An automated mining method, device and medium based on digital mine technology

Abstract

The application discloses an automatic mining method and device based on digital mine technology and a medium, and relates to the technical field of digital mine, which comprises the following steps: collecting ore body data and mining equipment operation data, correcting the time base, constructing a unified time axis, generating aligned data streams, and transmitting the data streams to an edge computing node through wireless communication; based on the edge computing node, the aligned data streams are adaptively cleaned and multi-scale fused to construct a state field covering the ore body; the state field is input into a pre-trained continuous time deep spatio-temporal prediction model to output a deformation field in a future period; based on an execution deviation sequence and a default degree sequence, an incremental learning sample is constructed on the edge computing node and fed back to a cloud platform to update the pre-trained continuous time deep spatio-temporal prediction model. The application significantly improves the state perception accuracy of multi-source ore body data and equipment operation data by constructing a continuous, stable and unified reference coordinate ore body state field.

Description

Technical Field

[0001] This invention relates to the field of digital mining technology, and in particular to an automated mining method, equipment and medium based on digital mining technology. Background Technology

[0002] With the development of digital mines and intelligent mining technologies, underground sensor networks can continuously acquire data on ground pressure, surrounding rock deformation, microseismic events, and mining equipment operation. Edge computing combined with cloud platforms is gradually becoming the mainstream architecture. At the same time, deep learning, especially continuous-time deep spatiotemporal prediction models, is used to characterize the deformation evolution trend of ore bodies, providing predictive basis for determining safety limits in mining procedures.

[0003] However, existing technologies generally suffer from the problem of a lack of a unified time reference for multi-source ore body data and equipment operation data. There are time alignment errors and scale inconsistencies in the construction of the state field, resulting in insufficient accuracy in subsequent deformation prediction and safety assessment. At the same time, most prediction models are mainly trained offline, making it difficult to use the deviation and safety breach degree generated during the execution process as closed-loop feedback for incremental model updates, and thus unable to adaptively track changes in working conditions in the continuous time dimension. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an automated mining method based on digital mining technology to address the problem of insufficient dynamic fusion and real-time adaptive updating of prediction models for multi-source orebody and equipment data under a unified time benchmark.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an automated mining method based on digital mining technology, which includes: collecting ore body data and mining equipment operation data, performing time reference correction, constructing a unified time axis, generating an aligned data stream, and transmitting it to an edge computing node via wireless communication;

[0008] Based on edge computing nodes, adaptive cleaning and multi-scale fusion of aligned data streams are performed to construct a state field covering the ore body.

[0009] Input the state field into a pre-trained continuous-time deep spatiotemporal prediction model and output the deformation field for future time periods;

[0010] The preset set of safety limit rules is compiled into a differentiable logic monitor, which performs continuous-time default assessment on the deformable field in future time periods, outputs a default sequence, and generates a spatially executable domain.

[0011] By combining the spatial executable domain with the deformation field of future time periods, mining strategy adjustment instructions and equipment scheduling instructions are generated. During the execution of the instructions, the equipment operating status and ore body status are collected and aligned with the deformation field of future time periods on a unified time axis. The ore body deviation and equipment deviation are calculated, and the execution deviation sequence is output.

[0012] Based on the execution bias sequence and default degree sequence, incremental learning samples are constructed on edge computing nodes and fed back to the cloud platform to update the pre-trained continuous-time deep spatiotemporal prediction model.

[0013] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the specific steps for constructing a unified timeline and generating aligned data streams, and transmitting them to edge computing nodes via wireless communication, are as follows.

[0014] Collect ore body data and mining equipment operation data, record the original timestamps corresponding to the ore body data and mining equipment operation data, perform time base correction, and generate a unified timeline;

[0015] By aligning ore body data with mining equipment operation data using a unified timeline, an aligned data stream is generated and transmitted to edge computing nodes via wireless communication protocol encoding.

[0016] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the specific steps for constructing the state field covering the ore body are as follows:

[0017] At the edge computing node, adaptive noise removal is performed on the ore body data and mining equipment operation data in the aligned data stream to generate primary cleaned data.

[0018] Based on the timestamps of the aligned data streams, perform missing data imputation and anomaly detection on the primary cleaned data, and output calibrated multi-source data;

[0019] The calibrated multi-source data are fused at multiple scales according to spatial location and temporal order to construct a state field covering the spatial region of the ore body.

[0020] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the specific steps for outputting the deformation field for future time periods are as follows:

[0021] The state field covering the ore body space is organized into a model input sequence in a continuous time format, input into a pre-trained continuous time depth spatiotemporal prediction model, and spatiotemporal extrapolation of the state field is performed to output an extrapolation data set.

[0022] Based on the inferred data set, the deformation trend and spatial change structure of the ore body in future time periods are read, and candidate deformation segments for future time periods are generated.

[0023] The candidate deformation segments for future time periods are organized according to a unified time axis and spatial grid, and the deformation field for future time periods is output.

[0024] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the steps of compiling a preset safety limit rule set into a differentiable logic monitor, performing continuous-time default assessment on the deformation field in future time periods, outputting a default sequence, and generating a spatially executable domain are as follows.

[0025] Based on the deformation field in the future time period, the deformation change of the ore body is calculated according to the adjacent time positions on the unified time axis, and a deformation sequence is constructed. A set of safety limit rules is set according to the mining regulations, and a logical reference is generated based on the set of safety limit rules. The preset safety limit rules are compiled into a differentiable logic monitor.

[0026] The deformed sequence is input into the differentiable logic monitor, and the default degree of future time periods is calculated in the differentiable logic monitor based on the logic reference. The default degree sequence of spatial domain judgment is output.

[0027] Based on the default degree sequence, determine the location range and non-executable location range that meet the preset safety limit rules within the spatial area of ​​the ore body, and generate preliminary executable area data of the spatial area.

[0028] Based on spatial connectivity and future deformation trends, the preliminary executable region data is processed to generate a spatial executable region.

[0029] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the specific steps for combining the spatially executable domain with the deformation field of future time periods to generate mining strategy adjustment instructions and equipment scheduling instructions are as follows.

[0030] Based on the spatial executable domain, the spatial range in which mining operations are allowed to be carried out in the future time period is determined, spatial constraints are generated, and continuous filtering of the spatial executable domain is performed on a unified time axis to output a set of candidate operation spatial locations;

[0031] Using spatial constraints and the set of candidate job locations as spatial screening conditions, the deformation field of future time periods is spatially screened on a unified time axis to construct a spatially constrained future deformation reference field.

[0032] By utilizing the future deformation reference field with spatial constraints, the consistency of the change magnitude and direction of the candidate job spatial location set in adjacent time locations is extracted and a change trend is generated. The change magnitude is then summarized within the spatial executable domain to generate a spatial executable domain risk benchmark.

[0033] Based on the consistency of change magnitude, change direction, change trend, and spatial executable domain risk benchmark, risk classification and priority ranking are performed to determine the corresponding operation mode for spatial location and generate mining strategy adjustment instructions.

[0034] On a unified timeline, mining strategy adjustment instructions and spatial constraints are combined at each time position, and equipment scheduling instructions are generated by prioritizing them.

[0035] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the step of collecting the equipment operating status and ore body status during instruction execution, aligning them with the deformation field of future time periods on a unified time axis, and obtaining an execution deviation sequence, includes the following specific steps.

[0036] During the execution process, the equipment operating status and ore body status are collected and the corresponding timestamps are recorded to generate an aligned equipment operating status sequence and ore body status sequence.

[0037] Based on a unified timeline, the equipment operation status sequence is aligned with the ore body status sequence and the deformation field of future time periods. Difference metrics are calculated based on spatial and temporal locations to generate an execution deviation sequence.

[0038] As a preferred embodiment of the automated mining method based on digital mining technology described in this invention, the steps of constructing incremental learning samples at edge computing nodes based on execution deviation sequences and default rates, feeding these incremental learning samples back to the cloud platform, and updating the pre-trained continuous-time deep spatiotemporal prediction model are as follows:

[0039] Based on the execution deviation sequence, the time and spatial locations where the deviation exceeds the preset deviation judgment conditions are identified on a unified time axis, deviation candidate segments are generated, and deviation candidate segments are correlated with the default degree sequence of future time periods to construct joint deviation segments.

[0040] The state field data corresponding to the joint bias fragment and the bias candidate position set are integrated at the edge computing node to generate incremental learning input fragments, and then organized into an incremental learning sample set for updating the pre-trained continuous-time deep spatiotemporal prediction model according to the continuous-time format.

[0041] Based on edge computing nodes, the incremental learning samples are formatted and validated. The validated incremental learning samples are then fed back to the cloud platform via wireless communication to generate cloud-based incremental learning samples.

[0042] The pre-trained continuous-time deep spatiotemporal prediction model is updated on the cloud platform based on incremental learning samples in the cloud, and the updated continuous-time deep spatiotemporal prediction model is synchronized to the edge computing nodes.

[0043] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the automated mining method based on digital mining technology as described in the first aspect of the present invention.

[0044] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the automated mining method based on digital mining technology as described in the first aspect of the present invention.

[0045] The beneficial effects of this invention are as follows: By correcting and aligning data streams with a time reference, and performing adaptive cleaning and multi-scale fusion at the edge, a continuous, stable ore body state field with unified reference coordinates is constructed, significantly improving the state perception accuracy of multi-source ore body data and equipment operation data; by inputting the state field into a pre-trained continuous-time depth spatiotemporal prediction model, and combining it with a differentiable logic monitor to evaluate the continuous-time default degree of the future deformation field, a spatially executable domain is quantified, enabling risk identification to have coherence and fine-grained expression; by using the ore body deviation, equipment deviation, and default degree sequence obtained during instruction execution to construct incremental learning samples at the edge computing nodes, and feeding them back to the cloud to update the prediction model parameters, the strategy generation can adaptively adjust with changes in geological conditions and working conditions, taking into account the safety, reliability, and operational efficiency of the mining process as a whole. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an automated mining method based on digital mining technology.

[0048] Figure 2 This is a flowchart for data cleaning and multi-scale fusion.

[0049] Figure 3 This is a flowchart for predicting the deformation field.

[0050] Figure 4 Flowchart for default assessment and spatial executable domain generation Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0054] Reference Figures 1-4 This is one embodiment of the present invention, which provides an automated mining method based on digital mining technology, including the following steps:

[0055] S1: Collect ore body data and mining equipment operation data, perform time base correction, construct a unified time axis, generate an aligned data stream, and transmit it to the edge computing node via wireless communication;

[0056] S1.1: Collect ore body data and mining equipment operation data, record the original timestamps corresponding to the ore body data and mining equipment operation data, perform time base correction, and generate a unified timeline;

[0057] By monitoring the mine interior in real time, data on ore bodies are collected, including the underground space environment, changes in surrounding rock, and spatial deformation trends. The original timestamps attached to each ore body data are also recorded.

[0058] By monitoring mining equipment operation data in real time and recording the original timestamp for each data point, the mining equipment operation data includes equipment location, equipment operating status, and equipment load.

[0059] Time reference calibration is performed on the original timestamps of the ore body data and the original timestamps of the mining equipment operation data. The master clock of the edge computing node is used as the time reference source. The acquisition end uses the IEEE1588PTP precision clock synchronization method to synchronize the local clock of the acquisition end with the time reference source.

[0060] Specifically, the acquisition end obtains the time deviation of the local clock relative to the time reference source through IEEE 1588PTP time synchronization message interaction, and calibrates the local clock of the acquisition end.

[0061] Based on the calibrated local clock of the calibration acquisition terminal, the time difference between the original timestamp of the ore body data and the original timestamp of the mining equipment operation data relative to the time reference source is compared and corrected. The original timestamp is calibrated to a calibration timestamp on a unified time axis, and the calibrated timestamps of the ore body data and the calibrated timestamps of the mining equipment operation data are output.

[0062] Based on the calibrated ore body data timestamps and the calibrated mining equipment operation data timestamps, each timestamp is mapped to a unique time position on a unified time axis, thus constructing a unified time axis.

[0063] S1.2: Align the ore body data with the mining equipment operation data using a unified timeline, generate an aligned data stream, and transmit it to the edge computing node via wireless communication protocol encoding;

[0064] Based on continuous time positions in a unified time axis, the timestamps of the corrected ore body data are matched with the corresponding time positions in the unified time axis:

[0065] When a certain time location corresponds to multiple ore body data, the time location is subdivided into consecutive sub-time locations, and the ore body data is recorded in the corresponding sub-time locations according to the collection order.

[0066] When no ore body data is collected at a certain time point, an empty record is retained at that time point, and a missing marker is recorded to maintain a continuous time structure on the unified time axis.

[0067] Based on continuous time positions on a unified time axis, the timestamps of the calibrated mining equipment operation data are matched with the corresponding time positions on the unified time axis:

[0068] When a certain time location contains multiple mining equipment operation data, all mining equipment operation data will be recorded at this time location in the original acquisition order;

[0069] When no mining equipment operation data is collected at a certain time point, an empty record is retained at that time point, and a missing marker is recorded to maintain a continuous time structure on the unified time axis.

[0070] The aligned orebody data is integrated with the aligned mining equipment operation data at each time point to generate an aligned data stream arranged in chronological order.

[0071] Missing records in the aligned data stream are filled using the following methods: linear interpolation is used to fill empty records in the ore body data, forward hold is used to fill empty records in the mining equipment operation data, and missing records are retained.

[0072] The aligned data stream is encoded using a wireless communication protocol. The encoded content includes at least the frame sequence number, time position identifier, data length, and check field. Packet loss detection and retransmission mechanisms are configured, and the data is transmitted to the edge computing node in real time via 5G wireless communication.

[0073] S2: Based on edge computing nodes, adaptive cleaning and multi-scale fusion of aligned data streams are performed to construct a state field covering the ore body;

[0074] S2.1: At the edge computing node, adaptive noise removal is performed on the ore body data and mining equipment operation data in the aligned data stream to generate primary cleaned data;

[0075] The edge computing node reads the aligned data stream arranged along a unified time axis and acquires the aligned ore body data and aligned mining equipment operation data at each time point.

[0076] Adaptive noise removal is performed on the ore body data, and a fixed-length sliding time window is set based on a unified time axis (example: 3 to 5 time positions).

[0077] Within the window, calculate the variation range of ore body data at consecutive time positions. The variation range is the absolute value of the difference between the ore body data at the next time position and the ore body data at the previous time position. The unit of variation range is the same as the unit of ore body data.

[0078] When there are differences in the time intervals of a unified time axis, the magnitude of change is normalized according to the corresponding time intervals, and a reference magnitude of change index is output.

[0079] Determine the preset ore body change threshold based on the reference change range index:

[0080] When the numerical difference between the ore body data at a certain time location and the adjacent time locations exceeds the preset ore body change threshold, the ore body data at this time location is marked as a noise point.

[0081] By using linear interpolation of ore body data from the previous and subsequent consecutive time positions to replace noise points, the time series of ore body data is kept continuous, and the noise-removed ore body data is output.

[0082] The rate of change of device position is determined based on the difference between device positions in adjacent time locations and the corresponding time interval. The frequency of device action state switching is determined based on the number of inconsistencies in state codes between adjacent time locations. The rate of change of device load is determined based on the difference in load values ​​between adjacent time locations.

[0083] When the rate of change of equipment position at a certain time location exceeds the preset equipment speed threshold, or the rate of change of equipment load exceeds the preset load change threshold, or the frequency of equipment action state switching is higher than the preset state switching threshold, the mining equipment operation data corresponding to this time location is marked as jump noise.

[0084] Linear interpolation is performed using mining equipment operation data from the previous and subsequent consecutive time positions to enable the mining equipment operation data to continuously express the actual operating status of the equipment over time, and output the mining equipment operation data after noise removal.

[0085] The noise-removed ore body data and the noise-removed mining equipment operation data are recorded as primary cleaning data along a unified time axis.

[0086] It should be noted that the preset ore body change threshold is configured based on the statistical distribution of reference change amplitude formed by historical data within the sliding window (example: take the 90% to 95% quantile of the reference change amplitude distribution); during operation, the quantile value is re-statistically calculated based on the latest window data at the set update interval to smoothly correct the ore body change threshold.

[0087] The basic ranges for equipment speed thresholds, load change thresholds, and state switching thresholds are determined based on the rated parameters in the equipment manual, and the initial thresholds are determined through historical operating data.

[0088] For example, the equipment speed threshold can be configured based on the historical equipment speed distribution (example: take the sum of the mean and three standard deviations of the historical speed distribution); the load change threshold can be configured based on the upper limit of the rated load change (example: take the value in the range of 80% to 100% of the upper limit of the rated load change); the state switching threshold can be configured based on the statistical results of the number of times the equipment action state switches per unit time (example: take the 95th percentile of the number distribution).

[0089] S2.2: Perform missing data filling and anomaly detection on the primary cleaned data based on the timestamps of the aligned data stream, and output calibrated multi-source data;

[0090] Read each time point in the primary cleaning data and check for missing ore body data and missing mining equipment operation data at each time point.

[0091] If there is a missing ore body data or mining equipment operation data at a certain time position, and there is similar valid data in both the preceding and following consecutive time positions, the similar data in the preceding and following positions are used as interpolation endpoints, and linear interpolation is used to generate filler values.

[0092] When a missing location has valid data of the same type on only one side, the valid data value of the most recent time location is used to directly replace the data at the missing location.

[0093] Anomaly detection is performed on the ore body data in the primary cleaning data by comparing the numerical difference of the ore body data between consecutive time positions with a preset ore body change threshold, and combining the change trend of the ore body data within the sliding time window.

[0094] When the numerical difference at a certain time point exceeds the preset ore body change threshold and is inconsistent with the overall change direction within the sliding time window, the ore body data corresponding to this time point is marked as an outlier.

[0095] Based on the changes in ore body data in the preceding and following consecutive time positions of the outlier, the average of the nearest time positions is used to adjust the outlier.

[0096] Anomaly detection is performed on the mining equipment operation data in the primary cleaning data by comparing the rate of change of equipment position, the frequency of equipment action state switching, and the rate of change of equipment load between consecutive time locations.

[0097] When the rate of change of equipment position at a certain time location exceeds the preset equipment speed threshold, or the rate of change of equipment load exceeds the preset load change threshold, or the frequency of equipment action state switching is higher than the preset state switching threshold, the mining equipment operation data corresponding to this time location will be marked as an abnormal record.

[0098] Based on the changes in mining equipment operation data in the preceding and following consecutive time positions of the abnormal record, the abnormal record is smoothly adjusted according to the trend of the preceding time position.

[0099] The ore body data and mining equipment operation data that have completed missing filling and anomaly detection are recorded in the time position order of a unified time axis as calibrated multi-source data.

[0100] S2.3: The calibrated multi-source data are fused at multiple scales according to spatial location and temporal order to construct a state field covering the spatial region of the ore body;

[0101] Based on the spatial location identifiers attached to each ore body data and each mining equipment operation data, the calibrated multi-source data are classified according to spatial location, so that each spatial location has corresponding time series data arranged on a unified time axis.

[0102] Within each spatial location, a multi-timescale sliding window is constructed based on the records of calibrated multi-source data at continuous time locations. The mean of local changes is calculated within the short-timescale sliding window (e.g., 3-5 time locations), and the overall change trend is calculated within the long-timescale sliding window (e.g., 10-20 time locations). The local change information and the overall change structure are then weighted and synthesized to generate the timescale fusion result.

[0103] Based on the spatial adjacency relationships of calibrated multi-source data in different spatial locations, at each time location, according to the spatial grid adjacency relationship corresponding to the spatial location identifier, a set of adjacent spatial locations sharing the boundary with the target spatial location is selected to form an adjacent spatial location set, and the spatial distance between the target spatial location and the adjacent spatial locations is calculated.

[0104] The weights of the inverse distance weighted average are determined based on the spatial distance between the target spatial location and its adjacent spatial locations. The weights decrease as the spatial distance increases. The weights are then normalized within the set of adjacent spatial locations, and the normalized weights of the inverse distance weighted average are output.

[0105] The temporal scale fusion results of adjacent spatial locations are weighted by normalized inverse distance weighted average to output the spatial scale fusion results of the target spatial location in time.

[0106] Furthermore, when there are adjacent spatial locations with zero spatial distance, the temporal scale fusion result corresponding to the spatial location with zero spatial distance is directly used as the spatial scale fusion result.

[0107] The temporal-scale fusion results and spatial-scale fusion results are combined according to the temporal order and spatial location of a unified time axis to generate a state field that covers the spatial region of the ore body and changes continuously over time.

[0108] S3: Input the state field into the pre-trained continuous-time deep spatiotemporal prediction model and output the deformation field for future time periods;

[0109] S3.1: Organize the state field covering the ore body space into a model input sequence in a continuous time format, input it into the pre-trained continuous time depth spatiotemporal prediction model, perform spatiotemporal deduction of the state field, and output the deduction data set;

[0110] Based on the state field that covers the spatial region of the ore body and changes continuously over time, multiple time positions are recorded on a unified time axis. Each time position contains multiple multi-source state quantities corresponding to multiple spatial positions. The corresponding multi-source state quantities are normalized, and the normalized state field recording results are generated under the unified time axis and spatial position index.

[0111] The normalized state field record result is used as the state field function.

[0112] The state field function is expressed as:

[0113] ;

[0114] In the formula, To indicate spatial location With time and location Below, the state field numerical value, composed of multiple source state variables, is used to describe the comprehensive state of this spatial location at this temporal location. For spatial location, This refers to the time and location.

[0115] Based on the state field function, the state field records are read sequentially according to the time position on a unified time axis. For each time position, the state field function values ​​at each spatial position are read in turn.

[0116] Arrange all these state field function values ​​in a fixed order to generate state field frames corresponding to the time positions.

[0117] Connect the state field frames corresponding to all time positions on a unified time axis in chronological order to output the state field input sequence.

[0118] The state field input sequence, spatial location, and temporal location are input together into a pre-trained continuous-time deep spatiotemporal prediction model. The state field input sequence is represented based on the state field function, and the state field input sequence is mapped into deformation indicators in future time periods.

[0119] The mapping of the state field input sequence to deformation indicators in future time periods is expressed as:

[0120] ;

[0121] In the formula, In spatial location With time and location The deformation indication quantity below, This relates the mapping between the pre-trained continuous-time deep spatiotemporal prediction model and the state field input sequence, spatial location, and temporal location.

[0122] Select a future time period on a unified time axis, read the corresponding deformation indicator for each time position and each spatial position within the future time period, record all deformation indicators according to time position and spatial position, and output a set of projection data covering the future time period.

[0123] It should be noted that the pre-trained continuous-time deep spatiotemporal prediction model is constructed based on historical state field data covering the spatial region of the ore body.

[0124] During the pre-training phase, ore body data and mining equipment operation data from historical acquisition periods are used to complete time alignment, cleaning, missing data filling, and multi-scale fusion according to a unified timeline, generating a state field sequence for historical time periods and using it as model input.

[0125] Using the actual ore body deformation field recorded within a historical period as a supervision signal, a supervised learning training dataset of state field sequence - actual ore body deformation field is constructed, and time sampling and spatial sampling are performed to generate training samples, including the historical time window length (example value: 12 to 48 time positions), prediction lead (example value: 6 to 24 time positions), and prediction time window length (example value: 6 to 24 time positions).

[0126] Furthermore, the continuous-time deep spatiotemporal prediction model adopts a structure of spatial representation encoding-continuous-time extrapolation-deformation decoding. The number of spatial representation encoding layers (example values ​​are 2 to 4 layers) and the hidden representation dimension (example values ​​are 64 to 256) are determined by the number of spatial locations, the dimension of multi-source state variables, the length of the historical time window, and the length of the prediction time window.

[0127] The error between the predicted deformation field and the historical actual ore body deformation field is used as the training objective. The loss function adopts mean squared error and is superimposed with a parameter regularization term. The training batch size (example values ​​are: 8 to 64), learning rate (example values ​​are: 1×10^-4 to 5×10^-3), parameter regularization coefficient (example values ​​are: 1×10^-6 to 1×10^-3), number of training rounds (example values ​​are: 50 to 300), and number of rounds for early stopping judgment (example values ​​are: 10 to 30 rounds) are used. The gradient-based optimization method is used to iteratively update the parameters and solidify the convergence results, and the trained continuous time depth spatiotemporal prediction model is output.

[0128] S3.2: Based on the inferred data set, read the deformation trend and spatial change structure of the ore body in future time periods, and generate candidate deformation segments for future time periods;

[0129] Based on the inferred data set, the start and end time positions of future periods are determined on a unified time axis. Within the future period, the deformation indication values ​​of each spatial location at each time position are read to generate a time series of deformation indication values ​​covering the future period.

[0130] Based on the deformation indicator time series, at each spatial location, the starting time position of the future period is selected as the reference time position. The deformation indicator at each time position in the future period is compared with the deformation indicator at the reference time position, and the deformation increment time series is output.

[0131] The deformation increment time series is represented as follows:

[0132] ;

[0133] In the formula, In spatial location With time and location The deformation increment below, This represents the starting time position of a future time period.

[0134] Within each spatial location, a sliding time window mean smoothing process is performed on the deformation increment time series. The sliding time window covers multiple consecutive time locations (e.g., covering 3 to 5 consecutive time locations) to suppress false detections caused by single-point jitter.

[0135] Based on the smoothed deformation increment time series, the change structure of each time position in the future period is analyzed. When a time position meets any of the following conditions, the time position is recorded as a key time position of spatial focus:

[0136] When the deformation increment at a certain time position is greater than the deformation increment at its immediate and next-to-immediate time positions, it is determined to be a local maximum point.

[0137] Starting from the beginning of the future time period, and moving along the direction of the end of the future time period, when the deformation increment shows a monotonically increasing relationship at multiple consecutive time positions (each increment is greater than the previous increment), the time position at the end of the monotonic subsequence is recorded as the continuous growth point.

[0138] All local maxima and all persistent growth points are merged and recorded in chronological order as a set of key time locations to focus on.

[0139] In the spatial dimension, at the same time location, the deformation increments of all spatial locations are formed into a spatial sequence according to the spatial location index order.

[0140] When the deformation increments of several adjacent spatial locations are all greater than the deformation increments of the outer spatial locations adjacent to both ends of the set of adjacent spatial locations, these adjacent spatial locations are combined into a continuous set of spatial locations.

[0141] Based on the set of key time locations and the set of continuous spatial locations in the spatial dimension, candidate fragments for future time period deformation are constructed within the future time period.

[0142] The candidate segments for deformation in the future time period include: the corresponding time location, the corresponding set of spatial locations, and the deformation indicators and deformation increments at these time and spatial locations.

[0143] S3.3: Organize the candidate deformation segments for future time periods according to a unified time axis and spatial grid, and output the deformation field for future time periods.

[0144] Based on candidate fragments of future time periods, each time position within the future time period is enumerated on a unified time axis, and all existing spatial positions in the state field are enumerated at each time position to form an index set of the deformation field of the future time period.

[0145] For each combination of time and spatial location in the index set, retrieval is performed in the candidate fragments of future time periods:

[0146] When a combination of a certain time location and a certain spatial location appears in the time location set and spatial location set of a future time period deformation candidate segment, the deformation indicator recorded in the future time period deformation candidate segment is directly used.

[0147] When a combination of time and space location does not appear in the time and space location sets of any future deformation candidate segments, the deformation indicator corresponding to that combination of time and space location is read from the inference data set, and the deformation indicator is used as the deformation field value corresponding to the combination of time and space location.

[0148] After filling the deformation field values ​​of all time and spatial locations on a unified time axis and spatial grid, all spatial locations and their corresponding deformation field values ​​are arranged in spatial grid order at each time location to generate the spatial deformation distribution of the time location.

[0149] The spatial deformation distribution at each time position is arranged sequentially according to the time position order of a unified time axis, and the deformation field covering future time periods is output.

[0150] S4: Compile the preset set of safety limit rules into a differentiable logic monitor, perform continuous-time default assessment on the deformable field in future time periods, output the default sequence, and generate a spatially executable domain;

[0151] S4.1: Based on the deformation field of future time periods, calculate the deformation change of the ore body according to the adjacent time positions on a unified time axis, construct the deformation sequence, set a safety limit rule set according to the mining regulations, generate a logical reference based on the set safety limit rule set, and compile the preset safety limit rule set into a differentiable logic monitor.

[0152] The deformation fields based on future time periods are arranged at multiple time positions on a unified time axis. At each time position, the deformation field values ​​at various spatial locations within the ore body spatial region are recorded.

[0153] Select two adjacent time points on a unified time axis, calculate the change between adjacent time points for each spatial point, and output the change.

[0154] Specifically, for each spatial location, the change between adjacent time locations is calculated and expressed as:

[0155] ;

[0156] In the formula, For spatial location In time location With time and location The amount of deformation change between them For spatial location In time location The deformation field values ​​under the following conditions For spatial location In time location The deformation field values ​​under the following conditions To unify a specific time position on the timeline arranged chronologically. To unify the time position in the timeline The next time position that is adjacent and follows in chronological order. To standardize the sequence number of time positions on the timeline.

[0157] The deformation changes are calculated repeatedly for all adjacent time locations. All deformation changes are arranged in chronological order at each spatial location to generate a deformation sequence covering future time periods.

[0158] Select safety limit clauses related to ore body deformation from the mining regulations text, establish a correspondence between the upper and lower deformation limits involved in the clauses and the spatial location, and output a set of safety limits oriented towards spatial location.

[0159] For spatial locations with safety limit constraints, the entry and exit times of the operation phase are identified based on the equipment action state sequence and timestamp sequence in the mining equipment operation data.

[0160] The entry and exit times are mapped to the start and end times on a unified time axis, respectively, to define the applicable time range of the safety limit clauses in spatial location.

[0161] Based on the set of safety limits, allowable deformation range boundaries are given for each combination of spatial and temporal locations. The upper and lower deformation limits are recorded together with the corresponding time range to generate a logical reference.

[0162] It should be noted that when there is a clause in the mining regulations that the surrounding rock displacement of a certain mining zone does not exceed the safe deformation limit during the prediction period, the mining zone can be divided into several spatial locations, and the upper limit of deformation and the applicable time range corresponding to each spatial location can be recorded in the logical reference.

[0163] S4.2: Input the deformed sequence into the differentiable logic monitor, calculate the default degree of future time periods in the differentiable logic monitor based on the logic reference, and output the default degree sequence of spatial domain judgment;

[0164] The deformation sequence and the deformation field of the future time period are used together as the input signal of the differentiable logic monitor.

[0165] The input signal is fed into a differentiable logic monitor, which compares the deformation field value and deformation sequence value with the safety limit range in the logic reference (example value range: –2mm-15mm) at each combination of spatial and temporal positions on a unified time axis, and outputs the corresponding default degree.

[0166] For spatial locations that simultaneously have upper and lower deformation limits, calculate the deviation of the deformation field value from the upper and lower limits at each time location within a future time period, and output the single-point default degree.

[0167] The single point of default is expressed as:

[0168] ;

[0169] ;

[0170] ;

[0171] In the formula, For spatial location In time location The degree of default is relative to the upper limit of deformation. When the deformation field value does not exceed the upper limit of deformation, the degree of default is zero. For spatial location In time location The degree of failure relative to the lower limit of deformation, when the deformation field value is lower than the lower limit of deformation. A value greater than zero indicates a default degree greater than zero, and the deformation field value is not lower than the lower limit of deformation. If the value is less than or equal to zero, the default rate is zero. For spatial location In time location Single point of default Spatial location recorded in logical reference The corresponding upper limit of deformation, Spatial location recorded in logical reference The corresponding lower limit value of deformation.

[0172] If a logical reference provides only an upper limit value for deformation at a certain spatial location but not a lower limit value, then the value at that spatial location is taken as... ;

[0173] If the logical reference only provides the lower bound of deformation and does not include the upper bound of deformation, then the spatial location is taken as... .

[0174] For each time location within the future period, calculate the single-point default score across all spatial locations, use the safety limit range as the normalization scale to normalize the single-point default score, and output the normalized single-point default score.

[0175] For each time location within a future period, calculate the normalized single-point default rate across all spatial locations.

[0176] The normalized single-point default scores corresponding to all spatial locations at the same time position are arranged in the spatial location index order of the spatial grid, and the spatial default score distribution corresponding to the time position is output.

[0177] Arrange the spatial default degree distributions corresponding to all time locations within the future period in chronological order to generate a default degree sequence covering the future period.

[0178] It should be noted that a certain spatial location satisfies the following conditions at all time locations within a future time period: When, the spatial location meets the safety limit rules throughout the entire future time period;

[0179] For scenarios where only an upper limit constraint on deformation exists, the lower limit value of deformation can be... Set to zero and ignore. Regarding the impact of deformation lower bound constraints, for scenarios with such constraints, the corresponding deformation lower bound values ​​can be recorded in the logical reference and... Perform measurement.

[0180] If a certain time location exists This indicates that the spatial location deviates from the safety limit at the corresponding time location.

[0181] A better approach is to use a differentiable logic monitor to generate default score values. The calculation is implemented using a smooth approximation method, which makes the output failure degree value change continuously with the deformation field value.

[0182] S4.3: Based on the default degree sequence, determine the location range and non-executable location range that meet the preset safety limit rules within the ore body spatial area, and generate preliminary executable area data of the spatial area;

[0183] Based on the default degree sequence, the single-point default degree of each spatial location is statistically analyzed in the future time period, and the default degree value of each spatial location is checked on a unified time axis to see if there are any time locations with a default degree value greater than zero.

[0184] For all time locations within the future time period, satisfying The spatiotemporal location is recorded as a spatial location that conforms to the safety limit rules;

[0185] For at least one time location in the future period, there exists The spatial location is recorded as a spatial location that does not comply with the safety limit rules.

[0186] After completing the compliance status marking of all spatial locations, the set of spatial locations that comply with the safety limit rules is recorded as the initial set of executable locations within the spatial region.

[0187] The set of spatial locations that do not conform to the safety limit rules is recorded as the set of non-executable locations within the spatial region.

[0188] Based on the initial set of executable and non-executable locations, preliminary executable area data of the spatial region is generated within the ore body spatial region.

[0189] S4.4: Based on spatial connectivity and future deformation trends, perform connectivity domain organization on the preliminary executable region data to generate a spatial executable domain;

[0190] Based on the preliminary executable region data, the spatial location arrangement method used in the spatial dimension determines the adjacency relationship between spatial locations (adjacency relationship is defined as two spatial locations sharing an edge in the spatial grid).

[0191] In the future, spatial connectivity analysis will be performed on the initial set of executable locations.

[0192] Initial executable locations that are connected to each other through adjacency are grouped into the same connected region.

[0193] Initially executable locations that are spatially isolated from each other will be divided into different connected regions.

[0194] Within each connected region, the deformation increment time series is used to examine the changes in deformation increment at each spatial location within the connected region over future time periods.

[0195] Within each connected region, the time series of deformation increments is analyzed:

[0196] When a spatial location exhibits a unidirectional increase in deformation increment over a long period of time in the future, the spatial location is marked as a spatial location where deformation continues to develop.

[0197] When the deformation increment of a certain spatial location fluctuates around zero or changes within a finite range in the future, the spatial location is marked as a spatial location where the deformation remains stable.

[0198] Within the connected region, while maintaining the initial set of executable locations, the edge locations of the connected region are filtered, retaining spatial locations that meet all of the following conditions as candidate locations for the spatial executable domain:

[0199] The default rate will be zero at all time points in the future period.

[0200] There are a sufficient number of preliminary executable locations that are adjacent to each other within the connected region, so that the connected region where the candidate location is located has a continuous set of spatial locations.

[0201] The deformation increment time series did not show a long-term unidirectional increase.

[0202] After filtering all connected regions, all candidate location sets are merged and recorded as a spatial executable domain.

[0203] It should be noted that the future time period of the spatial executable domain on the unified time axis is the time range, and the spatial location in the ore body spatial region is used as an index to give the location range in which mining operations are allowed to be performed in the future time period.

[0204] S5: Combine the spatial executable domain with the deformation field of future time periods to generate mining strategy adjustment instructions and equipment scheduling instructions. During the execution of instructions, collect the equipment operating status and ore body status, align them with the deformation field of future time periods on a unified time axis, calculate the ore body deviation and equipment deviation, and output the execution deviation sequence.

[0205] S5.1: Based on the spatial executable domain, determine the spatial range in which mining operations can be performed in the future time period, generate spatial constraints, and perform continuity filtering on the spatial executable domain on a unified time axis to output a set of candidate operation spatial locations;

[0206] Enumerate all time positions within a future time period on a unified time axis, and read the set of spatial positions corresponding to the spatial executable domain at each time position.

[0207] Perform a union summation on the set of spatial locations corresponding to all time locations within the future time period, and output the complete set of spatial locations covered by the spatial executable domain within the future time period as the candidate set of spatial locations to be filtered.

[0208] For each spatial location in the candidate selection spatial location set, determine whether the spatial location belongs to the spatial location set corresponding to the time location of the spatial executable domain on a unified time axis, and generate an executable tag sequence of spatial locations.

[0209] The executable marker sequence is filtered using a continuity filtering rule. When a spatial location has multiple consecutive adjacent time locations in a future time period that all satisfy the condition that the spatial location belongs to the set of spatial locations corresponding to the time locations of the spatial executable domain, the spatial location is judged to meet the continuity filtering condition.

[0210] When a spatial location does not meet the continuity filtering rules, the spatial location is judged as not meeting the continuity filtering conditions.

[0211] Spatial locations that meet the continuity filtering criteria are aggregated into a candidate job spatial location set, and the candidate job spatial location set is output.

[0212] S5.2: Using spatial constraints and the set of candidate job spatial locations as spatial screening conditions, spatial screening of deformation fields in future time periods is performed on a unified time axis to construct a future deformation reference field with spatial constraints;

[0213] Enumerate all time positions within a future time period on a unified time axis, and read the deformation field value of the future time period at each time position.

[0214] Using spatial constraints and the set of candidate job locations as spatial filtering conditions, at each time location, only the deformation field values ​​corresponding to spatial locations that meet the spatial filtering conditions are retained, and null values ​​are written for spatial locations that do not meet the spatial filtering conditions, thus constructing a future deformation reference field with spatial constraints.

[0215] The future deformation reference field that constructs spatial constraints is represented as:

[0216] ;

[0217] In the formula, As a future deformation reference field under spatial constraints, For the time position The set of spatial locations in the lower spatial executable domain that are allowed to perform mining operations. A null value indicates that the spatial location is excluded from mining operations at the corresponding time location.

[0218] It should be noted that the future deformation reference field of the spatial constraints only includes the deformation field values ​​corresponding to the set of candidate operation spatial locations at each time position, and does not include spatial locations with null values. Spatial locations with null values ​​are not included in the calculation of subsequent mining strategy adjustment instructions and equipment scheduling instructions.

[0219] S5.3: Utilize the future deformation reference field with spatial constraints to extract the consistency of the change magnitude and direction of the candidate job spatial location set in adjacent time locations and generate a change trend. Summarize the change magnitude within the spatial executable domain to generate a spatial executable domain risk benchmark.

[0220] On a unified time axis, enumerate all adjacent time position pairs within the future time period. For the candidate operation spatial position set, read the future deformation reference field values ​​of the spatial constraints for each spatial position, calculate the change amplitude between adjacent time positions, and record the change direction as upward, downward, or unchanged. Generate a change amplitude sequence and a change direction sequence (spatial positions with null values ​​marked in the future deformation reference field of the spatial constraints are not included in the calculation).

[0221] For each spatial location, count the number of times the upward direction and the number of times the downward direction occurs in the future time period, and count the number of adjacent switching between the upward and downward directions;

[0222] When the number of times the upward direction occurs is greater than the number of times the downward direction occurs, and the number of adjacent switching does not exceed the preset number of switching, the consistency of the output change direction is marked as upward consistency, and the output change trend is marked as a continuous growth trend.

[0223] When the number of occurrences of the downward direction is greater than the number of occurrences of the upward direction and the number of adjacent switching does not exceed the preset number of switching, the consistency of the output change direction is marked as downward consistency and the output change trend is marked as continuous downward trend. In other cases, the consistency of the output change direction is marked as inconsistent and the output change trend is marked as fluctuating trend.

[0224] For each adjacent time position pair, the change range of the candidate job spatial position set is arithmetically averaged within the spatial executable domain to obtain the time position risk benchmark.

[0225] Arrange all time location risk benchmarks within the future time period in chronological order to obtain the spatial executable domain risk benchmark sequence, and perform an arithmetic average on the spatial executable domain risk benchmark sequence to output the spatial executable domain risk benchmark.

[0226] It should be noted that the preset number of switching is a pre-set integer parameter used to limit the upper limit of the number of adjacent switching that the direction of change is allowed to occur in the future period (the example value is 1 to 3). For example, when the preset number of switching is 1, it means that one reversal of the direction of change is allowed, and when the preset number of switching is 2, it means that two reversals of the direction of change are allowed.

[0227] S5.4: Based on the consistency of change magnitude, change direction, change trend, and spatial executable domain risk benchmark, perform risk classification and priority ranking, determine the corresponding operation mode for spatial location, and generate mining strategy adjustment instructions;

[0228] On a unified time axis, read the change magnitude, direction consistency and trend of the candidate job spatial location set at each time position, and simultaneously read the spatial executable domain risk benchmark.

[0229] Risk classification is performed based on the magnitude of change and the risk benchmark of the spatial executable domain. The risk classification results are then corrected by combining the consistency of change direction and change trend, and a spatial location risk level label is output.

[0230] Within the set of candidate job locations, priority ranking is performed based on the spatial location risk level markers and changing trends, and the spatial location priority sequence is output.

[0231] Based on the risk classification results and changing trends, the corresponding operation mode for the spatial location is determined. The operation mode is selected between a stable operation mode and an operation mode that reduces the advance speed, and the correspondence between the spatial location and the operation mode is generated.

[0232] Mining strategy adjustment instructions are generated based on the correspondence between spatial location and operation mode and the spatial location priority sequence. The mining strategy adjustment instructions include at least the target spatial location, the corresponding operation mode, and the corresponding priority marker.

[0233] S5.5 combines mining strategy adjustment instructions with spatial constraints at each time position on a unified time axis, and generates equipment scheduling instructions by combining priority sorting.

[0234] Enumerate all time positions within a future period on a unified time axis, and read the target spatial position and corresponding operation mode associated with the corresponding time position in the spatial constraints and mining strategy adjustment instructions for each time position.

[0235] Perform an inclusion relationship determination on the target spatial location and the set of spatial locations recorded at the corresponding time location. If the inclusion relationship is true, retain the target spatial location and the corresponding operation method. If the inclusion relationship is false, remove the target spatial location and the corresponding operation method. Output the set of target spatial locations and corresponding operation methods that satisfy the spatial constraint.

[0236] The output order of target spatial location and corresponding operation mode is organized by combining the priority sorting results, and equipment scheduling instructions are generated. The equipment scheduling instructions include target spatial location, corresponding operation mode, corresponding time location and corresponding spatial limit range, and output the searchable records of equipment scheduling instructions at each time location in the future time period.

[0237] S5.6: During execution, collect the equipment operating status and ore body status and record the corresponding timestamps to generate an aligned equipment operating status sequence and ore body status sequence;

[0238] Throughout the entire timeframe of executing mining strategy adjustment instructions and equipment scheduling instructions, the operating status of the equipment is collected in real time, including equipment location, equipment action status, and equipment load, and a timestamp is recorded for each piece of equipment operating status.

[0239] The status of the ore body is collected in real time, including ore body deformation values, surrounding rock environmental indicators and structural changes, and a timestamp is recorded for each ore body status.

[0240] Based on the timestamp, the device operating status is arranged in chronological order to generate a device operating status sequence.

[0241] Arrange the states of the ore body in chronological order to generate an ore body state sequence.

[0242] The equipment operation status sequence and the ore body status sequence are integrated according to the continuous time position in a unified time axis to generate an aligned equipment operation status sequence and an aligned ore body status sequence.

[0243] S5.7: Based on a unified timeline, align the equipment operation status sequence with the ore body status sequence and the deformation field of future time periods, calculate the difference measure according to spatial and temporal location, and generate an execution deviation sequence;

[0244] Enumerate all time positions within a future period on a unified time axis, and at each time position, read the deformation field values, equipment operation status sequences, and corresponding time position records in the ore body status sequences for the future period.

[0245] For each time location, on the spatial grid, for each allowed spatial location corresponding to the spatial constraints, read the deformation field value corresponding to the spatial location where mining operations are allowed to be performed in the future deformation reference field of the spatial constraints, and the actual ore body deformation value recorded in the aligned ore body state sequence. The ore body deviation is calculated by comparing the future deformation field value with the actual ore body deformation value.

[0246] The deviation of the ore body is expressed as:

[0247] ;

[0248] In the formula, This represents the deviation of the ore body. The actual orebody state values ​​recorded in the orebody state sequence can be considered as the spatial location of the actual deformation field observed during the execution process. With time and location The values ​​below.

[0249] For each time point, the equipment position in the equipment operation status sequence is compared with the corresponding target position in the mining strategy adjustment instruction and equipment scheduling instruction. The equipment deviation is calculated, and the equipment position offset and the actual operation mode offset are recorded.

[0250] Arrange the ore body deviation and equipment deviation at each time point in a unified time axis order, and simultaneously record the set of allowable spatial locations corresponding to the spatial constraints at each time point, and output the execution deviation sequence.

[0251] Furthermore, based on the actual ore body deformation values ​​recorded in the ore body state sequence, and referring to a unified time axis and spatial grid, the time and spatial locations are organized, and an actual deformation field function is constructed during execution to represent the spatial location. With time and location The actual deformation field values ​​observed at the location.

[0252] S6: Based on the execution deviation sequence and default degree sequence, incremental learning samples are constructed on edge computing nodes and fed back to the cloud platform to update the pre-trained continuous-time deep spatiotemporal prediction model;

[0253] S6.1: Based on the execution deviation sequence, identify the time and spatial locations where the deviation exceeds the preset deviation judgment conditions on a unified time axis, generate deviation candidate segments, and associate the deviation candidate segments with the default degree sequence of future time periods to construct joint deviation segments;

[0254] Read all time positions in the execution deviation sequence on a unified time axis, and read the execution deviation values ​​of each spatial position within the ore body spatial region at each time position.

[0255] The absolute value of all execution deviations is calculated, and the deviation judgment threshold is determined based on the distribution characteristics of the absolute value of the execution deviation within the current operating cycle, and used as the preset deviation judgment condition.

[0256] The time and spatial locations where the absolute value of the execution deviation exceeds the deviation judgment threshold are combined as deviation points, and all deviation points are recorded as a set of deviation candidate locations according to their time and spatial locations.

[0257] Based on the set of deviation candidate locations, deviation points that are consecutively adjacent in time and adjacent in space are aggregated according to the temporal position order on a unified time axis and the adjacency relationship of spatial positions in the spatial grid to generate deviation candidate segments.

[0258] Based on the default degree sequence, the corresponding default degree value is found for the combination of time and space location recorded in each deviation candidate segment. Then, the execution deviation value and the default degree value are paired one by one within the deviation candidate segment to construct a joint deviation segment.

[0259] It should be noted that the deviation judgment threshold is determined based on the statistical distribution of the absolute value of the most recent execution deviation (the example value is the 90th quantile of the statistical distribution). During the operation, the statistical distribution of the absolute value of the latest execution deviation is recalculated at a fixed update interval to obtain the 90th quantile, and then smoothly updated based on the previous threshold.

[0260] S6.2: Integrate the state field data corresponding to the joint bias fragment and the bias candidate position set at the edge computing node to generate incremental learning input fragments, and organize them into an incremental learning sample set for updating the pre-trained continuous-time deep spatiotemporal prediction model according to the continuous-time format.

[0261] Based on the state field function that covers the spatial region of the ore body and changes continuously over time, the time position is recorded on a unified time axis, and the spatial position is recorded at each time position. The state field value is output through multi-source data fusion.

[0262] In the joint deviation segment, the temporal location set and spatial location set of each deviation candidate segment are read.

[0263] For each combination of time and space locations in the set of time locations and the set of space locations, extract the corresponding state field value from the state field function, extract the execution deviation value from the execution deviation sequence, and extract the default value from the default degree sequence.

[0264] The execution deviation value is taken as absolute value and then normalized according to the deviation judgment threshold. The normalized execution deviation value is then output.

[0265] The default score value is normalized according to the safe limit range determined by the upper and lower deformation values ​​corresponding to the spatial position in the logical reference, and the normalized default score value is output.

[0266] The normalized execution deviation value and the normalized default value are weighted and combined to output the joint deviation metric value at the corresponding time and spatial location combination.

[0267] Within the time range corresponding to the candidate deviation fragment, a state field input fragment is constructed for each spatial location, arranged in chronological order on a unified time axis, and a one-to-one correspondence is generated with the execution deviation value, default value, and joint deviation metric of the corresponding time location.

[0268] The state field input fragment is combined with the execution bias value, the default value, and the joint bias metric to form the incremental learning input fragment.

[0269] The incremental learning input segments are organized in a continuous time format. The state field input segment, execution bias value, default value and joint bias metric corresponding to each time position are recorded on a unified time axis, and the corresponding spatial position index is also recorded. The entire set of incremental learning input segments is recorded as the incremental learning sample set.

[0270] It should be noted that the default value is normalized according to the safe limit range determined by the upper and lower deformation values ​​corresponding to the spatial position in the logical reference. The normalization method is to divide the default value by the difference between the upper and lower deformation values.

[0271] If a logical reference provides only an upper limit value for deformation at a certain spatial location but does not include a lower limit value for deformation, then the lower limit value for deformation at that spatial location is zero.

[0272] If the logical reference only provides the lower limit of deformation and does not include the upper limit of deformation, then the upper limit of deformation is zero in spatial location.

[0273] S6.3: Based on edge computing nodes, perform format verification on incremental learning samples, and feed back the verified incremental learning samples to the cloud platform via wireless communication to generate cloud incremental learning samples;

[0274] On a unified timeline, perform format validation on the incremental learning sample set, read the incremental learning input segment corresponding to each time position one by one, and check whether the time position contains a complete state field input segment, execution bias value, default value and joint bias metric.

[0275] In the spatial dimension, check whether the spatial location index recorded in each time location is consistent with the spatial grid index used by the state field function, and check whether the records of each spatial location in the state field input segment correspond one-to-one with the corresponding spatial location index.

[0276] Compare the state field values ​​with the upper and lower deformation limits recorded at the corresponding spatial locations in the logical reference;

[0277] When there is a record that exceeds the upper limit of deformation or falls below the lower limit of deformation, and the absolute value of the excess is greater than the difference between the upper limit and the lower limit of deformation in the spatial location, the corresponding record will be marked as an abnormal record and the abnormal record will be removed.

[0278] Compare the relevant values ​​of the equipment with the rated parameters in the equipment manual; when there are abnormal records that exceed the rated parameters in the equipment manual, mark the corresponding records as abnormal records and remove them.

[0279] After completing the field integrity and index consistency checks, check whether there are any records with missing state field values ​​or missing execution bias values ​​in the incremental learning samples.

[0280] For records with missing data, mean interpolation is performed to complete them based on valid data of the same type at adjacent time positions, and the completed incremental learning sample set is used as the incremental learning sample that passes the format validation.

[0281] The verified incremental learning samples are encoded into data packets recognizable in the wireless communication protocol and sent to the cloud platform via the wireless communication link.

[0282] After receiving the data packets, the cloud platform decodes and reassembles the incremental learning samples, and recovers the complete cloud-based incremental learning samples under a unified time axis and spatial location index.

[0283] S6.4: Update the pre-trained continuous-time deep spatiotemporal prediction model on the cloud platform based on incremental learning samples in the cloud, and synchronize the updated continuous-time deep spatiotemporal prediction model to the edge computing nodes;

[0284] Load the pre-trained continuous-time deep spatiotemporal prediction model into the cloud platform, and read the state field input fragments, execution bias values, default values ​​and joint bias measures recorded in the incremental learning samples in the cloud on a unified time axis.

[0285] The state field input segment is used as the input to the pre-trained continuous-time deep spatiotemporal prediction model. The values ​​of the actual deformation field function constructed during the execution process at the corresponding time and spatial positions are used as the expected output. The joint bias metric is used as the weight of the training samples, and incremental updates are performed in the continuous-time deep spatiotemporal prediction model.

[0286] The updated continuous time-depth spatiotemporal prediction model is recorded as a new model version on the cloud platform and synchronized to the edge computing nodes via wireless communication.

[0287] This embodiment also provides a computer device applicable to automated mining methods based on digital mining technology, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the automated mining method based on digital mining technology as proposed in the above embodiment.

[0288] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0289] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the automated mining method based on digital mining technology as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0290] In summary, this invention significantly improves the state perception accuracy of multi-source orebody data and equipment operation data by: correcting and aligning data streams with time references, and performing adaptive cleaning and multi-scale fusion at the edge. It constructs a continuous, stable orebody state field with unified reference coordinates. By inputting the state field into a pre-trained continuous-time depth-spatial prediction model and combining it with a differentiable logic monitor to continuously assess the default degree of future deformation fields, a spatially executable domain is quantified, enabling risk identification to have coherence and fine-grained expression. Incremental learning samples are constructed on edge computing nodes using orebody deviation, equipment deviation, and default degree sequences obtained during instruction execution, and then fed back to the cloud to update prediction model parameters. This allows strategy generation to adaptively adjust with changes in geological conditions and operating conditions, comprehensively considering the safety, reliability, and operational efficiency of the mining process.

[0291] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automated mining method based on digital mining technology, characterized in that: include, The data collected from the ore body and the mining equipment operation data are time-referenced to construct a unified timeline and generate an aligned data stream, which is then transmitted to the edge computing node via wireless communication. Based on edge computing nodes, adaptive cleaning and multi-scale fusion of aligned data streams are performed to construct a state field covering the ore body. Input the state field into a pre-trained continuous-time deep spatiotemporal prediction model and output the deformation field for future time periods; The preset set of safety limit rules is compiled into a differentiable logic monitor, which performs continuous-time default assessment on the deformable field in future time periods, outputs a default sequence, and generates a spatially executable domain. By combining the spatial executable domain with the deformation field of future time periods, mining strategy adjustment instructions and equipment scheduling instructions are generated. During the execution of the instructions, the equipment operating status and ore body status are collected and aligned with the deformation field of future time periods on a unified time axis. The ore body deviation and equipment deviation are calculated, and the execution deviation sequence is output. Based on the execution deviation sequence and default degree sequence, incremental learning samples are constructed on edge computing nodes and fed back to the cloud platform to update the pre-trained continuous-time deep spatiotemporal prediction model. The process of compiling a preset set of security limit rules into a differentiable logic monitor, performing continuous-time violation assessments on the deformable field in future time periods, outputting a violation sequence, and generating a spatially executable domain involves the following steps: Based on the deformation field in the future time period, the deformation change of the ore body is calculated according to the adjacent time positions on the unified time axis, and a deformation sequence is constructed. A set of safety limit rules is set according to the mining regulations, and a logical reference is generated based on the set of safety limit rules. The preset safety limit rules are compiled into a differentiable logic monitor. The deformed sequence is input into the differentiable logic monitor, and the default degree of future time periods is calculated in the differentiable logic monitor based on the logic reference. The default degree sequence of spatial domain judgment is output. Based on the default degree sequence, determine the location range and non-executable location range that meet the preset safety limit rules within the spatial area of ​​the ore body, and generate preliminary executable area data of the spatial area. Based on spatial connectivity and future deformation trends, the preliminary executable region data is processed to generate a spatial executable region.

2. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The specific steps for constructing a unified timeline and generating aligned data streams, which are then transmitted to edge computing nodes via wireless communication, are as follows. Collect ore body data and mining equipment operation data, record the original timestamps corresponding to the ore body data and mining equipment operation data, perform time base correction, and generate a unified timeline; By aligning ore body data with mining equipment operation data using a unified timeline, an aligned data stream is generated and transmitted to edge computing nodes via wireless communication protocol encoding.

3. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The specific steps for constructing the state field covering the ore body are as follows. At the edge computing node, adaptive noise removal is performed on the ore body data and mining equipment operation data in the aligned data stream to generate primary cleaned data. Based on the timestamps of the aligned data streams, perform missing data imputation and anomaly detection on the primary cleaned data, and output calibrated multi-source data; The calibrated multi-source data are fused at multiple scales according to spatial location and temporal order to construct a state field covering the spatial region of the ore body.

4. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The specific steps for outputting the deformation field for future time periods are as follows: The state field covering the ore body space is organized into a model input sequence in a continuous time format, input into a pre-trained continuous time depth spatiotemporal prediction model, and spatiotemporal extrapolation of the state field is performed to output an extrapolation data set. Based on the inferred data set, the deformation trend and spatial change structure of the ore body in future time periods are read, and candidate deformation segments for future time periods are generated. The candidate deformation segments for future time periods are organized according to a unified time axis and spatial grid, and the deformation field for future time periods is output.

5. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The specific steps for combining the spatial executable domain with the deformation field of future time periods to generate mining strategy adjustment instructions and equipment scheduling instructions are as follows. Based on the spatial executable domain, the spatial range in which mining operations are allowed to be carried out in the future time period is determined, spatial constraints are generated, and continuous filtering of the spatial executable domain is performed on a unified time axis to output a set of candidate operation spatial locations; Using spatial constraints and the set of candidate job locations as spatial screening conditions, the deformation field of future time periods is spatially screened on a unified time axis to construct a spatially constrained future deformation reference field. By utilizing the future deformation reference field with spatial constraints, the consistency of the change magnitude and direction of the candidate job spatial location set in adjacent time locations is extracted and a change trend is generated. The change magnitude is then summarized within the spatial executable domain to generate a spatial executable domain risk benchmark. Based on the consistency of change magnitude, change direction, change trend, and spatial executable domain risk benchmark, risk classification and priority ranking are performed to determine the corresponding operation mode for spatial location and generate mining strategy adjustment instructions. On a unified timeline, mining strategy adjustment instructions and spatial constraints are combined at each time position, and equipment scheduling instructions are generated by prioritizing them.

6. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The process of collecting equipment operating status and ore body status during instruction execution, aligning them with the deformation field in future time periods on a unified time axis, yields an execution deviation sequence. The specific steps are as follows: During the execution process, the equipment operating status and ore body status are collected and the corresponding timestamps are recorded to generate an aligned equipment operating status sequence and ore body status sequence. Based on a unified timeline, the equipment operation status sequence is aligned with the ore body status sequence and the deformation field of future time periods. Difference metrics are calculated based on spatial and temporal locations to generate an execution deviation sequence.

7. The automated mining method based on digital mining technology as described in claim 1, characterized in that: The process involves constructing incremental learning samples on edge computing nodes based on the execution bias sequence and default degree, feeding these samples back to the cloud platform, and updating the pre-trained continuous-time deep spatiotemporal prediction model. The specific steps are as follows: Based on the execution deviation sequence, the time and spatial locations where the deviation exceeds the preset deviation judgment conditions are identified on a unified time axis, deviation candidate segments are generated, and deviation candidate segments are correlated with the default degree sequence of future time periods to construct joint deviation segments. The state field data corresponding to the joint bias fragment and the bias candidate position set are integrated at the edge computing node to generate incremental learning input fragments, and then organized into an incremental learning sample set for updating the pre-trained continuous-time deep spatiotemporal prediction model according to the continuous-time format. Based on edge computing nodes, the incremental learning samples are formatted and validated. The validated incremental learning samples are then fed back to the cloud platform via wireless communication to generate cloud-based incremental learning samples. The pre-trained continuous-time deep spatiotemporal prediction model is updated on the cloud platform based on incremental learning samples in the cloud, and the updated continuous-time deep spatiotemporal prediction model is synchronized to the edge computing nodes.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the automated mining method based on digital mining technology as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the automated mining method based on digital mining technology as described in any one of claims 1 to 7.

Images