosome includes an arrangement coding segment representing the mining sequence of the mining block to be recovered, an integer coding segment representing the allocation relationship between waste rock batches and filling units and the transportation path number, and a real number coding segment representing the geometric parameters of the stope, the effective bearing width of the pillar and the layer height.

[0024] Preferably, a directed dependency graph of mining task support units is constructed for candidate mining and filling schemes, and the cyclic dependency, effective support contact area ratio, maximum roof displacement and operation time constraints are verified according to the directed dependency graph of mining task support units; the existence of continuous available paths that meet the requirements of channel status, allowed passage direction and equipment type is verified according to the time-varying operation network model.

[0025] Preferably, the uncertainty of the state variable is determined based on the discrete states of the state variable and the state probabilities of each discrete state, and multiple state hypotheses are generated for the state variables that meet the uncertainty judgment conditions; when different state hypotheses cause changes in the set of feasible candidate solutions, the corresponding state variable is determined as a decision-sensitive state variable, and the decision-sensitive region is determined based on the spatial objects associated with the decision-sensitive state variable.

[0026] Preferably, a targeted data acquisition task is generated based on the decision-sensitive area; when there is no data acquisition device that meets the data type, data accuracy, and spatiotemporal accessibility conditions within the allowed acquisition time interval, the candidate acquisition scheme is re-screened using the state with the most unfavorable constraint on the scheme within the allowed state range of the state variable to be tested; when no feasible candidate scheme is found after using the most unfavorable state, the back-acquisition task associated with the decision-sensitive area is set to an unexecutable state.

[0027] Preferably, the benefit-type production evaluation indicators and cost-type production evaluation indicators of each candidate procurement scheme in the feasible candidate scheme set are extracted, each production evaluation indicator is standardized and a comprehensive evaluation score is calculated, and the candidate procurement scheme with the highest comprehensive evaluation score is determined as the selected procurement scheme.

[0028] Preferably, the mining status of recovered voxels, the records of completed backfilling operations, the status of permanently retained pillars, the status of channels that must be retained, and the status of irrevocable operations are determined as unadjustable states; the candidate range of effective support interfaces, the state probability of the candidate range of effective support interfaces, and the status of channels that have not yet been executed are determined as updatable inheritable states; the unadjustable states are used as the initial constraints for the next layer of unadjustable conditions, and the updatable inheritable states are used as the initial updatable states for the next layer.

[0029] This invention provides an intelligent and optimized room-and-pillar type upward layered waste rock backfill mining method. It has the following beneficial effects: 1. This invention enables the geometric state of the goaf, the waste rock filling state, the local mechanical state, and the underground operating conditions to jointly participate in the determination of the mining and filling scheme through the collaborative processing of mining area state modeling, on-site state updating, candidate mining and filling scheme generation, constraint screening, supplementary data collection, comprehensive evaluation, and cross-layer state transfer. This helps to reduce the impact of the disconnect between various state data on the scheme selection results.

[0030] 2. This invention calculates the probability field of filling state based on waste rock material records, visible filling surfaces, and the three-dimensional boundary of the actual goaf, and determines the candidate range of effective support interfaces accordingly. This allows the filling state judgment to simultaneously consider waste rock material input, spatial boundaries, and on-site measurement results, providing a basis for subsequent mining permits and pillar state judgments.

[0031] 3. This invention selects a local mechanical surrogate model or a local three-dimensional finite difference numerical model to calculate the local stability response based on the applicable domain of the local mechanical surrogate model, and updates the local rock mass mechanical state parameters in combination with on-site mechanical monitoring data, so that the local stability evaluation can correspond to the current mining state; at the same time, by verifying the pre-operation dependency conditions and spatiotemporal accessibility conditions, candidate mining and filling schemes that do not meet the support, operation sequence or access conditions can be eliminated.

[0032] 4. This invention identifies decision-sensitive areas by comparing changes in candidate sampling results under different state assumptions, and generates targeted data collection tasks for these areas, which helps to focus supplementary monitoring on state objects that affect scheme selection. By transmitting unadjustable states and updatable inherited states, it can provide a continuous data foundation for scheme generation and state updates at the next level. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the system architecture according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the calculation process of the probability field of the filling state and the derivation of the effective support interface in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the application domain determination and adaptive update process of the local mechanics proxy model in this embodiment of the invention. Figure 5 This is a flowchart illustrating the filtering process for job prerequisites and spatiotemporal reachability collaborative constraints in an embodiment of the present invention. Figure 6 This is a flowchart illustrating the decision-sensitive area identification and data closed-loop update process according to an embodiment of the present invention. Detailed Implementation

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Reference Figure 1 This invention provides an intelligent and optimized room-and-pillar type upward layered waste rock backfilling mining system, which includes a model building module, a state assimilation module, a scheme screening module, a data supplementation module, and a scheme selection module.

[0036] The model building module is used to construct voxel models of spatial state in mining areas, waste rock batch and backfill unit models, time-varying operation network models, and local mechanical proxy models.

[0037] The state assimilation module is used to determine the three-dimensional boundary of the actual goaf based on the three-dimensional point cloud data, calculate the probability field of the filling state based on the waste rock material records and the visible filling surface, and determine the candidate range of the effective support interface; it is also used to update the local rock mass mechanical state parameters based on the on-site mechanical monitoring data, and call the local mechanical proxy model or the local three-dimensional finite difference numerical model to calculate the local stability response.

[0038] The scheme screening module is used to generate candidate mining and filling schemes. It verifies the pre-operation dependency conditions based on the directed dependency graph of the mining task support unit, and verifies the spatiotemporal accessibility of the equipment and the waste rock batch it carries based on the time-varying operation network model, thus obtaining a set of feasible candidate schemes.

[0039] The data supplementation module is used to identify decision-sensitive areas that can change the screening results of candidate acquisition schemes, generate targeted data collection tasks, and update the corresponding status based on the supplemented data.

[0040] The scheme selection module is used to comprehensively evaluate the set of feasible candidate schemes, determine the selected scheme, and pass the unadjustable state and updatable inheritance state formed in the current layer to the next layer.

[0041] Reference Figure 2 This invention provides an intelligent and optimized room-and-pillar type upward layered waste rock backfill mining method, comprising the following steps: S100: Acquire 3D geological data of the ore body, waste rock material data, filling area data, underground passage data, and local mechanical numerical calculation samples to construct a voxel model of the spatial state of the mining area, a waste rock batch and filling unit model, a time-varying operation network model, and a local mechanical proxy model. S200: Based on the on-site 3D point cloud data, determine the actual 3D boundary of the goaf; based on the actual goaf 3D boundary, waste rock material records, and visible backfill surface, calculate the probability field of the backfill state under material balance constraints, and determine the candidate range of the effective support interface. Update the local rock mass mechanical state parameters based on the on-site mechanical monitoring data, and calculate the local stability response by calling the local mechanical proxy model or the local 3D finite difference numerical model according to the applicable domain of the local mechanical proxy model. S300 generates candidate mining and filling schemes based on the actual three-dimensional boundary of the goaf, the candidate range of the effective support interface, the local stability response, the batch status of waste rock and the channel status; verifies the pre-operational dependencies and spatiotemporal accessibility of each candidate mining and filling scheme to obtain a set of feasible candidate schemes. S400 generates multiple state hypotheses for uncertain state variables. When different state hypotheses cause changes in the permitting status of the stope, the retention status of the pillars, the batch allocation of waste rock, the transportation route, the closure status of the passage, or the feasibility of the candidate mining and filling schemes, it identifies the decision-sensitive area and generates a targeted data collection task. Based on the supplementary collected data, the corresponding states are updated, and a new set of feasible candidate schemes is obtained. S500 comprehensively evaluates the set of feasible candidate schemes based on at least two of the following: recovery rate, dilution rate, preparation work volume, production capacity, waste rock transportation and backfilling operation volume, and local stability response volume, and determines and outputs the selected mining and backfilling scheme; it takes the mining status of the recovered voxels formed in the current layer, the completed backfilling operation records, the status of permanently retained pillars, the status of channels that must be retained, and the status of irrevocable operations as the unadjustable initial constraints for the next layer; it takes the candidate range of the effective support interface and its status probability, as well as the status of the unexecuted channel plan, as the updatable inheritable status for the next layer.

[0042] The following describes the processing procedures of each functional module in steps S100 to S500.

[0043] The following describes the basic data model construction process in step S100. When the system initializes or enters the current mining layer, the model construction module executes S101 to S104 to form a voxel model of the spatial state of the mining area, a waste rock batch and backfill unit model, a time-varying operation network model, and a local mechanical proxy model.

[0044] S101, the model building module obtains the three-dimensional geological model data of the ore body in the target mining area, and performs voxelization processing on the target mining area under the unified three-dimensional coordinate system of the mining area to obtain a regular three-dimensional voxel set.

[0045] The side length of each voxel is determined based on the minimum operating width of the mining equipment and the required discretization accuracy of the ore body boundary. As one implementation method, the voxel side length is taken as one-third to one-half of the minimum operating width of the mining equipment. The minimum operating width of the mining equipment is the minimum passage width required for the equipment to pass through the target area during the current layered mining or backfilling operation. The required discretization accuracy of the ore body boundary is determined by the degree of boundary undulation in the 3D geological model of the ore body and the accuracy of the stope design.

[0046] The model building module configures a multi-dimensional attribute state vector for each voxel. This multi-dimensional attribute state vector includes voxel spatial coordinates, ore / rock category, rock mass mechanical parameters, mining status identifier, backfilling status identifier, and data source credibility label. The mining status identifier includes "awaiting mining" and "mined" states; the backfilling status identifier includes "unfilled," "partially filled," and "filled" states. Voxels in both the mined and unfilled states constitute the goaf area.

[0047] Rock mass mechanics parameters include the local uniaxial compressive strength and local elastic modulus of the rock mass, with units of MPa and GPa, respectively. The data source credibility tag indicates the source of the voxel state data, including geological exploration data, production design data, direct field measurement data, and state estimation data.

[0048] In S102, the model building module divides the waste rock generated from underground mining operations into multiple waste rock batches and establishes batch attribute data for each batch. The batch attribute data includes the waste rock batch number, waste rock source area, waste rock generation time, actual mass of the waste rock batch, loose bulk density, block size distribution parameters, fine particle content ratio, moisture content, and stackable volume. The loose bulk density is determined by on-site sampling measurement data or historical batch statistics under the same source area and block size distribution conditions; the stackable volume is determined by dividing the actual mass of the waste rock batch by the corresponding loose bulk density.

[0049] The actual batch weight of waste rock is obtained from on-site weighing equipment; the block size distribution parameters are obtained from image recognition equipment or screening statistics; the proportion of fine particles and moisture content are determined from on-site test data or batch sampling data.

[0050] The model building module divides the goaf to be filled into multiple filling units and establishes filling demand status data for each filling unit. The filling demand status data includes the filling unit number, target filling volume, waste rock lithology adaptation conditions, block size distribution adaptation conditions, fine particle content adaptation conditions, moisture content adaptation conditions, target values ​​of each attribute index, allowable deviation scale of each attribute index, entry channel identifier, coordinates of the predetermined unloading position, and spatial adjacency data between the filling unit and adjacent stops, adjacent pillars, and surrounding rock.

[0051] The model building module calculates the first [number] based on batch attribute data and filling requirement status data. The waste rock batch and the first The attribute compatibility between each filling unit is determined, and an attribute compatibility table between waste rock batches and filling units is generated. The attribute compatibility is calculated using the following formula: ; In the formula, Indicates the first The waste rock batch and the first The property compatibility between each filling unit is a dimensionless value. This indicates the total number of attribute metrics involved in the matching; Indicates the first Item attribute indicators; Indicates the first The weighting coefficient of the item attribute indicator, with a value range of 100%. And the sum of the weight coefficients of each attribute indicator is 1; Indicates the first The waste rock batch in the first The actual parameter values ​​under the item attribute indicator; Indicates the first The target value of the filling unit under the a-th attribute indicator; Indicates the first The filling unit for the first The allowable deviation scale for the attribute indicator, and with and They have the same dimensions.

[0052] Permissible deviation scale The design tolerances for the corresponding filling unit should be determined based on the historical filling quality inspection data of the target mine or the results of on-site filling tests, and should meet the following requirements. .

[0053] The model building module first determines whether the lithology of the waste rock batch meets the waste rock lithology compatibility conditions of the filling unit. If the conditions are met, the block size distribution attribute, fine particle content attribute, and moisture content attribute are used for attribute compatibility calculation. If the conditions are not met, the corresponding waste rock batch is identified as an unusable waste rock batch for the filling unit, and the attribute compatibility between the two is not calculated. The stackable volume of the waste rock batch is not included in the attribute compatibility calculation; it is verified separately by the material balance conditions in the waste rock batch allocation relationship. Weighting coefficients It is determined by the analytic hierarchy process or historical mine backfill quality data.

[0054] In this embodiment, the block size distribution attribute is represented by the median particle size of the waste rock batch, the fine particle content attribute is represented by the mass percentage of fine particles, and the moisture content attribute is represented by the moisture content of the waste rock batch. The weighting coefficients for the block size distribution attribute, fine particle content attribute, and moisture content attribute are 0.40, 0.35, and 0.25, respectively; the corresponding allowable deviation scales are 20 mm, 0.10, and 0.05, respectively. Both the mass percentage of fine particles and the moisture content are expressed as mass fractions. Attribute adaptation threshold. Take 0.80.

[0055] when Greater than or equal to the attribute adaptation threshold At that time, the model building module will... The waste rock batch is marked as the No. The available waste rock batches for each filling unit are recorded and written into the waste rock batch and filling unit attribute adaptation table.

[0056] Attribute adaptation threshold The value ranges from 0.75 to 0.90, and its specific value is determined by the historical backfilling quality inspection data of the target mine, the actual usage records of waste rock batches, or the on-site initialization parameters. The attribute adaptation threshold is used to determine whether the waste rock batch meets the material adaptation requirements of the backfilling unit, and does not directly indicate that the backfilling unit has formed an effective support.

[0057] S103, the model building module constructs a time-varying job network model. The time-varying job network model includes network nodes and network edges.

[0058] Network nodes include longwall faces, waste rock loading locations, waste rock temporary storage locations, backfilling units, equipment dwell locations, and monitoring equipment deployment locations. Network edges include transport roadways and connecting passages within the mining area.

[0059] The model building module configures the channel length, allowed traffic direction, channel status, device occupancy time interval, allowed device types, and traffic speed limits for each network edge. Channel status includes open, occupied, and closed states.

[0060] The model building module records the channel state of each network edge according to the discrete time interval, forming the network edge state data corresponding to each discrete time interval.

[0061] S104, the model building module obtains the local mechanical sample set calculated by the local three-dimensional finite difference numerical model. Based on the ore body's three-dimensional geological model, rock mechanics test data, and stope design parameters, the module determines the value range of each input parameter. Using the Latin hypercube sampling method, it generates 1000 local mechanical state vectors within the value range and inputs them into the local three-dimensional finite difference numerical model to obtain the corresponding local stability response. Finally, a local mechanical surrogate model is constructed based on the local mechanical sample set.

[0062] The local mechanics sample set includes multiple local mechanics samples. Each local mechanics sample includes a local mechanics state vector and a local stability response quantity corresponding to that local mechanics state vector.

[0063] In this embodiment, the local mechanical state vector consists of the goaf span, layer height, effective bearing width of the pillar, effective support contact area ratio, local uniaxial compressive strength of the rock mass, and local elastic modulus of the rock mass.

[0064] In this embodiment, the local stability response consists of the maximum principal stress, the maximum displacement of the top plate, and the volume ratio of the plastic zone.

[0065] As one implementation method, the local mechanics surrogate model employs a multilayer perceptron surrogate model. The multilayer perceptron surrogate model includes an input layer, two fully connected hidden layers, and an output layer. The input layer is 6-dimensional, the two fully connected hidden layers contain 64 and 32 neurons respectively, and both use linear rectified functions as activation functions. The output layer is 3-dimensional and uses a linear activation function.

[0066] The model building module divides the local mechanics sample set into a training sample set and a validation sample set in an 8:2 ratio. The training sample set is used to calculate the loss function and update the network weights; the validation sample set does not participate in the network weight update and is used to calculate the prediction error of the local mechanics surrogate model.

[0067] The model building module performs range normalization on the input and output parameters in both the training and validation sample sets, mapping each parameter to a value range of 0 to 1. The parameter range used for normalization is determined based on the training sample set and is also used for normalization of the validation sample set and the input local mechanical state vector. The adaptive moment estimation optimization method has an initial learning rate of 0.001, a batch size of 32, and a maximum training epoch of 1000. After each training epoch, the mean square error of the training and validation sample sets is calculated, and the network weights corresponding to the minimum mean square error of the validation sample set are saved.

[0068] When the mean square error of the training sample set is no greater than 1×10 -4 Alternatively, the reduction in the mean square error of the validation sample set relative to the historical minimum value within 30 consecutive training rounds is no greater than 1×10. -5 Training should be stopped when the mean square error of the validation sample set is no greater than 2.5 × 10⁻⁶. -3 At that time, the local mechanics proxy model is set to a callable state; the mean square error of the verification sample set is greater than 2.5 × 10⁻⁶. -3 At that time, the local mechanics proxy model is not set to a callable state.

[0069] The model building module determines the applicable domain of the local mechanics surrogate model based on the value range of each local mechanics state vector in the local mechanics sample set. The applicable domain includes the minimum and maximum values ​​of each input parameter, as well as the distance range between the normalized local mechanics state vector and the training samples.

[0070] The model building module calculates the Euclidean distance between each training sample and its nearest neighbor training sample, and determines the maximum value of the distance as the sample coverage distance threshold. When the Euclidean distance between the normalized local mechanical state vector to be input and the nearest training sample is not greater than the sample coverage distance threshold, it is determined that it meets the training sample coverage condition.

[0071] When the local mechanical state vector to be input is within the applicable domain, the local mechanical proxy model is in a callable state; when the local mechanical state vector to be input is outside the applicable domain, the model building module does not call the local mechanical proxy model and generates a local three-dimensional finite difference numerical model calculation request.

[0072] The following describes the state assimilation and local model update process in step S200. The state assimilation module receives on-site 3D point cloud data, waste rock material records, transportation records, and on-site mechanical monitoring data, and updates the actual goaf 3D boundary, filling state probability field, local rock mass mechanical state parameters, and local mechanical proxy model.

[0073] S201, Extraction of three-dimensional boundaries and updating of geometric state of actual goaf.

[0074] When the current layer enters a mining-filling optimization cycle, or after the completion of the mining operation unit in the current layer, the state assimilation module acquires the three-dimensional point cloud data of the goaf through the downhole three-dimensional laser scanning equipment, and performs outlier removal and spatial filtering on the three-dimensional point cloud data.

[0075] The state assimilation module performs initial point cloud registration based on the fixed features of the mine area's measurement control points or roadways, and then uses an iterative nearest-point algorithm for fine registration. During fine registration, the goal is to minimize the sum of squared distances between corresponding points, solving for the rotation matrix and translation vector required for point cloud coordinate transformation, until the change in registration error between two adjacent iterations does not exceed a preset convergence threshold. The preset convergence threshold is determined based on the measurement accuracy of the 3D laser scanning equipment.

[0076] After point cloud registration is completed, the state assimilation module generates the actual 3D boundary of the goaf based on the processed 3D point cloud data. It then calculates the spatial difference between the actual goaf boundary and the designed goaf boundary to obtain the actual exposed space, over-excavated area, and under-excavated area. The geometric residual includes the volume of the over-excavated area, the volume of the under-excavated area, and the sum of the two.

[0077] The state assimilation module updates the mining status of the corresponding voxels in the mining area spatial state voxel model to the mined status based on the actual three-dimensional boundary of the goaf. It also updates the filling status of the corresponding voxels that have not yet been filled to the unfilled status and updates the data source credibility label of the corresponding voxels to direct field measurement data.

[0078] S202, Calculation of the probability field of the filling state and determination of the candidate range of the effective support interface.

[0079] Reference Figure 3 The state assimilation module discretizes the filling unit to be evaluated into multiple filling state voxels and obtains the waste rock unloading location, unloading time sequence, actual mass of waste rock batch, stackable volume, block size distribution parameters, and visible filling surface obtained by scanning.

[0080] The stackable volume of a waste rock batch is determined based on the actual mass of the batch and the corresponding loose bulk density. The loose bulk density is determined by on-site sampling and measurement data or historical batch statistics. The angle of repose of the waste rock is determined by the lithology, block size distribution, and on-site stacking test results of the corresponding waste rock batch.

[0081] The state assimilation module generates multiple candidate states for waste rock spatial distribution using Monte Carlo simulation based on the unloading location, unloading time sequence, stackable volume, waste rock angle of repose, and visible filling surface. Each simulation adds simulated material units in batches according to the unloading time sequence; the simulated material units move along the direction of gravity from their corresponding unloading locations, selecting unoccupied voxels with lower elevations and satisfying the waste rock angle of repose condition from adjacent voxels, until no adjacent voxels satisfying the condition remain, at which point movement stops. The unloading location deviation, waste rock angle of repose, and simulated material unit movement direction are randomly selected within their respective measurement error ranges. Each candidate state consists of multiple simulated material units with preset volumes, and the sum of the volumes of all simulated material units equals the stackable volume of the corresponding waste rock batch. In this embodiment, 100 candidate states for waste rock spatial distribution are generated for each waste rock batch; the volume of a single simulated material unit equals the volume of a filling state voxel, and when the remaining stackable volume is less than the volume of a filling state voxel, the remaining stackable volume is used as the volume of the last simulated material unit.

[0082] For each candidate state of waste rock spatial distribution, the state assimilation module calculates the total volume of waste rock corresponding to the candidate state and determines whether it is within the allowable error range of the actual stackable volume. The allowable error range is determined based on the weighing equipment error and the loose bulk density measurement error, and is calculated by error propagation from the maximum allowable mass error of the weighing equipment and the loose bulk density measurement error.

[0083] When a candidate state exceeds the allowable error range, crosses the three-dimensional boundary of the actual goaf, or the nearest point distance between the outer surface of the candidate state and the visible filling surface within the scanning coverage area corresponding to the visible filling surface is greater than the measurement allowable error of the three-dimensional laser scanning device, the state assimilation module excludes the candidate state of the waste rock spatial distribution.

[0084] The state assimilation module counts the number of times each filling state voxel is occupied by waste rock and the number of times it comes into contact with the top plate in the remaining candidate states, and divides them by the total number of valid candidate states to obtain the waste rock occupation probability and the top plate contact probability of each filling state voxel, thus forming the filling state probability field.

[0085] The state assimilation module extracts voxels with a top plate contact probability not less than the support state probability threshold, and performs spatial connectivity analysis on the voxels according to the twenty-six neighborhood connectivity rules. The top plate contact area that satisfies the connectivity condition and the minimum support area condition is determined as the candidate range of the effective support interface.

[0086] The support state probability threshold and minimum support area are determined based on historical backfilling detection data, physical accumulation test data, or numerical simulation verification results of the target mine, and are written into the model during the current layer initialization. In this embodiment, the support state probability threshold is set to 0.80, and the minimum support area is set to 60% of the area of ​​the target roof contact region.

[0087] S203, Layer residual verification and local rock mass mechanical state parameter update.

[0088] The state assimilation module calculates the theoretical filling requirements of the current filling unit based on the actual three-dimensional boundary of the goaf and the boundary of the filling unit, and corrects the material input state based on the actual quality of the waste rock batch.

[0089] After completing the correction of geometric state and material input state, the state assimilation module calculates the expected filling volume and expected roof contact area based on the filling state probability field, and compares them with the theoretical filling requirement and the target roof contact area, respectively, to obtain the filling volume residual rate and the roof contact area residual rate. The filling volume residual rate is the absolute value of the difference between the expected filling volume and the theoretical filling requirement divided by the theoretical filling requirement; the roof contact area residual rate is the absolute value of the difference between the expected roof contact area and the target roof contact area divided by the target roof contact area.

[0090] In this embodiment, the allowable values ​​for both the filling volume residual rate and the roof contact area residual rate are set to 0.05. When the filling volume residual rate is greater than the allowable value, or the roof contact area residual rate is greater than the allowable value, the state assimilation module regenerates the candidate states for the spatial distribution of waste rock. If any residual rate after regeneration is still greater than the corresponding allowable value, the corresponding filling state variable is determined as the state variable to be measured, and the state variable to be measured, the corresponding filling unit number, and the residual rate exceeding the allowable value are transferred to the data supplementation module without initiating the inversion of local rock mass mechanical state parameters. When the filling volume residual rate is not greater than the allowable value and the roof contact area residual rate is not greater than the allowable value, the remaining mechanical residual is calculated.

[0091] The state assimilation module extracts on-site mechanical monitoring data corresponding to the current local area and the current monitoring period based on the spatial coordinates of the monitoring equipment and the data acquisition time. The on-site mechanical monitoring data includes at least one of displacement monitoring data, stress monitoring data, and microseismic monitoring data.

[0092] Displacement monitoring data is compared with the theoretical displacement response, and stress monitoring data is compared with the theoretical stress response. Microseismic monitoring data is used to determine the disturbed local area and the corresponding monitoring time interval, and is not directly subtracted from the stress or displacement values.

[0093] The state assimilation module compares the current local mechanical state vector with the applicable domain of the local mechanical surrogate model. When the state is within the applicable domain, the local mechanical surrogate model is used to calculate the theoretical mechanical response. When the state is outside the applicable domain, the local three-dimensional finite difference numerical model is used to calculate the theoretical mechanical response. The module then compares each theoretical mechanical response with the same type of field-measured mechanical response to obtain the remaining mechanical residual.

[0094] When at least one mechanical response index satisfies the following formula, the state assimilation module, within the allowable range of local rock mass mechanical state parameters, uses a genetic algorithm to invert the local uniaxial compressive strength and local rock mass elastic modulus: ,in, Indicates the first In the local region, the first Theoretical mechanical response; Indicates the corresponding first The mechanical response was measured on-site. Indicates the first The normalized scale for the mechanical response index has the same units as the corresponding theoretical and field-measured mechanical responses. The allowable range of the parameters is determined based on rock mechanics test data, geological exploration data, and historical inversion results.

[0095] The genetic algorithm encodes the local rock mass mechanical state parameters to be inverted into individual parameter sequences and generates an iterative population through selection, crossover, and mutation operations. In this embodiment, the genetic algorithm population size is 80, the crossover probability is 0.80, the mutation probability is 0.10, and a tournament selection method is used, randomly selecting 3 individuals from the population each time, and the individual with the smallest inversion objective function value is selected as the parent individual. The fitness of each individual is determined according to the inversion objective function, which is calculated according to the following formula: ; In the formula, Indicates the first The inversion objective function value for each local region is a dimensionless value. This represents the total number of mechanical response indices involved in the inversion. Indicates the first Mechanical response index; Indicates the first The weighting coefficients of the mechanical response index satisfy the following conditions: And the sum of all weight coefficients is 1.

[0096] Normalized scale The weighting is determined based on the allowable error of the corresponding monitoring equipment, the allowable deviation of the response index, or the historical observation range. In this embodiment, each mechanical response index uses the same weight. .

[0097] When the decrease in the minimum inversion objective function value relative to the previous minimum value for 20 consecutive generations is no greater than 1×10 -4 Alternatively, when the number of iterations reaches 200, the state assimilation module stops the inversion calculation and updates the rock mechanics state parameters of the corresponding local area using the individual parameter with the smallest inversion objective function value.

[0098] S204, Judgment and update of the applicable domain of the local mechanics proxy model.

[0099] Reference Figure 4 The state assimilation module compares the updated local rock mass mechanics state vector with the applicable domain of the local mechanics surrogate model. The applicable domain includes the training value range of each input parameter and the coverage conditions of the training samples within the corresponding state region.

[0100] When all input parameters of the local mechanical state vector are within the training range and the corresponding state region meets the training sample coverage condition, the state assimilation module calls the local mechanical proxy model to obtain the local stability response.

[0101] When any input parameter exceeds the training value range, or when the current state region does not meet the training sample coverage condition, the state assimilation module does not call the local mechanics proxy model.

[0102] The state assimilation module extracts the current local region from the voxel model of the spatial state of the mining area. Based on the actual three-dimensional boundary of the goaf, the candidate range of the effective support interface, the state of the adjacent area, and the updated local rock mass mechanical state parameters, it sets the geometric boundary, displacement boundary, and initial geostress conditions of the local three-dimensional finite difference numerical model.

[0103] The local three-dimensional finite difference numerical model adopts the Mohr-Coulomb elastoplastic constitutive model. The cohesion, internal friction angle, tensile strength, elastic modulus, and Poisson's ratio in the constitutive model are determined based on rock mechanics test results from the target mine. The actual goaf's three-dimensional boundary is set as a free boundary. Regions with a roof contact probability not less than the support state probability threshold and satisfying connectivity conditions are set as unidirectional contact boundaries, which transmit compressive stress but not tensile stress. Regions in the actual goaf's three-dimensional boundary where no unidirectional contact boundary is formed are set as free boundaries. The bottom boundary of the local three-dimensional finite difference numerical model constrains displacement in three coordinate directions, the side boundaries constrain normal displacement, and the top boundary applies vertical stress determined by the self-weight of the overlying strata. The initial geostress is determined based on field geostress test results; if field geostress test results are unavailable, it is determined based on the self-weight of the overlying strata and the lateral pressure coefficient used in the target mine.

[0104] A local three-dimensional finite difference numerical model is used to obtain a reference mechanical response by solving the mechanical equilibrium equations of a local region. The reference mechanical response consists of the maximum principal stress, the maximum displacement of the top plate, and the volume percentage of the plastic zone.

[0105] The state assimilation module takes the current local mechanical state vector as the input sample and the corresponding reference mechanical response as the output label to form supplementary training samples, and adds the supplementary training samples to the local mechanical sample set.

[0106] The state assimilation module retrains the local mechanics surrogate model based on the updated local mechanics sample set, following the sample partitioning method, normalization method, and training parameters in step S104. It then uses a validation sample set to determine whether the updated local mechanics surrogate model meets the preset prediction error requirements. When the mean square error of the validation sample set is no greater than 2.5 × 10⁻⁶, the model is considered as follows: -3 When the updated local mechanics proxy model is determined to meet the preset prediction error requirements, it is determined that the updated model meets the preset prediction error requirements.

[0107] When the updated local mechanics surrogate model meets the preset prediction error requirements, the state assimilation module saves the updated model parameters and redetermines the applicable domain of the local mechanics surrogate model based on the updated local mechanics sample set.

[0108] When the updated local mechanics surrogate model does not meet the preset prediction error requirements, the original local mechanics surrogate model and its applicable domain are retained, and the current local mechanics state is continued to be marked as an outside state of the applicable domain; the reference mechanical response output by the local three-dimensional finite difference numerical model is used for the current local region.

[0109] Step S200 outputs the actual three-dimensional boundary of the goaf, geometric residual, probability field of filling state, candidate range of effective support interface, residual rate of filling volume, residual rate of roof contact area, updated local rock mass mechanical state parameters, local stability response, and updated local mechanical surrogate model.

[0110] The following describes the candidate sampling scheme generation and collaborative constraint screening process in step S300. (Combined with...) Figure 5 As shown, the scheme selection module receives the updated voxel model of the spatial state of the mining area, the probability field of the filling state, the candidate range of the effective support interface, the local stability response, the batch attribute data of waste rock and the time-varying operation network model, generates candidate mining and filling schemes, and verifies the pre-operation dependency conditions and spatiotemporal reachability conditions in sequence.

[0111] S301, candidate sampling and filling schemes are generated.

[0112] The scheme selection module generates candidate mining and filling schemes using a genetic algorithm based on the ore chamber to be returned, available waste rock batches, filling units, and time-varying operation network model.

[0113] The chromosomes in the genetic algorithm employ a hybrid coding structure. The permutation coding segment represents the mining sequence of each ore block to be returned; the integer coding segment represents the allocation relationship between waste rock batches and backfilling units, as well as the transport path number; and the real number coding segment represents the stope geometry, effective pillar bearing width, and layer height. The transport path number is selected from the set of available transport paths in a connected state within the current planning period in the time-varying operation network model. Voxel states with locking indicators, permanently retained pillar states, must-retain passage states, and irrevocable operation states are not written into the chromosome and serve as unadjustable initial constraints during the generation of candidate mining and backfilling schemes. The value range of each coding parameter is determined by the voxel model of the mining area spatial state, the candidate range of the effective support interface, and the local stability response.

[0114] The scheme selection module calculates the fitness of candidate schemes based on the production completion rate, equipment utilization rate, and estimated total operation time. The calculation method is as follows: ; In the formula, Indicates the first The fitness of each candidate sampling scheme is a dimensionless value. Indicates the first The production completion rate of each candidate mining and filling scheme is the smaller of the ratio of the expected output to the planned output and 1. Indicates the first The equipment utilization rate of each candidate acquisition and supply scheme is determined by the average ratio of the planned working time to the corresponding available working time of each participating equipment; when the planned working time of any equipment exceeds its available working time, the corresponding chromosome is determined as an infeasible chromosome. and These represent the weighting coefficients for production completion rate and equipment utilization rate, respectively. Both are greater than 0 and less than 1, and their sum is 1. Indicates the first The estimated total operation time for each candidate extraction and filling scheme; This indicates the total planned time for the current shift cycle; both use the same time unit.

[0115] Weighting coefficient and This is determined based on the production volume requirements and equipment usage requirements in the current tiered production plan. In this embodiment, , .

[0116] The scheme selection module performs selection, crossover, and mutation operations based on the generated fitness of each chromosome. In this embodiment, the genetic algorithm uses a population size of 100, a crossover probability of 0.80, and a mutation probability of 0.10. A tournament selection method is used, randomly selecting 3 chromosomes from the population each time, and the chromosome with the highest generated fitness is used as the parent chromosome. Specifically, sequential crossover and exchange mutation are used for permutation coding segments, integer site crossover and random reset mutation are used for integer coding segments, and arithmetic crossover and boundary mutation are used for real number coding segments.

[0117] After each crossover or mutation, the scheme selection module checks whether there are duplicate or missing mine room numbers in the permutation coding segment, and checks whether the integer coding segment and the real number coding segment exceed the corresponding value range. When there are duplicate mine room numbers in the permutation coding segment, the duplicate mine room numbers are replaced with the missing mine room numbers in the order of their arrangement in the permutation coding segment; when the integer coding segment or the real number coding segment exceeds the corresponding value range, the coding value that exceeds the value range is corrected to the boundary value closest to that coding value.

[0118] After chromosome repair is completed, the scheme selection module continues to execute the genetic algorithm iteratively. The algorithm is considered successful if the improvement in the highest generated fitness over 30 consecutive generations relative to the previous highest generated fitness is no greater than 1 × 10⁻⁶. -4 The genetic algorithm iteration stops when the number of iterations reaches 300. After the iteration ends, the scheme selection module decodes the retained chromosomes to obtain multiple candidate mining and filling schemes. Each candidate mining and filling scheme includes stope geometry parameters, effective pillar bearing width, layer height, allocation relationship between waste rock batches and filling units, transportation path, mining operation sequence, and filling operation sequence.

[0119] S302, Job prerequisite dependency validation.

[0120] The scheme selection module constructs a directed dependency graph of mining task support units for each candidate mining and filling scheme. The directed dependency graph includes nodes for the ore storage room to be returned, ore pillar status, filling unit, transportation channel, monitoring task, and equipment evacuation task.

[0121] Directed dependency edges represent the preceding dependencies between corresponding nodes, including the dependency of the stope mining task on the support status of adjacent filling units, the dependency of the pillar parameter adjustment on the status of adjacent stopes, the dependency of the filling task on the status of the transportation channel, and the dependency of the channel closure task on the completion status of the monitoring task and the equipment evacuation task.

[0122] The solution selection module first performs loop detection on the directed dependency graph. When a circular dependency is detected, the corresponding candidate solution is determined to not meet the job prerequisite dependency conditions. When no circular dependency is detected, a topological sort is performed on the directed dependency graph to obtain the set of prerequisite tasks for each task.

[0123] For the For each ore chamber awaiting return, the scheme selection module determines the preconditions for support based on the ratio of the effective support contact area of ​​adjacent filling units and the current local stability response.

[0124] The effective support contact area ratio is determined by the ratio of the candidate area of ​​the effective support interface of adjacent filling units to the area of ​​the target roof contact area of ​​that filling unit. The area of ​​the target roof contact area is determined by the filling design data. The allowable lower limit of the effective support contact area ratio is determined by the filling quality inspection data, physical packing test data, or local mechanical calculation results of the target mine.

[0125] The maximum displacement of the roof in the local stability response should not exceed the allowable displacement value of the corresponding local area. The allowable displacement value is determined based on rock mechanics test data, mine safety design parameters, or the stability assessment results of a local three-dimensional finite difference numerical model.

[0126] The scheme selection module verifies whether the planned start time of the mining room to be returned meets the time requirements of its predecessor tasks according to the following formula.

[0127] ; In the formula, Indicates the first The planned start time for the mission to return to the mining site; Indicates the first A set of prerequisite tasks for the pending return mining mission; Indicates the prerequisite task The expected end time; Indicates the first The safe time interval corresponding to each pending mining task. The above time parameters use the same time unit.

[0128] Preliminary tasks include at least one of the following: confirming the support status of adjacent filling units, mining related stopes, monitoring, and equipment evacuation. Safety time interval. It is determined based on the type of preceding task, equipment operation records, and historical delay data of the target mine.

[0129] When the preconditions for the mining block to be returned are not met, or the planned start time of the mining block to be returned does not meet the aforementioned time constraints, the scheme screening module will determine the corresponding candidate mining and filling scheme as not meeting the preconditions for operation and exclude it.

[0130] S303, Spacetime reachability check.

[0131] The scheme selection module performs spatiotemporal reachability verification on candidate sampling schemes that meet the pre-operation dependency conditions.

[0132] The scheme selection module divides the current scheduling cycle into multiple discrete time intervals based on the operation time series of candidate procurement schemes. The step size of the discrete time interval is determined based on the expected travel time of the participating scheduling equipment through the shortest network edge in the time-varying operation network model, and is set to be no greater than half of the expected travel time.

[0133] The scheme selection module updates the open, occupied, and closed states of network edges in each discrete time interval according to the operation time sequence, and records the batches of waste rock carried by the equipment and transportation equipment and the network nodes where the data acquisition equipment is located.

[0134] For each mobile task, the solution selection module calculates the estimated time for the equipment and its batch of waste rock to enter and leave each network edge based on the channel length and the equipment's historical statistical travel speed under the corresponding channel type. The historical statistical travel speed is determined based on equipment positioning records and transportation operation records.

[0135] The scheme screening module verifies edge by edge whether the channel status, allowed passage direction and equipment type of the network edge meet the passage conditions when the equipment and the batch of waste rock it carries arrive at the corresponding network edge, and determines whether the passage time between adjacent network edges is continuous.

[0136] When there is no continuous available path for the equipment and the batch of waste rock it carries within the required time interval, or when the candidate mining and filling scheme causes the ventilation channel, equipment evacuation channel, waste rock transportation channel or monitoring channel that must be retained to be closed in advance, the scheme screening module will determine that the candidate mining and filling scheme does not meet the spatiotemporal accessibility conditions and exclude it.

[0137] The scheme selection module will form a set of feasible candidate schemes that simultaneously meet the pre-operation dependency conditions and the spatiotemporal accessibility conditions.

[0138] The following describes the process of identifying decision-sensitive areas and supplementing data in step S400. (Combined with...) Figure 6 As shown, the data supplementation module reads the probability field of the filling state, local rock mass mechanical state parameters, local stability response and time-varying operation network model, as well as the state variables to be measured, corresponding object identifiers and residual rates transmitted by the state assimilation module, identifies uncertain state variables that may change the screening results of candidate mining and filling schemes, and updates the corresponding states according to the supplemented data.

[0139] S401, Uncertain state variable extraction and state hypothesis generation.

[0140] The data supplementation module extracts state variables with multiple possible states from the infill state probability field, local rock mass mechanical state parameters, and time-varying operation network model.

[0141] For continuous state variables, the data supplementation module divides their allowable state range into 5 discrete state intervals of equal width based on monitoring error, parameter allowable range, or historical observation interval; for discrete state variables, the predefined state categories are directly used.

[0142] The data supplementation module calculates the state variables according to the following formula. Normalized state information entropy: ; In the formula, Represents state variables The normalized state information entropy is a dimensionless value; Represents state variables The total number of discrete states, and ; Indicates the discrete state number; Represents state variables In the first The probabilities of discrete states satisfy the following: And the sum of the probabilities of each discrete state is 1. When When the summation term is zero, the corresponding term is taken as 0.

[0143] When the state variable is the waste rock occupation state of a filling voxel, waste rock occupation and non-occupation are set as two discrete states. The probability of the waste rock occupation state is the waste rock occupation probability of the corresponding voxel, and the probability of the non-occupation state is 1 minus the waste rock occupation probability. When the state variable is the roof contact state, roof contact formation and non-formation are set as two discrete states. The probability of roof contact formation is the roof contact probability of the corresponding voxel, and the probability of non-formation is 1 minus the roof contact probability. When the state variable is a continuously monitored variable... The state variable is determined by the 30 most recent valid monitoring records for the current stratum; if fewer than 30 valid monitoring records have been obtained for the current stratum, all valid monitoring records already obtained are used; when the state variable is a local rock mass mechanical state parameter... Based on the effective individual parameters of the last generation population at the time of inversion termination, it falls into the first... The ratio of the number of individuals in each discrete state interval to the total number of valid individuals in the last generation population is used to determine the status. A valid individual is defined as an individual whose parameters are all within their allowable ranges. When the state variable is a channel state, the data supplementation module determines the channel state corresponding to each valid record based on channel planning records, equipment positioning records, and channel state detection records. The channel state belongs to the first The ratio of the number of valid records for each discrete state to the total number of valid records.

[0144] When the normalized state information entropy is not less than the uncertainty judgment threshold At that time, the data supplementation module determines the corresponding state variable as the state variable to be tested. Uncertainty judgment threshold. The value ranges from 0.70 to 0.90, and its specific value is determined based on historical monitoring data of the target mine and verification records of candidate mining and filling schemes. In this embodiment, the uncertainty judgment threshold is... Take 0.80.

[0145] The data supplementation module generates multiple state hypotheses based on the type of the state variable to be measured. For continuous state variables, the state hypotheses include at least the lower limit, median, and upper limit of the allowed state range; when historical probability distribution data exists, state hypotheses are generated supplementarily within the allowed state range according to the historical probability distribution. For discrete state variables, the state hypotheses include all allowed state categories of the state variable.

[0146] The data supplementation module writes each state hypothesis into the corresponding filling state probability field, local rock mass mechanical state parameters, or time-varying operation network model, and calls the scheme screening module to obtain the set of feasible candidate schemes corresponding to each state hypothesis.

[0147] S402, Determination of sensitive areas for decision-making.

[0148] The data supplementation module compares the set of feasible candidate solutions for the same state variable under different state assumptions.

[0149] When different state assumptions lead to changes in the permitting of stope mining, the status of pillar retention, the allocation relationship between waste rock batches and backfilling units, the transportation route, the closure status of passages, or the feasibility of candidate mining and backfilling schemes, the data supplementation module identifies the state variable to be measured as a decision-sensitive state variable.

[0150] The data supplementation module determines the decision-sensitive region based on the spatial objects corresponding to the decision-sensitive state variables. Specifically, the region corresponding to the filling state variable is the region where the corresponding filling state voxel is located; the region corresponding to the local rock mass mechanical state parameter is the corresponding local mechanical calculation region; and the region corresponding to the channel state variable is the roadway region represented by the corresponding network edge.

[0151] S403, Targeted data acquisition task generation and conservative state processing.

[0152] The data supplementation module generates targeted data collection tasks for decision-sensitive areas. These tasks include the spatial area to be collected, the status type to be confirmed, the required data type, data accuracy requirements, the allowed collection time interval, and associated candidate data collection schemes.

[0153] When the required data is three-dimensional point cloud data, the data accuracy requirement is determined based on the voxel side length of the voxel model of the spatial state of the mining area, and the spatial resolution of the point cloud is no greater than half of the corresponding voxel side length; when the required data is displacement, stress or microseismic data, the data accuracy requirement is determined based on the measurement accuracy of the corresponding monitoring equipment and the update requirements of local mechanical state parameters.

[0154] The data supplementation module reads the device type, sensor type, current location, and availability status of the data acquisition equipment. For mobile data acquisition equipment, the module determines whether it can reach the decision-sensitive area within the allowed data acquisition time interval based on the time-varying operation network model; for fixed data acquisition equipment, it determines whether its monitoring range covers the decision-sensitive area.

[0155] The data supplementation module sends targeted data acquisition tasks to data acquisition devices that meet the requirements of data type, data accuracy, and spatiotemporal accessibility.

[0156] When no data acquisition device meets the conditions within the allowed acquisition time interval, the data supplementation module uses the state value that is most unfavorable to the corresponding constraint within the allowed state range of the state variable to be measured, and re-executes the candidate acquisition scheme screening.

[0157] When the state variable to be measured is the uniaxial compressive strength or elastic modulus of the local rock mass, the lower limit of its allowable state range is adopted; when the state variable to be measured is the local stress, the upper limit of its allowable state range is adopted; when the state variable to be measured is the effective support state, the candidate range of the continuous effective support interface with the smallest area among the various state assumptions is taken as the conservative state; for the roof contact voxel that has not reached the support state probability threshold, it is treated as if no effective support has been formed.

[0158] If no feasible candidate solution is found after adopting the worst-case scenario, the data supplementation module will set the back-collection task related to the decision-sensitive area to an unexecutable state.

[0159] S404, Supplementary data collection verification and status update.

[0160] The data supplementation module receives supplementary data returned by the data acquisition device and verifies the data format, acquisition time, spatial coordinates, data integrity, and device identification of the supplementary data.

[0161] When the supplementary data is 3D point cloud data, the state assimilation module performs coordinate registration on the 3D point cloud data and updates the actual 3D boundary of the goaf or the visible filling surface according to the data acquisition object.

[0162] When the supplementary data is displacement monitoring data or stress monitoring data, the state assimilation module updates the on-site measured mechanical response of the corresponding local area.

[0163] When supplementary data is microseismic monitoring data, the state assimilation module extracts the location, number, and energy of microseismic events, and updates the disturbed local area and the corresponding monitoring time interval. Microseismic monitoring data does not directly replace displacement or stress monitoring data for mechanical residual calculation.

[0164] The state assimilation module recalculates the filling state probability field, the candidate range of the effective support interface, or the remaining mechanical residual based on the supplementary collected data, and updates the corresponding local rock mass mechanical state parameters and local stability response quantities.

[0165] The solution selection module re-executes the pre-job dependency condition verification and spatiotemporal reachability condition verification based on the updated status data to obtain an updated set of feasible candidate solutions.

[0166] Step S400 outputs the decision-sensitive area, directional data acquisition task, updated filling state probability field, local rock mass mechanical state parameters, channel state, and set of feasible candidate solutions.

[0167] The following describes the scheme selection and cross-layer state transfer process in step S500. The scheme selection module receives a set of feasible candidate schemes, performs standardized evaluation on each candidate scheme, determines the selected scheme, and transfers the state at the end of the current layer to the next layer.

[0168] S501, Standardization of evaluation indicators for candidate mining and filling schemes.

[0169] The scheme selection module extracts production evaluation indicators for each candidate mining and backfilling scheme from the set of feasible candidate schemes. The production evaluation indicators include at least two of the following: recovery rate, dilution rate, preparation work volume, total waste rock transportation distance, production capacity, backfilling operation volume, and local stability response.

[0170] Recovery rate is determined by the ratio of the expected mined ore quantity corresponding to the candidate mining and filling scheme to the current stratified ore quantity; dilution rate is determined by the ratio of the expected waste rock mixed in to the expected total mined ore and rock quantity; preparation work volume is determined by the excavation volume of the new preparation roadway corresponding to the candidate mining and filling scheme; total waste rock transportation distance is determined by the sum of transportation distances of each waste rock batch; production capacity is determined by the ratio of the expected mined ore quantity in the current stratification optimization cycle to the duration of the current stratification optimization cycle; filling operation volume is determined by the sum of the expected filling volumes of each filling unit; the maximum principal stress, maximum roof displacement, and plastic zone volume ratio in the local area are respectively used as cost-based evaluation indicators.

[0171] Among them, recovery rate and production capacity are benefit-type evaluation indicators. The higher the indicator value, the higher the evaluation result of the corresponding candidate mining and backfilling scheme. Dilution rate, preparation work volume, total waste rock transportation distance and backfilling operation volume are cost-type evaluation indicators. The lower the indicator value, the higher the evaluation result of the corresponding candidate mining and backfilling scheme.

[0172] For benefit-oriented evaluation indicators, the scheme selection module is standardized according to the following formula: ; For cost-based evaluation indicators, the scheme selection module is standardized according to the following formula: ; In the formula, Indicates the first The first candidate procurement scheme The standardized evaluation value is a dimensionless value between 0 and 1. Indicates the first The first candidate procurement scheme Original evaluation index values; and They represent the first and second feasible candidate solutions in the set of feasible solutions, respectively. The maximum and minimum values ​​of the original evaluation indicators, and their correlation with... Use the same units of measurement.

[0173] when At that time, it indicates that each candidate intake and intake scheme is in the first stage. If the values ​​of the evaluation indicators are the same, the scheme selection module will uniformly set the standardized evaluation value of this indicator for each candidate adoption scheme to 1.

[0174] S502, the selected sampling and filling scheme is determined.

[0175] The selected module is calculated according to the following formula: The overall evaluation score of each candidate procurement scheme: ; In the formula, Indicates the first The comprehensive evaluation score of each candidate sampling scheme is a dimensionless value; This represents the total number of production evaluation indicators, and ; Indicates the serial number of the production evaluation indicator; Indicates the first The weighting coefficients of the production evaluation indicators satisfy the following: Furthermore, the sum of the weight coefficients of all production evaluation indicators is 1.

[0176] Weighting coefficient Based on the current tiered production plan. When determining the weight coefficients using the analytic hierarchy process (AHP), the scheme selection module establishes a judgment matrix based on the relative importance of each production evaluation indicator. When the consistency ratio of the judgment matrix is ​​less than 0.10, each element in the normalized eigenvector is determined as the weight coefficient of the corresponding production evaluation indicator.

[0177] In this embodiment, the production evaluation indicators include recovery rate, dilution rate, excavation volume of newly added preparatory roadways, total waste rock transportation distance, production capacity, backfilling operation volume, maximum principal stress, maximum roof displacement, and plastic zone volume ratio, with corresponding weighting coefficients of 0.20, 0.15, 0.10, 0.10, 0.15, 0.10, 0.08, 0.07, and 0.05, respectively.

[0178] The scheme selection module sorts the candidate adoption schemes from high to low according to the comprehensive evaluation score, and determines the candidate adoption scheme with the highest comprehensive evaluation score as the selected adoption scheme.

[0179] When two or more candidate mining and filling schemes have the same comprehensive evaluation score, the candidate mining and filling scheme with the shorter expected total operation time shall be selected first; if the expected total operation time is still the same, the selected mining and filling scheme shall be determined according to the preset scheme number order.

[0180] The scheme selection module outputs the stope geometry parameters, pillar parameters, layer height, allocation relationship between waste rock batches and backfilling units, transportation path, mining operation sequence, and backfilling operation sequence corresponding to the selected mining and backfilling scheme.

[0181] S503, the current layering end status is determined.

[0182] After the current layer is executed, the selected mining and filling scheme module determines the end state of the current layer based on the actual three-dimensional boundary of the goaf, the probability field of the filling state, and the candidate range of the effective support interface output by the state assimilation module, as well as the pillar retention state, passage state, and operation completion state recorded in the mine production record. It then compares this state with the voxel model of the mine area spatial state before the selected mining and filling scheme was executed.

[0183] The current layering completion status includes voxel mining status, voxel filling status, actual pillar status, effective support interface candidate range and its status probability, channel status, and operation execution status.

[0184] The scheme selection module determines the unadjustable state and the updatable inheritable state based on the state comparison results. The unadjustable state is the state in which the current layer has been actually formed and further adjustment will change the existing mining, filling, passage or operation facts; the updatable inheritable state is the state that can be recalculated, corrected or replaced based on the field measurement data obtained from the next layer.

[0185] Unadjustable states include mining states where voxels have already been recovered, completed backfilling operations, permanently retained pillars, passages that must be retained, and irreversible operations that have begun or been completed. The scheme selection module sets a lock flag for unadjustable states.

[0186] The updatable inherited state includes the candidate range of effective support interfaces and their state probabilities. The state probability is the top plate contact probability of each filling state voxel constituting the candidate range of the effective support interface, and is stored along with the corresponding voxel spatial coordinates, as well as the channel plan state that has not yet been executed. The updatable inherited state does not have a locking flag and retains the corresponding data source and state probability.

[0187] S504, cross-layer state transfer.

[0188] The selected module of the scheme will form a cross-layered state dataset by combining the non-adjustable state and the updatable inherited state, as well as the corresponding object identifier, three-dimensional spatial location, state time and data source, and pass it to the mining area spatial state voxel model, waste rock batch and backfilling unit model, time-varying operation network model and task status record corresponding to the next layer.

[0189] For non-adjustable states, the voxel state is written into the mining area spatial state voxel model according to the voxel space coordinates, the filling unit state is written into the waste rock batch and filling unit model according to the filling unit number, the channel state is written into the time-varying operation network model according to the network edge number and the corresponding time interval, and the operation state is written into the corresponding task status record according to the task number, while maintaining the lock mark.

[0190] For updatable inherited states, the effective support interface candidate range and its state probability are written into the waste rock batch and filling unit model and the mining area spatial state voxel model of the next layer according to the filling unit number and voxel space coordinates; the channel plan state that has not yet been executed is written into the time-varying operation network model of the next layer according to the network edge number and the corresponding time interval, and can be updated according to the field measurement data obtained from the next layer. Step S500 outputs the selected mining and filling scheme and the cross-layer state dataset.

[0191] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smartly optimized room-and-pillar, top slicing, waste rock fill mining method characterized by, include: Construct a voxel model of the spatial state of the mining area, a waste rock batch and backfilling unit model, a time-varying operation network model, and a local mechanical proxy model; The three-dimensional boundary of the actual goaf is determined based on the three-dimensional point cloud data. The probability field of the filling state is calculated based on the waste rock material record and the visible filling surface, and the candidate range of the effective support interface is determined. The local rock mass mechanical state parameters are updated based on the mechanical monitoring data, and the local mechanical proxy model or local three-dimensional finite difference numerical model is called according to the applicable domain of the local mechanical proxy model to obtain the local stability response. Based on the actual three-dimensional boundary of the goaf, the candidate range of the effective support interface and the local stability response, candidate mining and filling schemes are generated, and the pre-operation dependencies and spatiotemporal accessibility conditions are verified to obtain a set of feasible candidate schemes. Generate at least two state hypotheses for uncertain state variables, write the state hypotheses into the corresponding filling state probability field, local rock mass mechanical state parameters or time-varying operation network model and re-screen them to obtain a set of feasible candidate schemes corresponding to each state hypothesis. When the set of feasible candidate solutions corresponding to different state assumptions is not the same, the decision-sensitive area is determined and a targeted data collection task is generated. Supplementary acquisition data is obtained according to the directional data acquisition task, and the channel state in the filling state probability field, local rock mass mechanical state parameters, or time-varying operation network model is updated and re-filtered based on the supplementary acquisition data; A comprehensive evaluation is performed on the set of feasible candidate solutions obtained from the re-screening, and the selected adoption solution is determined and output. The current layered state that cannot be adjusted is defined as an unadjustable state and passed on as the initial constraint of the next layer. The state that can be updated based on field measurement data is defined as an updatable inherited state and passed on as the initial state of the next layer.

2. The intelligently optimized room-and-pillar, top slicing, waste rock fill mining method of claim 1, wherein, Constructing a voxel model of the spatial state of the mining area and a batch and backfill unit model of waste rock, including: The mining area is voxelized to obtain a regular three-dimensional voxel set; a multi-dimensional attribute state vector is configured for each voxel, which includes voxel spatial coordinates, ore and rock type, rock mass mechanical parameters, mining status identifier and filling status identifier; The waste rock is divided into at least two waste rock batches and batch attribute data is configured. The goaf area to be filled is divided into at least two filling units and filling demand status data is configured. The batch attribute data includes block size distribution parameters, fine particle content ratio and moisture content. The filling demand status data includes the demand target value and allowable deviation scale of the corresponding attribute indicators. When a batch of waste rock meets the lithological suitability conditions of the filling unit, the attribute suitability is calculated, and the batch of waste rock with an attribute suitability not less than the attribute suitability threshold is determined as a usable batch of waste rock.

3. The intelligently optimized room-and-pillar, top slicing, waste rock fill mining method of claim 2, wherein, The attribute fit is calculated according to the following formula: ； In the formula, Indicates the first The waste rock batch and the first The attribute compatibility between individual filling units; This indicates the total number of attribute metrics involved in the matching; Indicates the first Item attribute indicators; Indicates the first The weighting coefficient of the item attribute indicator, with a value range of 100%. ,and ; Indicates the first The waste rock batch in the first The actual parameter values ​​under the item attribute indicator; Indicates the first The target value of the filling unit under the a-th attribute indicator; Indicates the first The filling unit for the first The allowable deviation scale for the attribute indicator. .

4. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Constructing the local mechanical proxy model includes: The local mechanical sample set is divided into a training sample set and a validation sample set; a multilayer perceptron surrogate model is adopted. The input of the multilayer perceptron surrogate model is the goaf span, layer height, effective bearing width of the pillar, effective support contact area ratio, local rock mass uniaxial compressive strength and local rock mass elastic modulus, and the output is the maximum principal stress, the maximum displacement of the roof and the volume ratio of the plastic zone. The network weights are updated using the training sample set, and the prediction error is determined using the validation sample set; the applicable domain is determined based on the range of values ​​of the input parameters in the training sample set and the distance between the normalized local mechanical state vector and the nearest training sample.

5. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Calculating the probability field of the filling state and determining the candidate range of effective support interfaces includes: The stackable volume of the waste rock batch is determined based on the actual mass and loose bulk density of the waste rock batch; the filling unit is discretized into at least two filling state voxels, and at least two candidate states of waste rock spatial distribution are generated based on the unloading location, unloading time sequence, stackable volume of the waste rock batch, angle of repose of the waste rock and the visible filling surface. Candidate states for waste rock spatial distribution that meet any of the following conditions are excluded: the absolute value of the difference between the total volume of waste rock and the stackable volume of the waste rock batch is greater than the volume allowable error; the candidate state for waste rock spatial distribution crosses the three-dimensional boundary of the actual goaf; the nearest point distance between the outer surface of the candidate state for waste rock spatial distribution and the visible filling surface within the scanning coverage area corresponding to the visible filling surface is greater than the measurement allowable error. The probability of waste rock occupation and the probability of roof contact are obtained by statistically analyzing the candidate states of the spatial distribution of remaining waste rock. The region where the probability of roof contact is not less than the support state probability threshold and meets the connectivity condition and the minimum support area condition is determined as the candidate range of the effective support interface.

6. The intelligent optimized room-and-pillar type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Updating the local rock mass mechanical state parameters includes: Among the local stability response quantities, the response quantity that has the same type as the field-measured mechanical response is determined as the theoretical mechanical response; when at least one mechanical response index satisfies the following formula, a genetic algorithm is used for inversion: ； The inversion objective function is: ； In the formula, Indicates the first The inversion objective function value for a local region; This represents the total number of mechanical response indicators; Indicates the mechanical response index number; Indicates the first In the local region, the first Theoretical mechanical response; Indicates the first In the local region, the first The mechanical response was measured on-site. Indicates the weighting coefficient. ,and ; Represents the normalization scale, and ; After the inversion stops, the local rock mass mechanical state parameters are updated using the individual parameter with the smallest inversion objective function value.

7. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Generating the candidate sampling scheme includes: A hybrid coding chromosome is adopted, with the permutation coding segment representing the mining sequence of the ore to be mined, the integer coding segment representing the allocation relationship between waste rock batches and backfilling units and the transportation route number, and the real number coding segment representing the stope geometric parameters, the effective bearing width of the pillar and the layer height; Repair the crossed or mutated chromosomes so that the permutation coding segment includes each mining room number to be returned once, and make the coding values ​​in the integer coding segment and the real number coding segment fall within the value range determined by the voxel model of the mining area spatial state, the candidate range of the effective support interface, and the local stability response quantity.

8. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Validating the job's prerequisites and spatiotemporal reachability includes: For each candidate mining and filling scheme, a directed dependency graph of mining task support units is constructed; when a circular dependency is detected, the candidate mining and filling scheme is excluded; when no circular dependency is detected, a topological sort is performed to obtain a set of preceding tasks. Verify that the effective support contact area ratio is not lower than the allowable lower limit of the effective support contact area ratio, the maximum displacement of the roof plate is not greater than the allowable displacement value, and the planned start time of the task to be returned to the mining chamber is not earlier than the sum of the latest expected end time of the set of preceding tasks and the safety time interval. The time-varying operation network model is used to verify whether there are continuous available paths for the equipment and the batch of waste rock carried by the equipment that meet the requirements of channel status, allowed passage direction and equipment type.

9. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Determining the decision-sensitive region and generating the targeted data acquisition task includes: Divide the allowable state range of continuous state variables into at least two discrete state intervals of equal width, and determine the predefined state categories of discrete state variables as discrete states; calculate the state variables according to the following formula. Normalized state information entropy: ； In the formula, This represents the total number of discrete states, and ; Indicates the discrete state number; Represents the state probability. ,and ; When the summation term is zero, the corresponding term is taken as 0. State variables whose normalized state information entropy is not less than the uncertainty judgment threshold are identified as the state variables to be measured. Write at least two state assumptions of the state variable to be tested into the corresponding filling state probability field, local rock mass mechanical state parameters or time-varying operation network model and re-select them; When the set of feasible candidate solutions corresponding to different state assumptions is different, the state variable to be tested is determined as the decision-sensitive state variable, and the region where the filling state voxel is located, the local mechanical calculation region, or the roadway region represented by the network edge is determined as the decision-sensitive region. Targeted data collection tasks are generated based on the decision-sensitive area, the type of state to be confirmed, the required data type, and the allowed collection time interval.

10. The intelligent optimized room-and-column type upward layered waste rock backfilling mining method according to claim 1, characterized in that, Determining the selected sampling scheme and transmitting the non-adjustable state and the updatable inherited state includes: Extract at least two production evaluation indicators from the set of feasible candidate solutions obtained from the re-screening: recovery rate, dilution rate, preparation work volume, total waste rock transportation distance, production capacity, backfilling work volume, and local stability response. The efficiency-based and cost-based production evaluation indicators are standardized. The comprehensive evaluation score is calculated based on the sum of the products of the standardized evaluation values ​​and the weighting coefficients. The candidate procurement scheme with the highest score is determined as the selected procurement scheme. The mining status of recovered elements, the records of completed backfilling operations, the status of permanently retained pillars, the status of channels that must be retained, and the status of irrevocable operations are identified as the unadjustable status and locked. The effective support interface candidate range, the state probability of the effective support interface candidate range, and the unexecuted channel plan state are determined as the updatable inheritable state; The non-adjustable state and the updatable inherited state are written into the cross-layer state dataset and passed to the next layer.

Images