Tungsten ore ground pressure intelligent early warning method based on multi-source data fusion

Abstract

The invention discloses a tungsten ore ground pressure intelligent early warning method based on multi-source data fusion, and relates to the technical field of mine safety monitoring and intelligent early warning, and the method comprises the following steps: obtaining and preprocessing multi-source monitoring data; analyzing the causal relationship between the micro-seismic event and the displacement rate change by using Granger causal test; then, in combination with causal correlation and seepage stress coupling data, graded early warning data are generated through a long-short-term memory network; and finally, displaying a result on a three-dimensional visualization platform, and generating a disaster avoidance route through an optimization algorithm. According to the method, the D-S evidence theory is utilized to carry out fusion processing on multi-source heterogeneous data under the uncertain condition, and intelligent and accurate early warning of the ground pressure of the tungsten mine is achieved; according to the tungsten ore ground pressure intelligent early warning method based on multi-source data fusion, a visual and operable emergency response scheme is provided for mine management personnel, and the decision-making efficiency and the practical value of mine safety production are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mine safety monitoring and intelligent early warning, and particularly relates to an intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion. Background Art

[0002] In deep mine mining, due to the complex geological environment, high stress concentration, and variable hydrogeological conditions, it is crucial to master the mechanical and hydraulic evolution characteristics of underground rock masses in real time to ensure mine safety. To achieve accurate early warning, a multi-source monitoring system needs to be constructed relying on various sensors such as stress gauges, displacement gauges, microseismic instruments, and water pressure gauges. However, due to differences in the time resolution, spatial density, and unit dimensions of the data collected by different sensors, directly conducting joint analysis faces the problem of inconsistent spatio-temporal scales. Therefore, how to effectively map heterogeneous data such as stress, displacement, microseismicity, and water pressure to a unified spatio-temporal grid and maintain the integrity of key feature information has become a major technical difficulty in current multi-source data fusion analysis. In addition, during the construction of the early warning model, multi-source monitoring data is often limited by factors such as on-site environmental interference, equipment drift, and sampling errors, resulting in different types of noise mixed in the original signal. To improve data quality, specialized filtering algorithms such as wavelet denoising, moving average, or Kalman filtering are usually designed for various types of data. However, excessive processing may cause some weak but critical precursor information to be eliminated, affecting the sensitivity of the subsequent early warning model. At the same time, in the fusion analysis stage, how to reasonably determine the weights of each monitoring index in the early warning model and screen representative coupling parameter combinations is also an important bottleneck for achieving high-precision risk identification. Especially when constructing a seepage-stress coupling model, due to the inhomogeneity and anisotropy characteristics of rock mass materials, it is difficult to accurately obtain key physical parameters such as the permeability coefficient tensor, which restricts the reliability and generality of the model. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion to solve the problems existing in the prior art.

[0004] To achieve the above purpose, the present invention provides the following technical solution: An intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion, the method comprising: S1. Obtain multi-source monitoring raw data; S2. For the ground pressure distribution data in the multi-source monitoring raw data, use wavelet transform to remove high-frequency noise to generate denoised ground pressure data, for the surrounding rock displacement data, eliminate equipment drift through a filtering algorithm to generate corrected displacement data, for the microseismic event frequency and microseismic energy amplitude, use the short-time average and long-time average algorithms to extract effective microseismic event data, and for the water pressure change data, extract the water pressure trend data through moving average filtering to obtain preprocessed feature data; S3. By processing the microseismic event frequency, microseismic energy amplitude in the preprocessed feature data and the displacement rate change in the corrected displacement data, the Granger causality test method is used to calculate the event occurrence time sequence and the causal lag time, determine the causal influence intensity of the microseismic event on the displacement rate change, and obtain the causal correlation data; S4. Obtain the causal correlation data and the seepage stress coupling data as inputs, and use the long short-term memory network to perform time series processing on the event occurrence time sequence, displacement rate change and water pressure trend data within the time window length, generate the hidden state data including the hidden state dimension, and obtain the hierarchical early warning data; S5. Render the ground pressure distribution data, the spatial distribution range of the microseismic event frequency and the water pressure trend data in the hierarchical early warning data through a 3D visualization platform, and use the A optimization algorithm to generate an evacuation route according to the high-risk early warning level and the spatial distribution range, and obtain the emergency decision-making path data.

[0005] Preferably, S2 further includes calculating the influence of the water pressure change data on the spatial distribution range of the surrounding rock deformation by combining the de-noised ground pressure data, water pressure trend data and corrected displacement data in the preprocessed feature data, and generating the seepage stress coupling data.

[0006] Preferably, after obtaining the causal correlation data in S3, the causal correlation data and the seepage stress coupling data are obtained as inputs, and the long short-term memory network is used to perform time series processing on the event occurrence time sequence, displacement rate change and water pressure trend data within the time window length, and generate the hidden state data including the hidden state dimension.

[0007] Preferably, S3 further includes calculating the ground pressure anomaly probability through the hidden state dimension in the hidden state data, combining the input feature weights to generate the ground pressure risk index, and obtaining the risk assessment data representing the anomaly probability in the future time period.

[0008] Preferably, in S3, the D-S evidence theory is used to perform uncertainty reasoning on the preprocessed feature data from different sensors, and a multi-parameter joint confidence is constructed.

[0009] Preferably, the multi-parameter joint confidence is calculated based on the evidence weights corresponding to different monitoring indicators, and the evidence weights are set according to the feature importance results of the random forest model.

[0010] Preferably, the wavelet transform used in S2 is the Daubechies5 wavelet basis, and the decomposition level is not less than 3 layers, which is used to extract the effective frequency band of 0.1 Hz - 10 Hz in the ground pressure distribution data and enhance the suppression ability of the high-frequency blasting interference.

[0011] Preferably, in S2, the Kalman filtering algorithm is used for filtering the surrounding rock displacement data, and an environmental temperature compensation mechanism is combined to correct the equipment measurement error caused by thermal expansion and contraction.

[0012] Preferably, when extracting the water pressure trend data in S2, a moving average filter is used, and the window length is set to 60 seconds to smooth the periodic disturbance and retain the mutation characteristics.

[0013] Preferably, in S5, the A optimization algorithm is used to calculate the shortest disaster avoidance path for the generated emergency decision path data. Combining with the high-risk space area in the hierarchical early warning data, the priority of the path nodes is dynamically adjusted to realize real-time path update.

[0014] It can be seen from the above technical solutions that the present invention has the following beneficial effects: This intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion maps the monitoring data such as stress, displacement, microseismicity, and water pressure accurately to a unified coordinate system by constructing a three-dimensional geological model of the mine, and uses a unified time synchronization mechanism to solve the problem of inconsistent spatio-temporal scales in data acquisition, improving the accuracy and comparability of multi-source information fusion analysis. For the noise characteristics of different types of monitoring data, wavelet denoising, moving average, Kalman filtering and other algorithms are respectively used for directional noise reduction processing, effectively suppressing interference while retaining key precursor information, providing high-quality data input for model training. By introducing the Granger causality test method to quantify the causal relationship between microseismic events and displacement rate and other indicators, and combining with the seepage-stress coupling model to identify the driving effect of water pressure change on rock mass deformation, the quantitative description of the ground pressure evolution mechanism is realized. By using the random forest model to analyze the importance of each monitoring feature, dynamically screening key coupling indicators and adjusting their weights in the model, taking into account both model accuracy and calculation efficiency, improving the reliability of ground pressure risk discrimination. Using the long short-term memory network (LSTM) to deeply model multi-source time series data, generating risk prediction results containing hidden state expressions, and supporting hierarchical early warning and dynamic update of ground pressure anomalies by fusing multi-parameter confidence indices. Combining with a three-dimensional visualization platform and the A optimization algorithm to generate a spatial disaster avoidance path, providing an intuitive and operable emergency response plan for mine management personnel, comprehensively improving the decision-making efficiency and practical value of mine safety production. Description of the Drawings

[0015] Figure 1 It is a flow chart of the method of the present invention; Figure 2 It is a signal transmission schematic diagram of the present invention. Detailed Embodiments

[0016] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] As Figure 1-2 shown, the present invention provides a technical solution: an intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion, characterized in that the method includes: S1. Obtain multi-source monitoring raw data; S2. For the ground pressure distribution data in the multi-source monitoring raw data, wavelet transform is used to remove high-frequency noise to generate denoised ground pressure data. For the surrounding rock displacement data, a filtering algorithm is used to eliminate equipment drift to generate corrected displacement data. For the microseismic event frequency and microseismic energy amplitude, the short-time average and long-time average algorithms are used to extract effective microseismic event data. For the water pressure change data, a moving average filter is used to extract the water pressure trend data, and preprocessed feature data is obtained; S3. Through the microseismic event frequency in the preprocessed feature data, the microseismic energy amplitude, and the change in displacement rate in the corrected displacement data, the Granger causality test method is used to calculate the event occurrence time sequence and the causal lag time, and the causal influence intensity of the microseismic event on the change in displacement rate is determined to obtain causal correlation data; S4. Obtain the causal correlation data and the seepage stress coupling data as inputs, and use a long short-term memory network to perform time series processing on the event occurrence time sequence, the change in displacement rate, and the water pressure trend data within the time window length to generate hidden state data including the hidden state dimension, and obtain hierarchical early warning data; S5. Render the ground pressure distribution data, the spatial distribution range of the microseismic event frequency, and the water pressure trend data in the hierarchical early warning data through a three-dimensional visualization platform, and use the A optimization algorithm to generate an evacuation route according to the high-risk early warning level and the spatial distribution range to obtain emergency decision-making path data.

[0018] Based on the theoretical framework of multi-source data fusion and causal association analysis, the present invention fuses ground pressure, microseismic, water pressure, and displacement data to establish a unified spatio-temporal data framework. First, in terms of data fusion, the D-S evidence theory is used for uncertainty reasoning, and corresponding weights are assigned to different sensor data, and the event credibility is jointly formed. The joint confidence calculation formula is as follows: , ; Among them, is the evidence weight of different sensing sources, Propose for disaster events to determine the credible fusion relationship between multi-source data. Further, based on the random forest model, rank the importance of features and screen key coupling indicators such as the combination of "stress gradient + microseismic value" to improve the data dimensionality reduction and feature screening effects.

[0019] In terms of correlation modeling, construct a "seepage-stress coupling model" to reveal the internal physical mechanism of the influence of variables such as microseismic and water pressure changes on the stability of surrounding rock. The model expression is: ; Among them, is the permeability coefficient tensor, is the pore water pressure, is the volume strain, Coefficient. The above equation characterizes the dynamic regulation mechanism of fluid seepage on the response of in-situ stress. Subsequently, through Granger causality test, quantitatively evaluate the leading warning ability of microseismic events to displacement rate, and improve timeliness and sensitivity.

[0020] In the time series modeling part, introduce the LSTM long short-term memory network, take causal correlation data and coupled stress data as inputs, model the multi-variable dynamic sequence within the time window, and generate a high-dimensional warning output with time state memory. Finally, through a three-dimensional visualization platform for information space projection, and based on the A optimization algorithm to plan the evacuation path, complete the closed-loop warning system from perception, recognition to response.

[0021] S2 also includes calculating the influence of water pressure change data on the spatial distribution range of surrounding rock deformation according to the de-noised ground pressure data, water pressure trend data and corrected displacement data in the preprocessed feature data, and combining the permeability coefficient tensor and volume strain to generate seepage stress coupling data.

[0022] After completing the preprocessing, by integrating the three types of data of ground pressure, water pressure and displacement, further evaluate the influence range of water pressure dynamics on surrounding rock deformation. This process is based on physical coupling modeling, analyzing the pressure conduction path and deformation response mechanism during the propagation of water pressure change in multi-dimensional space. The permeability coefficient tensor is used to reflect the seepage ability of the medium in different directions, and the volume strain reflects the response characteristics of the surrounding rock under water pressure disturbance. By fusing these physical parameters with the preprocessed data, establish the mapping relationship between water pressure change and surrounding rock deformation, realize the accurate modeling of the spatial distribution of surrounding rock deformation under the influence of seepage, and generate seepage stress coupling data based on this, providing the basic support for subsequent causal analysis and warning modeling.

[0023] This method significantly improves the ability of spatial modeling of ground pressure monitoring data and can quantify the specific impact area of water pressure changes on the stability of surrounding rocks. By constructing a physical coupling relationship, data processing is no longer limited to surface analysis, but delves into the underlying geological dynamic mechanism, thereby enhancing the scientificity and reliability of the early warning system. Compared with traditional independent data preprocessing methods, the steps described in the present invention provide a higher-resolution risk identification ability and have stronger adaptability and interpretability for the ground pressure environment in complex mining areas.

[0024] After obtaining the causal correlation data in S3, the causal correlation data and seepage stress coupling data are obtained as inputs, and a long short-term memory network is used to perform time series processing on the event occurrence time series, displacement rate change, and water pressure trend data within the time window length, generating hidden state data containing hidden state dimensions.

[0025] After obtaining the causal correlation data, in order to further explore the potential patterns of multi-source features evolving over time, the causal information and seepage stress coupling data are jointly input into a long short-term memory network (LSTM). LSTM has the ability to capture long-term dependence relationships and is particularly suitable for processing ground pressure evolution data with time series characteristics. During the time series processing, the system dynamically models multi-dimensional variables such as the occurrence time series of microseismic events, displacement rate changes, and water pressure trends within a specified time window. LSTM selects key information through its gating mechanism, retains the states highly relevant to risk evolution, and outputs high-dimensional hidden state data for expressing the potential ground pressure risk level at the current time period. This hidden state, as an intermediate representation form, can comprehensively reflect the time series characteristics and correlation strength of multi-source information, providing deep data support for subsequent risk level assessment and path optimization.

[0026] By introducing the LSTM network for time series modeling, the early warning system can effectively depict complex geological evolution paths and improve the ability to perceive abnormal evolution trends. Compared with traditional static feature analysis methods, this method can capture potential time series patterns and dynamic interaction relationships in multi-source data, greatly enhancing the prediction accuracy and stability of the model. The generation of hidden state dimensions also has strong expressive ability and can be used as a unified characterization variable to participate in subsequent modules such as risk judgment, level division, and spatial display, contributing to the structuring, quantification, and intelligentization of early warning results.

[0027] S3 also includes calculating the ground pressure anomaly probability using the sigmoid activation function type through the hidden state dimensions in the hidden state data, generating a ground pressure risk index by combining the input feature weights, and obtaining risk assessment data representing the anomaly probability in the future time period.

[0028] After the long short-term memory network completes the time series modeling and outputs the hidden state data, the system further uses the hidden state dimension as the input, and maps the hidden state to a ground pressure anomaly probability value between 0 and 1 through the sigmoid activation function. This probability represents the likelihood of a ground pressure anomaly event occurring at the current moment and within the future time window. To enhance the multi-factor expression ability of risk assessment, an input feature weight mechanism is also introduced to weight and integrate the influence degrees of key input variables (such as displacement rate, water pressure trend, microseismic frequency, etc.) on the prediction result. The system calculates the contribution degree of each feature to the anomaly probability, constructs a fusion ground pressure risk index, quantifies the ground pressure evolution trend under the combined action of multi-source indicators, and finally forms future-oriented risk assessment data, which serves as the basis for subsequent early warning grading and response decision-making.

[0029] This step enhances the interpretability and quantification ability of risk identification. The sigmoid function ensures that the output of the anomaly probability has good normalization characteristics, facilitating intuitive understanding and hierarchical management; the introduction of feature weights strengthens the interpretability and transparency of the model, enabling the early warning system to not only give results but also feedback the key influences, facilitating decision-making optimization. The finally generated risk assessment data has both timeliness and structure, which helps to achieve dynamic monitoring and real-time risk prediction, and further improves the practical application value and deployment efficiency of the early warning system.

[0030] In S3, the D-S evidence theory is used to perform uncertainty reasoning on the preprocessed feature data from different sensors and construct a multi-parameter joint confidence.

[0031] In the process of handling step S3, to solve the inconsistency problems caused by different precisions, sampling methods and response time lags among multi-source monitoring data, the Dempster-Shafer (D-S) evidence theory is introduced for uncertainty fusion analysis. By taking the preprocessed feature data output by different sensors (including ground pressure data, microseismic features, displacement rate, water pressure change trend, etc.) as independent evidence sources, the basic credibility assignment functions for the ground pressure anomaly event are constructed respectively. The system sets the weights according to the signal-to-noise ratio, historical accuracy and response sensitivity of each sensor, and performs multi-parameter joint calculation through the D-S synthesis rule to obtain the joint confidence value for a certain potential ground pressure event. This joint confidence is used to comprehensively evaluate the degree of common support of multi-source data for the possibility of an event occurring, and is a credible basis for subsequent causal analysis and risk modeling.

[0032] Through the introduction of the D-S evidence theory, the fusion processing of multi-source heterogeneous data under uncertain conditions is realized, significantly improving the accuracy and fault tolerance of anomaly recognition. Compared with traditional weighted average or simple logical judgment methods, the D-S method has stronger information integration ability and robustness, and can maintain stable decision-making output when part of the sensor data is missing or in conflict. The construction of joint confidence not only enhances the adaptability of the system in complex monitoring environments, but also provides a more credible input basis for subsequent early warning models.

[0033] The multi-parameter joint confidence is calculated based on the evidence weights corresponding to different monitoring indicators, and the evidence weights are set according to the feature importance results of the random forest model.

[0034] To improve the scientificity and objectivity of evidence weight setting in D-S evidence fusion, when constructing the multi-parameter joint confidence, the system introduces a random forest model to evaluate the importance of various monitoring indicators in the preprocessed feature data. As an ensemble learning algorithm, the random forest can quantify the contribution degree of input variables in classification or prediction tasks by constructing multiple decision tree models, so as to generate the relative importance scores of each feature. Based on the scoring results, the system assigns different weights to the evidence corresponding to each monitoring indicator (such as microseismic frequency, ground pressure fluctuation amplitude, displacement rate, water pressure change rate, etc.), as the basis for basic probability assignment in the D-S evidence theory. On this basis, by fusing the evidence information with weight assignment, a joint confidence result that better conforms to the actual observation law is generated. This method not only improves the sensitivity of the evidence synthesis process to the value of the data itself, but also strengthens the intelligent judgment ability of the entire system in dealing with data complexity and uncertainty.

[0035] Introducing the feature importance results of the random forest model into the weight assignment link of the D-S evidence theory improves the accuracy and adaptability of multi-source data fusion. This method makes the influence of each type of sensor data in the final decision consistent with its actual contribution to the early warning effect, enhancing the adaptive adjustment ability of the model. Compared with the traditional fixed weight setting method, this method can dynamically adjust the fusion ratio of each evidence source when facing new data or scenario changes, improving the generalization ability of the model under different mining areas and different monitoring backgrounds.

[0036] The wavelet transform used in S2 is the Daubechies5 wavelet basis, and the decomposition level is not less than 3 layers, which is used to extract the effective frequency band of 0.1Hz - 10Hz in the ground pressure distribution data and enhance the ability to suppress high-frequency blasting interference.

[0037] In the preprocessing stage of ground pressure distribution data, the system uses the Daubechies5 (db5) wavelet basis for multi-level wavelet decomposition. This wavelet basis has good time-frequency localization characteristics and strong smoothness, and is suitable for processing ground pressure monitoring signals with mutation or non-stationary characteristics. Through wavelet decomposition of no less than 3 levels, the characteristic components of different frequency bands in the signal can be effectively separated. The system further reconstructs the frequency bands at each decomposition scale, retains the effective information components in the range of 0.1 Hz to 10 Hz, and filters out the interference components with frequencies higher than this range, especially the high-frequency interference caused by engineering blasting, equipment noise, etc., so as to obtain more stable and real ground pressure change trend data.

[0038] Adopting the Daubechies5 wavelet basis and combining with the multi-level decomposition strategy can effectively improve the denoising accuracy in the preprocessing stage of ground pressure signals. Especially in the face of high-frequency strong interference backgrounds such as blasting, it has stronger filtering ability and information retention performance. By focusing on the 0.1 Hz - 10 Hz frequency band, not only the low-frequency information closely related to ground pressure evolution is retained, but also the sensitivity to key change patterns is enhanced, thus improving the accuracy and stability of subsequent risk identification and causal analysis, and providing a more reliable data basis for ground pressure anomaly identification.

[0039] In S2, the Kalman filter algorithm is used for the filtering process of surrounding rock displacement data, and an environmental temperature compensation mechanism is combined to correct the equipment measurement error caused by thermal expansion and contraction.

[0040] The surrounding rock displacement monitoring data is often affected by environmental temperature fluctuations, resulting in the thermal expansion and contraction effect of the sensor during long-term operation, thus introducing systematic measurement deviations. To improve the accuracy and stability of displacement data, this method uses the Kalman filter algorithm to dynamically estimate and optimize the displacement data in the preprocessing stage. The Kalman filter is based on a linear state space model, which can combine the current observation value and the predicted state to correct the measurement error in real time, improving the smoothness and reliability of the data. On this basis, the system introduces an environmental temperature compensation mechanism. By real-time monitoring the temperature change around the equipment, a temperature-displacement error mapping relationship is established to compensate and correct the system offset caused by thermal expansion and contraction, further eliminating the interference of non-structural errors on the monitoring data.

[0041] Adopting the dual mechanism of Kalman filter combined with temperature compensation significantly improves the accuracy and credibility of surrounding rock displacement data. This method can not only dynamically respond to measurement errors and environmental fluctuations, but also effectively isolate temperature interference factors, retain the real geological displacement trend, and enhance the quality of basic data for subsequent microseismic coupling analysis and causal modeling. Compared with traditional static filtering methods, this method has stronger adaptive ability and real-time update characteristics, and is suitable for long-term stable monitoring in complex mine environments.

[0042] When extracting the water pressure trend data in S2, a moving average filter is adopted, and the window length is set to 60 seconds, which is used to smooth the periodic disturbances and retain the mutation characteristics.

[0043] In the process of preprocessing the water pressure monitoring data, in order to enhance the stability and effectiveness of trend extraction, this method uses a moving average filter to smooth the original water pressure signal. The system sets the filter window length to 60 seconds to ensure that on the basis of covering the fluctuations in a certain period of time, the periodic disturbances caused by sampling errors, equipment perturbations, local water inrushes, etc. can be effectively suppressed. At the same time, this window length takes into account the response speed and data smoothing degree, so that the water pressure mutations occurring in a short period of time can still be effectively identified and retained, avoiding the over-smoothing of key signals. The filtering process continuously updates the input signal, calculates the average value within the sliding window point by point, and generates a representative water pressure trend curve, providing clear and stable data support for subsequent coupling modeling and risk identification.

[0044] Using a 60-second moving average filter to process the water pressure data can effectively improve the stability of trend extraction, retain the mutation characteristics related to abnormal events at the same time, and enhance the sensitivity to the precursor signals of risks. Compared with the traditional static filtering method, the sliding window mechanism can update the processing results in real time, with good real-time performance and adaptability, especially suitable for the continuous water pressure monitoring scenario under the complex hydrogeological conditions in tungsten mines. This method provides high-quality input for subsequent causal relationship analysis and time series modeling, improving the accuracy and reliability of the overall early warning system.

[0045] The emergency decision-making path data generated in the above S5 uses the A optimization algorithm to calculate the shortest disaster avoidance path, combines the high-risk spatial areas in the hierarchical early warning data, dynamically adjusts the priority of path nodes, and realizes real-time path update.

[0046] During the process of generating the emergency response path, the system is based on the mine topology after 3D spatial modeling and uses the A* optimization algorithm to search for paths through each passable node, calculating the shortest disaster avoidance path from the current location to the safe shelter area. The A* optimization algorithm combines a heuristic function and a cost function to comprehensively evaluate the path cost and has efficient optimal path solving capabilities. On this basis, the system further integrates the high-risk area data output by the hierarchical early warning module, uses the spatial locations with a high probability of abnormal ground pressure as constraints, and dynamically adjusts the priorities of the nodes associated with this area during path search. For example, when a certain area is marked as the "red warning" level, the priorities of its related nodes are significantly reduced, and the system automatically avoids the path options that cross the risk area. In addition, the system supports a real-time path update mechanism. Based on changes in the warning level or spatial risk distribution, it re-evaluates and calculates the current path to ensure that the path planning always maintains optimal safety. This mechanism meets the dynamic response requirements during the evolution of sudden disasters and provides efficient and feasible disaster avoidance route guidance for on-site operators.

[0047] This embodiment combines an efficient path planning algorithm with dynamic early warning data to achieve the intelligent generation and real-time update of the emergency disaster avoidance path. By introducing the warning level factor into the path planning, it effectively avoids high-risk areas and improves the safety and practicality of the path. Compared with traditional static route planning, this method has stronger environmental adaptability and dynamic adjustment capabilities, can significantly improve the emergency response efficiency in the event of sudden ground pressure in the mine, reduce casualties and equipment losses, and provide strong guarantee for safe production.

[0048] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent early warning method for tungsten mine ground pressure based on multi-source data fusion, characterized in that The method includes: S1. Obtain multi-source monitoring raw data; S2. For the ground pressure distribution data in the multi-source monitoring raw data, use wavelet transform to remove high-frequency noise to generate denoised ground pressure data. For the surrounding rock displacement data, eliminate equipment drift through a filtering algorithm to generate corrected displacement data. For the microseismic event frequency and microseismic energy amplitude, use the short-time average and long-time average algorithms to extract effective microseismic event data. For the water pressure change data, extract the water pressure trend data through moving average filtering to obtain preprocessed feature data; S3. Through the microseismic event frequency and microseismic energy amplitude in the preprocessed feature data and the displacement rate change in the corrected displacement data, use the Granger causality test method to calculate the event occurrence time sequence and causal lag time, and determine the causal influence intensity of the microseismic event on the displacement rate change to obtain causal correlation data; S4. Obtain the causal correlation data and seepage stress coupling data as inputs, and use a long short-term memory network to perform time sequence processing on the event occurrence time sequence, displacement rate change, and water pressure trend data within the time window length to generate hidden state data including hidden state dimensions to obtain hierarchical early warning data; S5. Render the ground pressure distribution data, the spatial distribution range of the microseismic event frequency, and the water pressure trend data in the hierarchical early warning data through a 3D visualization platform, and use the A optimization algorithm to generate an evacuation route according to the high-risk early warning level and the spatial distribution range to obtain emergency decision-making path data.

2. The intelligent early warning method for tungsten mine ground pressure based on multi-source data fusion according to claim 1, wherein: S2 also includes calculating the influence of the water pressure change data on the spatial distribution range of the surrounding rock deformation by combining the denoised ground pressure data, water pressure trend data, and corrected displacement data in the preprocessed feature data with the permeability coefficient tensor and volumetric strain to generate seepage stress coupling data.

3. The intelligent early warning method for tungsten mine ground pressure based on multi-source data fusion according to claim 2, characterized in that: After obtaining the causal correlation data in S3, obtain the causal correlation data and seepage stress coupling data as inputs, and use a long short-term memory network to perform time sequence processing on the event occurrence time sequence, displacement rate change, and water pressure trend data within the time window length to generate hidden state data including hidden state dimensions.

4. An intelligent pre-warning method for tungsten mine ground pressure based on multi-source data fusion according to claim 3, characterized in that: S3 also includes calculating the ground pressure anomaly probability through the hidden state dimension in the hidden state data, combining with the input feature weights to generate a ground pressure risk index to obtain risk assessment data representing the anomaly probability in the future time period.

5. The intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion according to claim 1, wherein: In S3, use the D-S evidence theory to perform uncertainty reasoning on the preprocessed feature data from different sensors to construct a multi-parameter joint confidence.

6. The intelligent early warning method for tungsten mine ground pressure based on multi-source data fusion according to claim 5, characterized in that: The multi-parameter joint confidence is calculated based on the evidence weights corresponding to different monitoring indicators, and the evidence weights are set according to the feature importance results of the random forest model.

7. The intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion according to claim 1, characterized in that: The wavelet transform used in S2 is the Daubechies5 wavelet basis, and the decomposition level is not less than 3 layers, which is used to extract the effective frequency band of 0.1Hz - 10Hz in the ground pressure distribution data to enhance the suppression ability of blasting high-frequency interference.

8. The intelligent early warning method for tungsten mine ground pressure based on multi-source data fusion according to claim 1, characterized in that: In S2, the filtering process of the surrounding rock displacement data uses the Kalman filtering algorithm and combines an environmental temperature compensation mechanism to correct the equipment measurement error caused by thermal expansion and contraction.

9. The intelligent ground pressure early warning method for tungsten mines based on multi-source data fusion according to claim 1, wherein: When extracting the water pressure trend data in S2, a moving average filter is used with a window length set to 60 seconds to smooth the periodic disturbances and retain the mutation characteristics.

10. A method for intelligent early warning of tungsten mine ground pressure based on multi-source data fusion according to claim 1, characterized in that: For the emergency decision-making path data generated in S5, the A* optimization algorithm is used to calculate the shortest disaster avoidance path. Combining with the high-risk spatial areas in the hierarchical early warning data, the priority of the path nodes is dynamically adjusted to achieve real-time path update.

Images