A method and system for optimizing gold mine extraction processes based on machine learning
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN RONGDA GOLD CO LTD
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-09
Smart Images
Figure CN122174655A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of process optimization technology, specifically to a method and system for optimizing gold mining processes based on machine learning. Background Technology
[0002] Gold mining typically relies on intelligent scheduling systems to continuously optimize the mining sequence and stope structure parameters, achieving a dynamic balance between efficiency and cost under complex operating conditions and resource constraints. The core task of such systems is to continuously generate executable mining plans under conditions of constantly changing production cycles, strong operational coupling, and frequent environmental disturbances, while maximizing output efficiency and resource utilization while meeting safety boundaries.
[0003] In the rolling optimization process, intelligent scheduling systems, in pursuit of local optima, often make small-scale iterative adjustments to the stope structure parameters over multiple production cycles. While each adjustment may seem reasonable in the current period, its cumulative effect across cycles creates a long-term deviation from the design baseline. This long-term deviation alters the geometric constraints and bearing conditions of the stope structure, causing the stress paths of the surrounding rock to continuously migrate and rearrange throughout historical mining activities. In deep environments, this migration exhibits a clear path-dependent characteristic, meaning that early deviations alter the direction and magnitude of subsequent stress redistribution, gradually evolving locally optimal parameter iterations into a globally unfavorable stress pattern. When the stress pattern evolves unfavorably under the drive of long-term accumulated deviations, events such as local collapses, dynamic instability, or the unavailability of critical passages may occur on-site, forcing the mining sequence to jump, work tasks to be interrupted, and triggering frequent rearrangements. Due to the strong coupling and cascading dependencies between mining operations, these events quickly propagate to transportation organization, support arrangements, and work rhythm, causing production rhythm disruptions, expanded downtime windows, and cost overruns, while significantly increasing the risk of safety accidents and systemic shutdowns.
[0004] The fundamental flaw in existing technologies lies in the fact that rolling optimization typically focuses on single-cycle feasibility checks and local target improvements, solidifying engineering constraints into static or weakly time-varying boundaries. It lacks explicit tracking and quantitative expression of the cumulative deviations of stope structural parameters across cycles, and also lacks a mechanism to establish a learnable mapping between these cumulative deviations and the evolution of the surrounding rock stress path, feeding this mapping back into the rolling optimization process. Because it cannot proactively identify and adaptively adjust constraints for hidden risks that are effective in the current period but unfavorable after accumulating across cycles, existing technologies tend to continuously accumulate unfavorable stress states under efficiency targets, which then manifest as sudden instability and unavailability in later stages. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method and system for optimizing gold mine mining processes based on machine learning, which can effectively solve the problems mentioned in the background technology.
[0006] To achieve the above objectives, the first aspect of the present invention is implemented through the following technical solution: a method for optimizing gold mining process based on machine learning, comprising: collecting production plan data and process execution data of gold mining process, analyzing to obtain process execution dataset of gold mining process, performing differential cumulative analysis on process execution dataset to obtain parameter cumulative offset vector of gold mining process.
[0007] Based on the production plan data and parameter cumulative offset vector of the gold mining process, the characteristic vector of the surrounding rock stress path in the gold mining process is comprehensively constructed.
[0008] By inputting the characteristic vector of the surrounding rock stress path during the gold mining process into the safety margin decay prediction model, the decay amount of the safety reserve coefficient and the risk value of the instability threshold crossing during the gold mining process are obtained.
[0009] Based on the decrease in the safety reserve coefficient and the risk value of the instability threshold in the gold mining process, a dynamic safety constraint set for the gold mining process is obtained, and the gold mining process parameters are optimized according to the dynamic safety constraint set.
[0010] Furthermore, the method for collecting production plan data and process execution data of the gold mining process is as follows: extracting production plan data and process execution data of the gold mining process.
[0011] Production planning data for the gold mine mining process includes the mining sequence, mining task number, planned start time, planned end time, planned output, and planned equipment information. The mining sequence is encoded as a mining sequence sequence with the mining sequence number increasing. The mining sequence sequence is used to represent the order of operations and the dependencies between tasks at the planning level, and provides a unified index for subsequent production cycle division and historical cumulative sequence.
[0012] The process execution data for gold mining includes the mining operation task number, actual start time, actual end time, actual output, and stope structure parameters. The stope structure parameters include pillar width, stope span, segment height, mining step distance, and mining-cutting ratio.
[0013] Furthermore, the method for obtaining the process execution dataset of the gold mine mining process is as follows: after obtaining the production plan data and process execution data, a task-level association relationship between the production plan data and the process execution data is established based on the mining operation task number, and the association verification between the production plan data and the process execution data is performed to obtain the process execution dataset of the gold mine mining process.
[0014] Furthermore, the method for differential cumulative analysis of the process execution dataset is as follows: based on the pre-set benchmark stope structure parameters and combined with the process execution dataset, a comprehensive analysis is performed to obtain the cumulative offset vector of parameters in the gold mine mining process. The cumulative offset vector of parameters includes the cumulative offset vector of stope structure parameters and the mining operation task number and preset production cycle identifier corresponding to the cumulative offset vector of stope structure parameters.
[0015] Furthermore, the method for comprehensively constructing the characteristic vector of the surrounding rock stress path in the gold mining process is as follows: collect the production plan data and cumulative parameter offset vector of the target stope in the gold mining process, analyze the mining sequence sequence in the production plan data, and obtain the historical mining sequence of the target stope and the list of spatial adjacency relationships of the target stope in the gold mining process.
[0016] The cumulative offset vector of the parameters is processed to obtain the cumulative offset data of the target stope and all directly spatially adjacent stopes of the target stope.
[0017] Based on the historical mining sequence, spatial adjacency list, and cumulative offset data, the target mining area's surrounding rock stress path feature vector is obtained by vectorizing and assembling according to preset encoding and combination rules.
[0018] Furthermore, the training method for the safety margin decay prediction model is as follows: construct a safety margin decay training dataset, perform supervised training using the safety margin decay training dataset, and obtain the safety margin decay prediction model.
[0019] The method for constructing the safety margin decay training dataset is as follows: The characteristic vectors of the surrounding rock stress path of the target mining area in the historical production process are collected as input features of the safety margin decay prediction model. The input features are labeled with target labels to form a complete training sample. All training samples are summarized to construct a safety margin decay training dataset. The target labels include the safety reserve coefficient decay amount label and the instability threshold crossing risk value label.
[0020] Furthermore, the method for obtaining the dynamic safety constraint set of the gold mining process is as follows: collect the decrease in safety reserve coefficient and the risk value of instability threshold crossing in the gold mining process, analyze the decrease in safety reserve coefficient, and obtain the dynamic safety reserve coefficient constraint value of the gold mining process.
[0021] By analyzing the risk value of the instability threshold crossing, the adjustment range constraint of the mining site structure parameters in the gold mining process is obtained.
[0022] By analyzing the risk value of instability threshold crossing and its position in the mining sequence, the adjacency constraint of mining sequence in the gold mine mining process is obtained.
[0023] By combining the dynamic safety reserve coefficient constraint value, the adjustment range constraint of the mining site structure parameter, and the adjacent constraint of the mining sequence, a dynamic safety constraint set for the gold mine mining process is obtained.
[0024] Furthermore, the method for optimizing the configuration of gold mining process parameters based on the dynamic safety constraint set is as follows: collect the dynamic safety constraint set, analyze the adjustment range constraints of the stope structure parameters in the dynamic safety constraint set, and obtain the updated values of the stope structure parameters that meet the constraints.
[0025] The adjacency constraints of the mining sequence in the dynamic safety constraint set are analyzed to obtain the start time of the mining operation that meets the constraints.
[0026] The updated values of the mining site structure parameters are combined with the start time of the mining operation to obtain the optimized configuration data of the gold mining process parameters, and then the optimized configuration is carried out based on the optimized configuration data of the gold mining process parameters.
[0027] A second aspect of the present invention provides a machine learning-based gold mine mining process optimization system, comprising: a differential cumulative analysis module, used to collect production plan data and process execution data of the gold mine mining process, analyze to obtain a process execution dataset of the gold mine mining process, perform differential cumulative analysis on the process execution dataset, and obtain a cumulative offset vector of parameters of the gold mine mining process.
[0028] The feature vector construction module is used to comprehensively construct the feature vector of the surrounding rock stress path in the gold mining process based on the production plan data and parameter cumulative offset vector of the gold mining process.
[0029] The feature vector analysis module is used to input the feature vector of the surrounding rock stress path in the gold mining process into the safety margin decay prediction model to obtain the decay of the safety reserve coefficient and the risk value of the instability threshold during the gold mining process.
[0030] The optimization configuration module is used to process the dynamic safety constraint set of the gold mining process based on the decrease of the safety reserve coefficient and the risk value of the instability threshold crossing during the gold mining process, and to optimize the configuration of the gold mining process parameters according to the dynamic safety constraint set.
[0031] The present invention has the following beneficial effects: This invention performs differential cumulative analysis on the process execution dataset to form a parameter cumulative offset vector. This parameter cumulative offset vector is then bound to the mining operation task number and the preset production cycle identifier. This allows the long-term cumulative deviation caused by small-scale iterative adjustments within multiple production cycles to be continuously tracked and quantified. In this way, the deviation accumulation process, which was originally scattered in different cycles and different task records, is uniformly mapped to a computable vector input. This enhances the identifiability and interpretability of rolling optimization for long-term offset risks and reduces the probability of hidden decay of the safety reserve coefficient and subsequent cycle instability caused by single verification passing but cross-cycle cumulative distortion.
[0032] This invention constructs a historical mining sequence and a mining-adjacency relationship list based on the mining sequence sequence in the production plan data. It concatenates the cumulative offset vectors of parameters of the target stope and its directly adjacent spatial stopes in spatial orientation order. At the same time, it introduces the order encoding of the adjacent surfaces being analyzed in the historical sequence to obtain a fixed-dimensional feature vector of the surrounding rock stress path. This achieves a unified structured encoding of the mining sequence, the exposure sequence of adjacent surfaces, and the long-term offset of structural parameters. This allows the changes in the surrounding rock stress path to have a computable expression with both temporal order constraints and spatial adjacency constraints. This reduces the sensitivity of rolling optimization to changes in the original data form and the number of stopes, improves the consistency, reusability, and comparability of inputs between different production cycles, and provides a more stable and discriminative feature input foundation for subsequent safety margin decay prediction models.
[0033] This invention inputs the characteristic vector of the surrounding rock stress path into the safety margin attenuation prediction model, simultaneously outputting the attenuation of the safety reserve coefficient and the risk value of crossing the instability threshold. Based on this, a dynamic safety constraint set is generated. The dynamic safety reserve coefficient constraint value, the adjustment range constraint of the stope structure parameter, and the adjacent constraint of the mining sequence are combined into executable restrictive rules. Then, the stope structure parameters and the start time of the mining operation are matched, updated, and arranged in sequence. This ensures that the process parameter configuration strategy is output under the premise that the safety reserve coefficient calculated by numerical simulation meets the dynamic safety reserve coefficient constraint value. Thus, the risk prediction results are directly transformed into parameter adjustment boundaries and mining sequence boundaries, enhancing the consistency of safety constraints and the stability of process parameter configuration in rolling optimization under complex working conditions, and reducing the risk of skipping, cycle breakage, and systemic shutdown caused by crossing the hidden instability threshold. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation
[0035] The technical solutions of 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.
[0036] Please see Figure 1 The first aspect of this invention provides a technical solution: a method for optimizing gold mining process based on machine learning, comprising collecting production plan data and process execution data of gold mining process, analyzing to obtain process execution dataset of gold mining process, performing differential cumulative analysis on process execution dataset to obtain parameter cumulative offset vector of gold mining process.
[0037] It should be explained that this process aims to unify and quantify the task sequence constraints in the mining plan with the actual execution deviations of the stope structure parameters, so that the continuous deviations of the stope structure parameters in the time dimension can be tracked and accumulated in vector form, providing a stable and reusable input for the construction of the characteristic vector of the stress path of the surrounding rock.
[0038] The methods for collecting production planning data and process execution data during the gold mining process are as follows: Based on the planned scheduling information and on-site mining execution records recorded in the production management system, production plan data and process execution data of the gold mine mining process are extracted respectively. The mining operation task number is used as the consistency association key for the two types of data to form a data alignment structure that can be used for periodic analysis. The specific implementation steps are as follows: For production planning data in the gold mining process, the mining sequence, mining task number, planned start time, planned end time, planned output and planned equipment information are extracted. The mining sequence is encoded into a mining sequence sequence with the mining sequence number increasing. The mining sequence sequence is used to represent the order of operations and the dependencies between tasks at the planning level, and provides a unified index for subsequent production cycle division and historical cumulative sequence.
[0039] For the process execution data of gold mining, the mining operation task number, actual start time, actual end time, actual output and stope structure parameters are extracted. The stope structure parameters include pillar width, stope span, segment height, mining step distance and mining-cutting ratio. The stope structure parameters are used to characterize the structural configuration at the field execution level. The actual start time and actual end time are used to determine the interval to which the task belongs in the production cycle.
[0040] After obtaining production planning data and process execution data, a task-level association relationship between the production planning data and process execution data is established based on the mining operation task number. The association between the production planning data and process execution data is then verified to obtain the process execution dataset of the gold mine mining process.
[0041] The production planning data and process execution data are correlated and validated as follows: When multiple process execution data points correspond to the same mining operation task number, they are sorted by actual start time and merged into a single task execution record; when missing fields exist in the process execution data, they are supplemented using the most recent valid record for the same mining operation task number; when the task boundaries of the process execution data and production planning data intersect, the task records are pruned based on actual start and end times to ensure that each task record can be completely mapped to a unique cycle interval within the preset production cycle. Through the above correlation and validation, a process execution dataset is obtained, indexed by task sequences and centered on stope structure parameters.
[0042] The method for performing differential cumulative analysis on the process execution dataset is as follows: Based on the pre-defined benchmark stope structure parameters and the process execution dataset, a comprehensive analysis is conducted to obtain the cumulative parameter offset vector of the gold mine mining process. The cumulative parameter offset vector is used to characterize the long-term cumulative deviation set between the actual execution of the structural parameters of the entire mining area and the original design. The specific implementation steps are as follows: The benchmark stope structural parameters include pillar width, stope span, sub-section height, mining step distance, and mining-to-cut ratio. A corresponding target value is assigned to each parameter as a reference for differential calculations. These benchmark stope structural parameters remain consistent throughout the analysis to ensure comparability of offsets across different production cycles.
[0043] The process execution data is divided into cycles according to the preset production cycle. The process execution data within the same preset production cycle are grouped into a cycle data set for that production cycle. The records in the cycle data set are sorted according to the mining operation task number and the actual start time, so that the offset calculation within the cycle can follow the time sequence of task execution.
[0044] Within each preset production cycle, the actual stope structure parameters used in the gold mine mining process are compared with the benchmark stope structure parameters one by one to obtain the stope structure parameter offset vector for the current production cycle. The stope structure parameter offset vector includes pillar width offset, stope span offset, segment height offset, mining step offset, and mining-cutting ratio offset. Each offset value is calculated according to the rule of subtracting the benchmark parameter value from the actual parameter value. The positive or negative sign of the offset value is used to indicate the deviation direction of the actual execution relative to the benchmark stope structure parameters.
[0045] After obtaining the offset vector of the stope structure parameters for the current production cycle, the offset vectors of all stope structure parameters generated in the historical production cycles are algebraically summed to obtain the cumulative offset vector of the stope structure parameters in the gold mine mining process. The cumulative offset vector of the stope structure parameters includes the cumulative offset value of the pillar width, the cumulative offset value of the stope span, the cumulative offset value of the segment height, the cumulative offset value of the mining step distance, and the cumulative offset value of the mining-cutting ratio. The summation rule is: according to the time sequence of the production cycle, the offset values of the same parameter dimension are added cycle by cycle to form a continuously accumulated offset trajectory across cycles.
[0046] The cumulative offset vector of the stope structural parameters is bound to the corresponding mining operation task number and preset production cycle identifier, and then summarized to obtain the cumulative offset vector of parameters that characterizes the actual execution of the structural parameters of the entire mining area compared with the original design. This cumulative offset vector can show the formation process of deviation accumulation in the task sequence dimension, and show the direction and magnitude of deviation accumulation in the stope structural parameter dimension, providing an input basis for the subsequent comprehensive construction of the surrounding rock stress path feature vector.
[0047] It should be explained that the above process achieves task-level alignment between production plan data and process execution data by using mining operation task numbers. Within a preset production cycle, the offset of the stope structure parameters relative to the benchmark stope structure parameters is differentially calculated and algebraically accumulated across cycles to obtain the parameter cumulative offset vector. This allows the continuous deviation of pillar width, stope span, segment height, mining step distance, and mining-cut ratio in the time dimension to be continuously tracked and quantified. Thus, the formation process of deviation accumulation is presented in the task sequence dimension, and the direction and magnitude of deviation accumulation are presented in the stope structure parameter dimension. This provides a consistent, reusable, and comparable input basis for the subsequent comprehensive construction of the surrounding rock stress path feature vector based on production plan data and parameter cumulative offset vector.
[0048] Based on the production plan data and parameter cumulative offset vector of the gold mining process, the characteristic vector of the surrounding rock stress path in the gold mining process is comprehensively constructed.
[0049] It should be explained that this process aims to unify and quantify the mining sequence and cumulative offset vector of parameters in the production plan data, so that the mining-adjacency relationship of the target stope in the historical mining sequence and the cumulative offset data of the target stope and spatially adjacent stopes can be extracted, encoded and spliced in the form of feature vectors, providing stable and reusable input for subsequent rolling optimization.
[0050] The method for comprehensively constructing the characteristic vector of the surrounding rock stress path in the gold mining process is as follows: The production plan data and cumulative offset vector of parameters of the target mining area in the gold mining process are collected. The mining sequence sequence in the production plan data is analyzed to obtain the historical mining sequence of the target mining area and the list of spatial adjacency relationships of the target mining area.
[0051] The mining sequence sequence centered on the corresponding target stope is extracted from the production plan data. The cumulative offset data of the target stope and its spatially adjacent stopes are extracted from the cumulative offset vector. Through specific coding and combination rules, a feature vector is generated that can quantitatively characterize the load path changes experienced by the surrounding rock during historical mining activities. The specific construction method is as follows: Locate the mining operation task number corresponding to the target mining site from the production plan data, and extract the position number of the mining operation task number in the mining sequence.
[0052] In the mining sequence, select mining operations that have been completed before the position number, with a quantity not exceeding a preset number of N, to form a historical mining sequence of finite length, and encode the spatial adjacency relationship of the historical mining sequence.
[0053] For each mined stop in the historical mining sequence, the spatial relationship between the mined stop and the target stop in the 3D orebody model is determined: if the two stops share a boundary or the distance between them is less than a preset distance threshold, they are considered spatially adjacent, and the orientation of the adjacent surface is recorded; if the two stops do not share a boundary and the distance between them is not less than the preset distance threshold, they are not considered spatially adjacent. Based on the mining order and spatial adjacency relationship of the historical mining sequence, a mining-adjacency relationship list is generated, containing the historical mined stop number, mining order, and spatial adjacency relationship between the historical mined stop and the target stop. This extracts the historical mining sequence and the spatial adjacency relationship list of the target stop in the gold mine mining process.
[0054] The cumulative offset vector of the parameters is processed by arranging the cumulative offset vector of the parameters of the target stope and the cumulative offset vector of the parameters of all directly adjacent stops in the target stope according to the spatial orientation order of the directly adjacent stops and the target stope, so as to obtain the cumulative offset data of the target stope.
[0055] Based on the mining sequence and spatial topology, all mining areas directly adjacent to the target mining area are identified, regardless of whether these mining areas have been completed. The cumulative offset vectors of the target mining area and its directly adjacent mining areas are extracted from the cumulative offset vector set and denoted as the cumulative offset data of the target mining area and all its directly adjacent mining areas.
[0056] Based on the historical mining sequence, spatial adjacency list, and cumulative offset data, the target mining area's surrounding rock stress path feature vector is obtained by vectorizing and assembling according to preset encoding and combination rules.
[0057] The process of constructing the characteristic vector of the surrounding rock stress path involves structurally encoding the historical mining sequence and mining-adjacency relationship list represented by the mining sequence sequence in the production plan data, and splicing the long-term cumulative deviation of the target stope and its directly adjacent spatial stopes represented by the parameter cumulative offset vector according to the spatial orientation order. This allows the change of the surrounding rock stress path to have a computable expression with both temporal order constraints and spatial adjacency constraints. In this way, the discrete mining sequence, adjacent surface exposure sequence and long-term structural parameter offset are uniformly mapped to a fixed-dimensional numerical sequence. This reduces the sensitivity of the rolling optimization time to changes in the original data form and the number of stopes, improves the input consistency and reusability between different production cycles, and provides a more stable and comparable characteristic input basis for the subsequent safety margin decay prediction model to infer the correlation of the surrounding rock stress path change trend.
[0058] By inputting the characteristic vector of the surrounding rock stress path during the gold mining process into the safety margin decay prediction model, the decay amount of the safety reserve coefficient and the risk value of the instability threshold crossing during the gold mining process are obtained.
[0059] The safety margin decay prediction model, based on the correlation information of mining sequence represented by the characteristic vector of the surrounding rock stress path and the long-term cumulative deviation represented by the cumulative parameter offset vector, infers the trend of surrounding rock stress path changes and outputs the decay of the safety reserve coefficient and the risk value of instability threshold crossing during the gold mine mining process. It should be explained that this process aims to unify and quantify the correlation information of mining sequence represented by the characteristic vector of the surrounding rock stress path and the long-term cumulative deviation represented by the cumulative parameter offset vector, enabling the correlation inference of the trend of surrounding rock stress path changes and outputting the decay of the safety reserve coefficient and the risk value of instability threshold crossing, providing stable input for subsequent rolling optimization. The decay of the safety reserve coefficient characterizes the magnitude of the decay of the safety reserve coefficient relative to the working conditions at the rolling optimization time within the mining cycle, while the risk value of instability threshold crossing characterizes the risk intensity of the surrounding rock stress path crossing the instability threshold within the mining cycle; the risk value of instability threshold crossing increases with increasing risk intensity.
[0060] The training method for the safety margin decay prediction model is as follows: Construct a safety margin decay training dataset, and use this dataset for supervised training to obtain a safety margin decay prediction model, specifically: First, a safety margin decay training dataset is constructed for model training. Then, a gradient boosting decision tree is selected as the core ensemble tree model architecture for supervised training and evaluation until the model performance meets the preset requirements.
[0061] A safety margin decay training dataset is constructed for model training. The construction of the safety margin decay training dataset is completed by collecting and associating multi-source data in the historical production process. For each historical mining area that has been completed, the feature vector of the surrounding rock stress path constructed at the last rolling optimization moment before the start of mining is extracted from the archived production plan data and process execution data. This vector is used as the input feature of a training data. Then, two target labels are labeled for this training data.
[0062] The first target label is the safety reserve coefficient decay label. The safety reserve coefficient decay is obtained through the following process: Before the start of mining in the stope, a calibrated rock mechanics numerical simulation software is used. Using the mining sequence and the cumulative parameter offset vector defined at that moment as boundary conditions, the planned start time, planned end time, planned output, and benchmark stope structural parameters are input into the software to analyze and obtain the theoretical safety reserve coefficient at the end of the complete mining cycle. After the actual mining in the stope ends, using the same mining sequence and the cumulative parameter offset vector defined as boundary conditions, the actual start time, actual end time, actual output, and stope structural parameters are input into the software to obtain the actual safety reserve coefficient. The difference between the theoretical and actual safety reserve coefficients is the safety reserve coefficient decay label corresponding to this training data point.
[0063] The second target label is the instability threshold crossing risk value label, which is a binary label, taking a value of 0 or 1. When a rock mass instability event occurs during or after mining in the stope, the instability threshold crossing risk value is recorded as 1; when no rock mass instability event meeting the preset definition occurs during or after mining in the stope, the instability threshold crossing risk value is recorded as 0. Rock mass instability events include, but are not limited to, roof collapse areas exceeding an area threshold or monitored rockburst energy exceeding an energy threshold. The input features are encapsulated with the two target labels to form a complete training sample; this process is repeated for multiple historical stopes to construct a safety margin decay training dataset containing multiple samples.
[0064] A gradient boosting decision tree is selected as the core ensemble tree model architecture to construct a safety margin decay prediction model. The safety margin decay prediction model is a multi-task learning framework. The input is the feature vector of the surrounding rock stress path of a single historical mining area, which is a fixed-dimensional numerical array. The safety margin decay prediction model has two parallel output branches: the first regression output branch predicts the continuous decay of the safety reserve coefficient; the second classification output branch predicts the discrete instability threshold crossing risk value. Both branches share the low-level features extracted from the multi-layer decision tree ensemble. The safety margin decay training dataset is divided into training and test sets, and the training set is used for supervised training of the safety margin decay prediction model. The training process adopts the iterative training rules of gradient boosting decision trees; the regression output branch uses the mean squared error loss function to measure the difference between the predicted value of the safety reserve coefficient decay and the true label; the classification output branch uses the cross-entropy loss function to measure the difference between the predicted value of the instability threshold crossing risk value and the true label. The loss functions of the two branches are weighted and summed with a preset weight ratio to form the total loss. The training process updates the multi-layer decision tree set according to the iterative training rules, so that the total loss function decreases monotonically. The performance of the model is continuously monitored using the test set, and an early stopping method is used to prevent the model from overfitting. The early stopping method's judgment rules include: when the total loss on the test set no longer decreases after the number of consecutive iterations reaches the preset threshold for stopping iterations, the iterative training is stopped, resulting in a safety margin decay prediction model that can be used for production environment prediction.
[0065] By using the characteristic vector of the surrounding rock stress path as a unified input, the mining sequence correlation information in the production plan data and the long-term cumulative deviation represented by the parameter cumulative offset vector are consistently encoded. This enables the safety margin decay prediction model to synchronously output the decay of the safety reserve coefficient and the risk value of instability threshold crossing during the target stope's mining cycle at the rolling optimization time. The decay of the safety reserve coefficient is calculated by comparing the planned start time, planned end time, planned output and benchmark stope structure parameters with the actual start time, actual end time, actual output and stope structure parameters under the same boundary conditions, ensuring consistent labeling and reproducibility. The risk value of instability threshold crossing is formed by labeling rock mass instability events to create a supervised risk discrimination target. This allows the prediction results to be directly used for the subsequent generation of dynamic safety constraint sets and the optimization configuration of gold mining process parameters, improving the stability of rolling optimization and the effectiveness of safety constraints.
[0066] Based on the decrease in the safety reserve coefficient and the risk value of the instability threshold in the gold mining process, a dynamic safety constraint set for the gold mining process is obtained, and the gold mining process parameters are optimized according to the dynamic safety constraint set.
[0067] The method for processing the dynamic safety constraint set obtained in the gold mine mining process is as follows: The generation of the dynamic safety constraint set is a process of dynamically calculating and combining a series of restrictive rules based on the output of the safety margin decay prediction model. The specific generation method is as follows: Analyzing the decay of the safety reserve coefficient yields the dynamic safety reserve coefficient constraint value for the gold mine mining process. Specifically, this involves reading the safety reserve coefficient decay output from the regression branch of the safety margin decay prediction model for the target stope. A baseline lower limit for the safety reserve coefficient is set, representing the minimum allowable safety factor for the current mining method and rock mass conditions. The dynamic safety reserve coefficient constraint value is the baseline lower limit plus the safety reserve coefficient decay. This dynamic constraint value indicates that in subsequent optimizations, under any process parameter configuration, the safety reserve coefficient obtained from numerical simulation calculations for the target stope must be greater than or equal to the dynamic safety reserve coefficient lower limit constraint.
[0068] Analyzing the risk value of instability threshold crossing yields constraints on the adjustment range of stope structure parameters during gold mine mining. Specifically, the risk value of instability threshold crossing for the target stope, output by the classification branch of the safety margin decay prediction model, is read. When the risk value of instability threshold crossing is 1, a strengthened constraint mechanism is triggered. This mechanism includes: multiplying the lower limit constraint of the dynamic safety reserve coefficient by a preset strengthening coefficient to obtain a more stringent dynamic safety reserve coefficient constraint value; and matching the dynamic safety reserve coefficient constraint value to obtain the adjustment range constraint of stope structure parameters. This constraint stipulates that the adjustment range of the target stope's stope structure parameters relative to the current actual value in the next production cycle must not exceed a preset maximum allowable adjustment percentage. It also stipulates that the pillar width can only increase, and the increase cannot exceed a preset maximum allowable increase percentage; and the stope span can only decrease, and the decrease cannot exceed a preset maximum allowable decrease percentage. When the risk value of instability threshold crossing is 0, the adjustment range constraint of stope structure parameters adopts a preset default allowable adjustment percentage range.
[0069] The adjustment range constraint of the stope structure parameter is obtained by matching the dynamic safety reserve coefficient constraint value. The specific process is as follows: the dynamic safety reserve coefficient constraint value is matched with the stope structure parameter adjustment range constraint corresponding to each dynamic safety reserve coefficient constraint value interval stored in the process optimization database. The adjustment range constraint of the stope structure parameter corresponding to the interval where the dynamic safety reserve coefficient constraint value is located is statistically recorded as the stope structure parameter adjustment range constraint of the target stope.
[0070] An analysis of the instability threshold crossing risk value and its position in the mining sequence yields the mining sequence adjacency constraints for the gold mine mining process. Specifically, the generation of the mining sequence adjacency constraints considers both the instability threshold crossing risk value of the target stope and its position in the mining sequence. When the instability threshold crossing risk value is 1, the mining sequence adjacency constraint stipulates that the start time of the target stope's mining operation in the optimized new mining sequence must be later than the end time of all its directly spatially adjacent stops that are already under mining or planned for mining, with a preset fixed time buffer period reserved. When the instability threshold crossing risk value is 0, the mining sequence adjacency constraint is not activated. The dynamic safety reserve coefficient constraint value, the stope structure parameter adjustment range constraint, and the mining sequence adjacency constraint are combined to form a dynamic safety constraint set for the target stope in the current gold mine mining process.
[0071] The method for optimizing gold mining process parameters based on dynamic safety constraint sets is as follows: Collect a set of dynamic safety constraints, analyze the adjustment range constraints of the stope structure parameters in the set of dynamic safety constraints, and obtain the updated values of the stope structure parameters that meet the constraints.
[0072] The adjacency constraints of the mining sequence in the dynamic safety constraint set are analyzed to obtain the start time of the mining operation that meets the constraints.
[0073] The updated values of the mining site structure parameters are combined with the start time of the mining operation to obtain the optimized configuration data of the gold mining process parameters, and then the optimized configuration is carried out based on the optimized configuration data of the gold mining process parameters.
[0074] This configuration method directly relies on a dynamic safety constraint set. Through item-by-item matching and sequential adjustment, it deterministically modifies and arranges the stope structure parameters and mining sequence of the target stope, ultimately outputting a process parameter configuration strategy that satisfies the dynamic safety constraint set. This strategy performs matching configuration based on the dynamic safety constraint set and updates it according to the dynamic safety reserve coefficient constraint value, the stope structure parameter adjustment range constraint, and the mining sequence adjacency constraint.
[0075] The specific configuration method is as follows: Read the dynamic safety constraint set for the target stope, and clarify the dynamic safety reserve coefficient constraint value, the stope structure parameter adjustment range constraint, and the mining sequence adjacency constraint.
[0076] The process involves adjusting the stope structure parameters to obtain the current actual stope structure parameter values, including pillar width, stope span, segment height, mining step distance, and cutting ratio. Based on the allowable adjustment directions, preset maximum allowable adjustment percentages, and default allowable adjustment percentage ranges specified in the stope structure parameter adjustment range constraints, the allowable value range for each parameter is obtained. The current actual stope structure parameter value is compared with this range. If the current actual stope structure parameter value is within the range, it remains unchanged; otherwise, it is updated according to the range boundary to form an updated stope structure parameter value that conforms to the constraints.
[0077] When the current actual stope structure parameter value is not within the value range, the process of updating it according to the value range boundary is as follows: Based on the stope structure parameter adjustment range constraint in the dynamic safety constraint set, the value ranges of pillar width, stope span, segment height, mining step distance and mining-cut ratio are determined respectively, and the current actual stope structure parameter value of the target stope is compared with the corresponding value range item by item; when a stope structure parameter value falls outside the value range, the stope structure parameter value is updated to the corresponding value range boundary value, thereby obtaining the stope structure parameter update value that meets the constraints.
[0078] When the current actual stope structure parameter values are not within the range, updates are performed according to the range boundaries. The specific process involves reading the allowable adjustment directions, preset maximum allowable adjustment percentages, and preset default allowable adjustment percentage ranges specified in the stope structure parameter adjustment range constraints for pillar width, stope span, segment height, retreat step distance, and mining-cut ratio. It also reads the target values for pillar width, stope span, segment height, retreat step distance, and mining-cut ratio from the benchmark stope structure parameters. The range boundaries are determined based on the allowable adjustment directions, which include allowable increase only, allowable decrease only, and allowable bidirectional adjustment. The preset maximum allowable adjustment percentage and preset default allowable adjustment percentage are also considered. The percentage range is used to determine the allowable offset of the value interval boundary. The allowable offset is calculated by multiplying the target value of the benchmark stope structure parameter by the preset maximum allowable adjustment percentage, or by multiplying the target value of the benchmark stope structure parameter by the product interval corresponding to the preset default allowable adjustment percentage range. When the stope structure parameter adjustment range constraint is configured to use the preset maximum allowable adjustment percentage, the value interval boundary is determined by the allowable offset corresponding to the preset maximum allowable adjustment percentage. When the stope structure parameter adjustment range constraint is configured to use the preset default allowable adjustment percentage range, the value interval boundary is determined by the allowable offset corresponding to the preset default allowable adjustment percentage range.
[0079] After determining the boundary of the value range, a step-by-step comparison is performed on the actual current value of the target stope's structural parameters. The comparison rules are as follows: when the actual value of the pillar width of the target stope is less than the lower boundary of the pillar width value range, the actual value of the pillar width is updated to the lower boundary of the pillar width value range; when the actual value of the pillar width of the target stope is greater than the upper boundary of the pillar width value range, the actual value of the pillar width is updated to the upper boundary of the pillar width value range; and when the actual value of the pillar width of the target stope is within the pillar width value range, it remains unchanged. The comparison and update rules for stope span, segment height, mining step distance, and mining-cutting ratio are consistent with those for pillar width. Lower boundary clamping and upper boundary clamping are performed on the stope span value range, segment height value range, mining step distance value range, and mining-cutting ratio value range, respectively, so that the update result of each stope structural parameter is within the corresponding value range.
[0080] The updated pillar width, stope span, segment height, mining step distance, and mining-cutting ratio are combined to form updated stope structure parameters that meet the constraints. These updated stope structure parameters are then written into the gold mining process parameter optimization configuration data for subsequent optimization of gold mining process parameters based on the dynamic safety constraint set.
[0081] The mining sequence arrangement is executed, the mining sequence adjacency constraints are read and their activation status is determined. When the mining sequence adjacency constraints are activated, the planned mining status and operation end time of all directly adjacent mining areas of the target mining area are obtained. The operation end time of the mining areas that are already in mining or planned for mining is determined, and the start time of the mining operation of the target mining area is set later than the operation end time with a fixed time buffer period. When the mining sequence adjacency constraints are not activated, the start time of the mining operation of the target mining area remains unchanged in the original production plan sequence.
[0082] The updated values of the mining site structure parameters are combined with the start time of the mining operation to obtain the optimized configuration data of the gold mining process parameters, and then the optimized configuration is carried out based on the optimized configuration data of the gold mining process parameters.
[0083] By transforming the safety reserve coefficient decay and instability threshold crossing risk value output by the safety margin decay prediction model into a dynamic safety constraint set, the target stope can obtain a dynamic safety reserve coefficient constraint value that is updated in real time with changes in working conditions during rolling optimization. The dynamic safety reserve coefficient constraint value is combined with the stope structure parameter adjustment range constraint and the mining sequence adjacency constraint to form deterministic constraints on the adjustment direction and range of stope structure parameters such as pillar width and stope span, as well as the start time of mining operations. Thus, under the premise that the safety reserve coefficient obtained by numerical simulation calculation of the target stope is greater than or equal to the dynamic safety reserve coefficient constraint value, the executable configuration of gold mining process parameters and the safe arrangement of mining sequence can be realized. This reduces the risk intensity of the surrounding rock stress path crossing the instability threshold during the mining cycle and improves the consistency of safety constraints and the stability of process parameter configuration under the conditions of production plan adjustment and on-site execution deviation.
[0084] A second aspect of the present invention provides a machine learning-based gold mine mining process optimization system, comprising: a differential cumulative analysis module, used to collect production plan data and process execution data of the gold mine mining process, analyze to obtain a process execution dataset of the gold mine mining process, perform differential cumulative analysis on the process execution dataset, and obtain a cumulative offset vector of parameters of the gold mine mining process.
[0085] The feature vector construction module is used to comprehensively construct the feature vector of the surrounding rock stress path in the gold mining process based on the production plan data and parameter cumulative offset vector of the gold mining process.
[0086] The feature vector analysis module is used to input the feature vector of the surrounding rock stress path in the gold mining process into the safety margin decay prediction model to obtain the decay of the safety reserve coefficient and the risk value of the instability threshold during the gold mining process.
[0087] The optimization configuration module is used to process the dynamic safety constraint set of the gold mining process based on the decrease of the safety reserve coefficient and the risk value of the instability threshold crossing during the gold mining process, and to optimize the configuration of the gold mining process parameters according to the dynamic safety constraint set.
[0088] It should be noted that a machine learning-based gold mine mining process optimization method and system also includes a process optimization database, which stores benchmark stope structure parameters obtained by analyzing historical data, preset quantity N, preset distance threshold, predefined order, lower limit of benchmark safety reserve coefficient, preset reinforcement coefficient, preset maximum allowable adjustment percentage, preset maximum allowable increase percentage, preset maximum allowable decrease percentage, preset default allowable adjustment percentage range, fixed time buffer period, and stope structure parameter adjustment range constraints corresponding to each dynamic safety reserve coefficient constraint value interval.
[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0090] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for optimizing gold mine mining processes based on machine learning, characterized in that, include: Collect production planning data and process execution data of gold mining process, analyze to obtain process execution dataset of gold mining process, perform differential cumulative analysis on process execution dataset to obtain parameter cumulative offset vector of gold mining process; Based on the production plan data and parameter cumulative offset vector of the gold mine mining process, the characteristic vector of the surrounding rock stress path in the gold mine mining process is comprehensively constructed. By inputting the characteristic vector of the surrounding rock stress path in the gold mining process into the safety margin decay prediction model, the decay of the safety reserve coefficient and the risk value of the instability threshold crossing in the gold mining process are obtained. Based on the decrease in the safety reserve coefficient and the risk value of the instability threshold in the gold mining process, a dynamic safety constraint set for the gold mining process is obtained, and the gold mining process parameters are optimized according to the dynamic safety constraint set.
2. The method for optimizing gold mine mining processes based on machine learning according to claim 1, characterized in that, The method for collecting production planning data and process execution data during the gold mining process is as follows: Extract production planning data and process execution data from the gold mining process; The production planning data for the gold mine mining process includes the mining sequence, mining task number, planned start time, planned end time, planned output and planned equipment information. The mining sequence is encoded as a mining sequence sequence with the mining sequence number increasing. The mining sequence sequence is used to represent the order of operations and the dependencies between tasks at the planning level, and provides a unified index for subsequent production cycle division and historical cumulative sequence. The process execution data for gold mining includes the mining operation task number, actual start time, actual end time, actual output, and stope structure parameters. The stope structure parameters include pillar width, stope span, segment height, mining step distance, and mining-cutting ratio.
3. The method for optimizing gold mine mining processes based on machine learning according to claim 2, characterized in that, The method for obtaining the process execution dataset of the gold mine mining process is as follows: After obtaining production planning data and process execution data, a task-level association relationship between the production planning data and process execution data is established based on the mining operation task number. The association between the production planning data and process execution data is then verified to obtain the process execution dataset of the gold mine mining process.
4. The method for optimizing gold mine mining processes based on machine learning according to claim 3, characterized in that, The method for performing differential cumulative analysis on the process execution dataset is as follows: Based on the pre-set benchmark stope structure parameters and combined with the process execution dataset, the cumulative parameter offset vector of the gold mine mining process is obtained through comprehensive analysis. The cumulative parameter offset vector includes the cumulative offset vector of stope structure parameters and the mining operation task number and preset production cycle identifier corresponding to the cumulative offset vector of stope structure parameters.
5. The method for optimizing gold mine mining processes based on machine learning according to claim 1, characterized in that, The method for comprehensively constructing the characteristic vector of the surrounding rock stress path during the gold mine mining process is as follows: Collect production plan data and cumulative parameter offset vectors of target mining areas in the gold mining process, analyze the mining sequence sequence in the production plan data, and obtain the historical mining sequence of target mining areas and the list of spatial adjacency relationships of target mining areas in the gold mining process. The cumulative offset vector of the parameters is processed to obtain the cumulative offset data of the target mining area; Based on the historical mining sequence, spatial adjacency list, and cumulative offset data, the target mining area's surrounding rock stress path feature vector is obtained by vectorizing and assembling according to preset encoding and combination rules.
6. The method for optimizing gold mine mining processes based on machine learning according to claim 1, characterized in that, The training method for the safety margin decay prediction model is as follows: Construct a safety margin decay training dataset, and use the safety margin decay training dataset for supervised training to obtain a safety margin decay prediction model. The method for constructing the safety margin decay training dataset is as follows: The characteristic vectors of the surrounding rock stress path of the target mining area in the historical production process are collected as input features of the safety margin decay prediction model. The input features are labeled with target labels to form a complete training sample. All training samples are summarized to construct a safety margin decay training dataset. The target labels include the safety reserve coefficient decay amount label and the instability threshold crossing risk value label.
7. The method for optimizing gold mine mining processes based on machine learning according to claim 1, characterized in that, The method for obtaining the dynamic safety constraint set of the gold mining process is as follows: The decay of the safety reserve coefficient and the risk value of the instability threshold during the gold mining process are collected. The decay of the safety reserve coefficient is analyzed to obtain the dynamic safety reserve coefficient constraint value during the gold mining process. By analyzing the risk value of the instability threshold crossing, the adjustment range constraint of the stope structure parameters in the gold mining process is obtained; By analyzing the risk value of instability threshold crossing and its position in the mining sequence, the adjacency constraint of mining sequence in the gold mine mining process is obtained. By combining the dynamic safety reserve coefficient constraint value, the adjustment range constraint of the mining site structure parameter, and the adjacent constraint of the mining sequence, a dynamic safety constraint set for the gold mine mining process is obtained.
8. The method for optimizing gold mine mining processes based on machine learning according to claim 1, characterized in that, The method for optimizing the configuration of gold mining process parameters based on a dynamic safety constraint set is as follows: Collect a set of dynamic safety constraints, analyze the adjustment range constraints of the stope structure parameters in the set of dynamic safety constraints, and obtain updated values of the stope structure parameters that meet the constraints. The adjacency constraints of the mining sequence in the dynamic safety constraint set are analyzed to obtain the start time of the mining operation that meets the constraints. The updated values of the mining site structure parameters are combined with the start time of the mining operation to obtain the optimized configuration data of the gold mining process parameters, and then the optimized configuration is carried out based on the optimized configuration data of the gold mining process parameters.
9. A machine learning-based gold mine mining process optimization system, characterized in that, include: The differential cumulative analysis module is used to collect production plan data and process execution data of the gold mine mining process, analyze them to obtain the process execution dataset of the gold mine mining process, perform differential cumulative analysis on the process execution dataset, and obtain the cumulative offset vector of parameters of the gold mine mining process. The feature vector construction module is used to comprehensively construct the feature vector of the surrounding rock stress path in the gold mining process based on the production plan data and parameter cumulative offset vector of the gold mining process. The feature vector analysis module is used to input the feature vector of the surrounding rock stress path in the gold mining process into the safety margin decay prediction model to obtain the decay of the safety reserve coefficient and the risk value of the instability threshold crossing in the gold mining process. The optimization configuration module is used to process the dynamic safety constraint set of the gold mining process based on the decrease of the safety reserve coefficient and the risk value of the instability threshold crossing during the gold mining process, and to optimize the configuration of the gold mining process parameters according to the dynamic safety constraint set.