Multi-source data fusion deep engineering rockburst early warning method and system adaptive to multiple processes

Abstract

The invention discloses a multi-source data fusion deep engineering rockburst early warning method and system adaptive to multiple processes, and belongs to the technical field of deep underground engineering rockburst disaster early warning. According to the early warning method, a standardized rockburst early warning special database is constructed by collecting multi-source monitoring data and rockburst case data under different construction procedures, procedure differentiation and rockburst precursor characteristics are extracted through data preprocessing, and a multi-source data fusion rockburst early warning model with a built-in procedure dynamic adaptation weight module is constructed. And finally, real-time rockburst risk grading early warning and model closed-loop optimization are realized. According to the invention, the core defects of single monitoring dimension, unadaptive construction process difference and lagging rock burst early warning blank in the prior art are solved, the precision, adaptability and engineering practicability of rock burst early warning are greatly improved, and the method is suitable for full-period rock burst early warning of underground engineering such as deep tunnels and the like.

Description

Technical Field

[0001] This invention belongs to the field of rockburst disaster early warning technology for deep underground engineering, specifically involving the design of a rockburst early warning method and system for deep engineering that adapts to multi-process multi-source data fusion. Background Technology

[0002] Currently, the scale and depth of rock engineering projects in my country, such as deep transportation tunnels, underground caverns for water conservancy and hydropower, and deep mineral mining, are continuously increasing. Deep rock masses are generally located in complex geological environments characterized by high ground stress, high ground temperature, high karst water pressure, and strong excavation disturbance. Rockburst disasters occur frequently and their intensity continues to rise. Their suddenness, instantaneity, and strong destructiveness directly threaten the lives of construction workers, the safety of equipment and property, and the progress of project construction, and have become the core problem restricting the safe and efficient construction of deep underground engineering projects.

[0003] Rockburst early warning is a core component of disaster prevention and mitigation systems for deep engineering. Currently, the industry has conducted extensive research on technologies such as stress monitoring, deformation monitoring, and microseismic monitoring, resulting in a series of single-dimensional rockburst early warning solutions that have achieved preliminary predictions of rockburst hazards in specific scenarios. However, in practical engineering applications, existing technologies still have significant shortcomings, making it difficult to meet the full-cycle, high-precision rockburst prevention and control needs of complex and dynamic construction scenarios in deep engineering.

[0004] Most existing technologies employ a single monitoring dimension, making it difficult to comprehensively capture the multi-dimensional precursor characteristics of rockbursts, resulting in high rates of missed and false alarms. Furthermore, they fail to consider the differences in rockburst occurrence patterns and risk characteristics across different tunnel construction stages, using fixed and uniform warning thresholds and feature weights, leading to a disconnect between warning results and actual on-site risks, thus limiting their engineering practicality. In addition, existing technologies primarily focus on instantaneous rockbursts during excavation, lacking targeted warning schemes for delayed rockbursts with highly concealed precursors and uncertain occurrence times, creating significant technical blind spots and failing to achieve full-coverage early warning for rockbursts in deep engineering. Most existing rockburst early warning technologies are only designed for instantaneous rockbursts during excavation, lacking a long-term evolution precursor identification system for delayed rockbursts, resulting in significant warning blind spots.

[0005] In summary, existing rockburst early warning technologies are insufficient to meet the safety and control requirements of complex construction scenarios in deep engineering. The industry urgently needs a high-precision intelligent early warning technology that can adapt to multiple construction procedures, integrate multi-source monitoring data, and cover rockbursts. Summary of the Invention

[0006] The purpose of this invention is to address the problems of existing rockburst early warning technologies, such as limited monitoring dimensions, lack of adaptation to differences in construction procedures, and the gap in delayed rockburst early warning. This invention proposes a multi-source data fusion method and system for rockburst early warning in deep engineering, adapting to multiple construction procedures. With differentiated adaptation to construction procedures as its core, it achieves complementary fusion of multi-source monitoring data, accurate identification of rockburst precursors, and dynamic adaptation of early warning for different construction scenarios. This significantly reduces the false alarm and missed alarm rates of rockburst early warning, fills the technological gap in delayed rockburst early warning, and improves the accuracy and engineering practicality of rockburst early warning in deep engineering.

[0007] The technical solution of the present invention is as follows: In a first aspect, the present invention provides a method for early warning of rockburst in deep engineering by fusing multi-source data adapted to multiple processes, comprising the following steps: S1. Collect multi-source monitoring data and rockburst case data under different construction procedures of deep engineering, archive and divide the datasets according to the preset structure, and build a standardized rockburst early warning special database.

[0008] S2. Noise reduction, spatiotemporal registration and standardization preprocessing are performed on the multi-source monitoring data in the standardized rockburst early warning database. Based on the preprocessed data, differentiated rockburst precursor features adapted to different construction procedures and rockburst precursor features covering instantaneous and delayed rockbursts are extracted to obtain a standardized multi-source feature dataset.

[0009] S3. Construct a multi-source data fusion rockburst early warning model adapted to multiple processes. Use a standardized multi-source feature dataset as the core input and construction process feature parameters as auxiliary input. Supervised training and optimization of the early warning model are performed to obtain the trained early warning model.

[0010] S4. Real-time acquisition of multi-source monitoring data and current construction process information from deep engineering sites, inputting the pre-processed data into the trained early warning model, outputting rockburst risk classification and early warning results, and transmitting the rockburst risk classification and early warning results back to the standardized rockburst early warning database to achieve closed-loop iterative optimization.

[0011] Furthermore, the multi-source monitoring data in step S1 includes full-cycle monitoring data in three dimensions: stress, microstrain, and vibration.

[0012] The stress monitoring data includes the rate of change of dynamic stress in the surrounding rock, stress concentration factor, and stress disturbance amplitude, which are collected in real time by passive wireless stress sensors.

[0013] The micro-strain monitoring data includes the time series values ​​of micro-strain across the entire cross section of the surrounding rock, strain accumulation rate, strain spatial distribution characteristics, and local strain anomalies, all acquired through a passive wireless stress sensor.

[0014] Vibration monitoring data includes the frequency, energy, spectral jitter characteristics, source location, and event evolution of rock mass fracture vibration signals acquired through passive wireless stress sensors.

[0015] Furthermore, the construction procedures in step S1 include four core procedures for deep tunnel engineering: ventilation and slag removal, hazard removal and support, drilling and charging, and blasting. When collecting the three-dimensional monitoring data corresponding to each procedure, the operating status of the construction equipment, the work area and personnel and equipment distribution information of the corresponding procedure are collected simultaneously, as well as the rock burst occurrence records under the corresponding procedure. The rock burst occurrence records include the rock burst occurrence time, intensity and frequency.

[0016] Furthermore, the rockburst case data in step S1 includes instantaneous rockburst case data throughout the entire tunnel excavation process, as well as full-cycle monitoring data of the surrounding rock and delayed rockburst case data within 72 hours after excavation. The standardized rockburst early warning database archives data according to a four-dimensional structure of construction procedures, monitoring dimensions, time series, and rockburst event tags, and divides it into training set, validation set, and test set in a 7:2:1 ratio.

[0017] Furthermore, in step S2, wavelet transform algorithm is used to denoise the original multi-source monitoring data, removing interference signals from construction equipment vibration and environmental noise, and completing outlier removal and missing value completion. The discrete expression for wavelet transform denoising is: in Represents wavelet coefficients, Indicates the scale factor. Indicates the translation factor. Represents the original discrete signal. Indicates the discrete-time sampling point index. This represents the wavelet basis function.

[0018] Spatiotemporal registration uses the GPS and BeiDou timing systems as a reference to perform time synchronization calibration on multi-source monitoring data from different sensors, sampling frequencies, and processes. At the same time, spatial positioning alignment is completed based on the tunnel excavation face mileage to ensure the comparability of multi-source monitoring data in the same time and space.

[0019] The min-max normalization method is used to standardize and preprocess multi-source monitoring data, eliminating dimensional differences between data of different dimensions. The min-max normalization calculation formula is as follows: in This represents the normalized data value. Represents the original data value. and These represent the minimum and maximum values ​​of the data in this dimension, respectively.

[0020] Furthermore, the method for extracting the differentiated rockburst precursor features in step S2 is as follows: for the rockburst occurrence patterns of different construction procedures, extract the stress mutation features, micro-strain accumulation features and vibration signal spectrum features under the corresponding procedures, and eliminate the feature interference brought about by the operation itself.

[0021] The method for extracting precursor features of rockbursts is as follows: for instantaneous rockbursts, extract short-term abrupt precursor features; for delayed rockbursts, extract long-term evolution precursor features.

[0022] Furthermore, the multi-source data fusion rockburst early warning model in step S3 includes a multi-source feature fusion backbone module, a construction procedure dynamic adaptation weight module, and a rockburst early warning decision module.

[0023] The multi-source feature fusion backbone module is constructed based on an attention mechanism and a multi-scale temporal feature extraction network. It is used to automatically learn the feature weights of monitoring data in three dimensions: stress, microstrain, and vibration, and to complete the complementary fusion of multi-dimensional precursor features. The feature fusion weight calculation formula of the multi-source feature fusion backbone module is as follows: in Indicates the first Attention weights for each feature This represents the scoring function. Indicates the first One original feature, Indicates the first One original feature, Represents the total number of features. This indicates the features after fusion.

[0024] The construction process dynamic adaptation weight module takes the characteristic parameters of the construction process as input, and the dynamic adjustment formula for the process weight of the construction process dynamic adaptation weight module is as follows: in Indicate process Next The adaptation weights of each feature, This represents the scoring function for the attention mechanism. Represents the basic weights of the features. This represents the Sigmoid activation function. and These are the network training parameters.

[0025] The rockburst early warning decision module is built on a multi-task learning network to simultaneously identify the precursor features of instantaneous and delayed rockbursts and output rockburst risk prediction results.

[0026] The supervised training is labeled with the time window period, intensity level and occurrence probability of rockbursts. The model hyperparameters are optimized through the validation set and the early warning accuracy of the model is verified through the test set.

[0027] Furthermore, the rockburst risk classification and early warning results in step S4 include the rockburst occurrence time window, rockburst intensity level, rockburst occurrence probability and risk level.

[0028] The time window for rockburst occurrence covers the risk persistence window within 72 hours after lagging rockburst excavation.

[0029] Rockburst intensity is classified into four levels according to industry standards.

[0030] The probability of a rock eruption is quantified using a sigmoid function and divided into four risk levels, as shown in the formula: in Indicates the probability of a rock eruption. This represents the output value of the fully connected layer of the model. Represents the Sigmoid activation function, when The risk level is low when the percentage is less than 30%, and ≤30% is considered a risk level. The risk level is medium when the risk is less than 60%, and 60% or less... The risk level is high when the percentage is less than 90%. When the risk level is ≥90%, the risk level is extremely high.

[0031] Secondly, the present invention provides a multi-source data fusion deep engineering rockburst early warning system adapted to multiple processes, including a multi-dimensional sensor array, an edge computing gateway, a cloud server, and an early warning terminal that enable real-time data transmission between each other via 5G or fiber optic communication links.

[0032] A multi-dimensional sensor array is used to collect real-time monitoring data on three dimensions: stress, micro-strain, and vibration in deep engineering sites.

[0033] Edge computing gateways are deployed at engineering sites for preprocessing of on-site collected data, real-time identification of construction procedures, local data caching, and encrypted transmission.

[0034] The cloud server is equipped with the aforementioned multi-source data fusion method for deep engineering rockburst early warning, which is adapted to multiple processes and is used for standardized database storage, early warning model training and updating, multi-source data fusion calculation, and early warning decision logic execution.

[0035] The early warning terminal is used for the simultaneous release of tiered early warning information through multiple channels and the issuance of emergency instructions.

[0036] Furthermore, the multi-dimensional sensor array includes multiple sets of sensors for detecting different rock layer depths and stresses in different radial and axial directions.

[0037] The edge computing gateway has a built-in construction process identification unit, which can obtain the current construction process information in real time by accessing the construction log and the equipment operation system.

[0038] The cloud server includes a database module, a model training module, an early warning calculation module, and an iterative optimization module, which respectively implement data management, model training, early warning calculation, and closed-loop optimization functions.

[0039] The early warning terminals include tunnel-mounted audible and visual alarms, emergency broadcasts, on-site monitoring platforms, and mobile apps.

[0040] The beneficial effects of this invention are: (1) This invention constructs a complementary monitoring system with three dimensions: stress, microstrain, and vibration. By integrating multi-source data, it deeply mines multi-dimensional precursor features, solving the pain points of existing technologies with single monitoring dimensions and insufficient ability to capture rockburst precursors, and significantly reducing the false alarm rate and false alarm rate of rockburst early warning.

[0041] (2) This invention takes the differentiated adaptation of construction procedures as the core line, embeds the procedure adaptation logic into the entire process from data collection and feature extraction to model early warning. Through the dynamic adaptation weight module of construction procedures, the feature weights and early warning thresholds under different procedures are automatically adjusted. Customized early warning is achieved for high-occurrence and high-hazard procedures. This solves the defects of existing technologies such as fixed threshold early warning, failure to adapt to differences in construction scenarios, and insufficient early warning targeting. It significantly improves the matching degree between early warning results and actual engineering.

[0042] (3) This invention designs special precursor feature extraction logic and identification scheme for instantaneous rockbursts and delayed rockbursts respectively, filling the technical gap of delayed rockburst early warning in the prior art, realizing full-cycle rockburst early warning for deep engineering, and fully ensuring the safety of personnel and equipment throughout the construction process.

[0043] (4) This invention constructs a complete technical process from data acquisition and model training to real-time early warning and closed-loop iteration, and designs an early warning system that can be engineered and implemented. All monitoring equipment and technologies have been engineered and applied, and have strong on-site adaptability and promotion value. They can be directly applied to rockburst disaster prevention and control in various deep rock engineering projects such as deep tunnels and underground caverns. Attached Figure Description

[0044] Figure 1 The diagram shown is a flowchart of a deep engineering rockburst early warning method adapted to multiple processes and using multi-source data fusion, provided in Embodiment 1 of the present invention.

[0045] Figure 2The diagram shown is an architecture diagram of a multi-source data fusion deep engineering rockburst early warning system adapted to multiple processes, provided in Embodiment 1 of the present invention. Detailed Implementation

[0046] Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, and are not intended to limit the scope of the invention.

[0047] Example 1: This invention provides a method for early warning of rockbursts in deep engineering through multi-source data fusion adapted to multiple processes, such as... Figure 1 As shown, it includes the following steps S1~S4: S1. Collect multi-source monitoring data and rockburst case data under different construction procedures of deep engineering, archive and divide the datasets according to the preset structure, and build a standardized rockburst early warning special database.

[0048] In this embodiment of the invention, the multi-source monitoring data includes full-cycle monitoring data in three dimensions: stress, micro-strain, and vibration.

[0049] The stress monitoring data includes the rate of change of dynamic stress in the surrounding rock, stress concentration factor, and stress disturbance amplitude, which are collected in real time by passive wireless stress sensors.

[0050] The micro-strain monitoring data includes the time series values ​​of micro-strain across the entire cross section of the surrounding rock, strain accumulation rate, strain spatial distribution characteristics, and local strain anomalies, all acquired through a passive wireless stress sensor.

[0051] Vibration monitoring data includes the frequency, energy, spectral jitter characteristics, source location, and event evolution of rock mass fracture vibration signals acquired through passive wireless stress sensors.

[0052] In this embodiment of the invention, the construction process includes four core processes for deep tunnel engineering: ventilation and muck removal, hazard removal and support, drilling and charging, and blasting. When collecting monitoring data from three dimensions corresponding to each process, the operating status of construction equipment, the work area, and the distribution of personnel and equipment for the corresponding process are simultaneously collected, along with rockburst occurrence records for the corresponding process. This allows for the differentiation of rockburst patterns across different processes from the data source. The rockburst occurrence records include the time, intensity, and frequency of rockburst occurrences.

[0053] In this embodiment of the invention, the rockburst case data includes instantaneous rockburst case data throughout the entire tunnel excavation process, as well as full-cycle monitoring data of the surrounding rock within 72 hours after excavation and delayed rockburst occurrence case data. The data labels the precursor characteristics and temporal evolution patterns of the corresponding rockbursts, achieving full coverage of both instantaneous and delayed rockburst data. The rockburst case data is used to provide rockburst event labeling information, supporting subsequent feature extraction, model training, and accuracy verification.

[0054] In this embodiment of the invention, the standardized rockburst early warning database archives data according to a four-dimensional structure of construction procedures, monitoring dimensions, time series, and rockburst event tags. Invalid, duplicate, and abnormal data are removed using the 3σ criterion, and the database is divided into training set, validation set, and test set in a 7:2:1 ratio to form the standardized rockburst early warning database.

[0055] S2. Noise reduction, spatiotemporal registration and standardization preprocessing are performed on the multi-source monitoring data in the standardized rockburst early warning database. Based on the preprocessed data, differentiated rockburst precursor features adapted to different construction procedures and rockburst precursor features covering instantaneous and delayed rockbursts are extracted to obtain a standardized multi-source feature dataset.

[0056] In this embodiment of the invention, the db4 wavelet transform algorithm is used to perform five-level decomposition and reconstruction noise reduction on the original multi-source monitoring data, removing interference signals from construction equipment vibration and environmental noise, while retaining effective characteristic signals of rockburst precursors. Simultaneously, outlier removal is performed, and missing values ​​are filled in using linear interpolation, thus achieving data cleaning. The discrete expression for wavelet transform noise reduction is as follows: in Represents wavelet coefficients, Indicates the scale factor. Indicates the translation factor. Represents the original discrete signal. Indicates the discrete-time sampling point index. This represents the wavelet basis function.

[0057] Spatiotemporal registration uses the GPS and BeiDou timing systems as a reference to perform time synchronization calibration on multi-source monitoring data from different sensors, sampling frequencies, and processes. At the same time, spatial positioning alignment is completed based on the tunnel excavation face mileage to ensure the comparability of multi-source monitoring data in the same time and space.

[0058] The min-max normalization method is used to standardize and preprocess multi-source monitoring data, eliminating dimensional differences between data of different dimensions. The min-max normalization calculation formula is as follows: in This represents the normalized data value. Represents the original data value. and These represent the minimum and maximum values ​​of the data in this dimension, respectively.

[0059] In this embodiment of the invention, the method for extracting differentiated rockburst precursor features is as follows: for the rockburst occurrence patterns of different construction procedures, stress mutation features, micro-strain accumulation features, and vibration signal spectrum features under the corresponding procedures are extracted respectively. The feature interference generated by the operation itself (such as vibration signals of blasting operations and strain disturbances of support equipment operations) is filtered out by the process background signal elimination algorithm, thereby reducing false alarms caused by the operation.

[0060] The method for extracting precursor features of rockburst is as follows: for instantaneous rockbursts, short-term abrupt change precursor features such as stress change rate, strain abrupt change amount, and high-frequency energy ratio of vibration signal are extracted; for delayed rockbursts, long-term evolution precursor features such as strain accumulation rate, stress-strain hysteresis coefficient, and low-frequency component ratio of vibration signal are extracted. This solves the problem of delayed rockburst precursors being hidden and difficult to identify, and finally obtains a standardized multi-source feature dataset.

[0061] S3. Construct a multi-source data fusion rockburst early warning model adapted to multiple processes. Use a standardized multi-source feature dataset as the core input and construction process feature parameters as auxiliary input. Supervised training and optimization of the early warning model are performed to obtain the trained early warning model.

[0062] In this embodiment of the invention, the multi-source data fusion rockburst early warning model includes a multi-source feature fusion backbone module, a construction procedure dynamic adaptation weight module, and a rockburst early warning decision module.

[0063] The multi-source feature fusion backbone module is constructed based on an attention mechanism and a multi-scale temporal feature extraction network. The input is preprocessed rockburst precursor features, which are used to automatically learn the feature weights of monitoring data in three dimensions: stress, micro-strain, and vibration. This completes the complementary fusion of multi-dimensional precursor features. The feature fusion weight calculation formula for the multi-source feature fusion backbone module is as follows: in Indicates the first Attention weights for each feature This represents the scoring function. Indicates the first One original feature, Indicates the first One original feature, Represents the total number of features. This indicates the features after fusion.

[0064] The construction process dynamic adaptation weight module takes the construction process feature parameters (one-hot encoded vectors) as input and automatically learns the warning thresholds and weight coefficients for each feature dimension under different processes. This enables dynamic adaptation of the warning logic to the construction scenario: For the hazard mitigation and support process, it automatically increases the weights of micro-strain and vibration features, reduces the warning threshold, and improves warning sensitivity; for the drilling and explosive loading process, it automatically increases the weights of stress mutation features, extends the warning lead time, and ensures the evacuation window for personnel and equipment. The dynamic adjustment formula for the process weights in the construction process dynamic adaptation weight module is as follows: in Indicate process Next The adaptation weights of each feature, This represents the scoring function for the attention mechanism. Represents the basic weights of the features. This represents the Sigmoid activation function. and These are the network training parameters.

[0065] The rockburst early warning decision module is built on a multi-task learning network. It takes the fused features output by the multi-source feature fusion backbone module and the weight parameters output by the construction procedure dynamic adaptation weight module as inputs. It simultaneously sets two sub-tasks: instantaneous rockburst identification and delayed rockburst identification, so as to realize the synchronous and accurate identification of the two types of rockburst precursor features and output the rockburst risk prediction results.

[0066] In this embodiment of the invention, supervised training and optimization of the model are as follows: Based on the standardized rockburst early warning database constructed in step S1, the standardized multi-source feature dataset extracted in step S2 is used as input, and the time window period, intensity level, and probability of rockburst occurrence are used as labels. The cross-entropy loss function is used to supervise the training of the model. The model learning rate, batch size, time window length, and other hyperparameters are optimized through the validation set, and the early warning accuracy of the model is verified through the test set. Finally, a multi-source data fusion rockburst early warning model that is adapted to multiple processes is obtained after training.

[0067] S4. Real-time acquisition of multi-source monitoring data and current construction process information from deep engineering sites, inputting the pre-processed data into the trained early warning model, outputting rockburst risk classification and early warning results, and transmitting the rockburst risk classification and early warning results back to the standardized rockburst early warning database to achieve closed-loop iterative optimization.

[0068] In this embodiment of the invention, the rockburst risk classification and early warning results in step S4 include the rockburst occurrence time window, rockburst intensity level, rockburst occurrence probability and risk level.

[0069] Among them, the time window for rockburst occurrence covers the risk duration window within 72 hours after the excavation of delayed rockburst, which is divided into medium- and long-term trend prediction (3 to 7 days in the future), short-term warning (within the next 2 hours), and emergency warning for impending disaster (within the next 2 hours).

[0070] Rockburst intensity is classified into four levels according to domestic geotechnical engineering industry standards: Level I (minor rockburst), Level II (moderate rockburst), Level III (strong rockburst), and Level IV (violent rockburst).

[0071] The probability of a rock eruption is quantified using a sigmoid function and divided into four risk levels, as shown in the formula: in Indicates the probability of a rock eruption. This represents the output value of the fully connected layer of the model. Represents the Sigmoid activation function, when The risk level is low when the percentage is less than 30%, and ≤30% is considered a risk level. The risk level is medium when the risk is less than 60%, and 60% or less... The risk level is high when the percentage is less than 90%. When the risk level is ≥90%, the risk level is extremely high.

[0072] In this embodiment of the invention, the early warning results are simultaneously released through the early warning terminal, and corresponding emergency response plans are matched for different risk levels. The actual feedback results of the on-site early warning (whether a rockburst occurred, the actual rockburst intensity, the occurrence time, etc.) are transmitted back to the standardized rockburst early warning database constructed in step S1. The model is retrained and its parameters are optimized every 3 months to continuously improve the model's early warning accuracy and adaptability to working conditions.

[0073] Example 2: This invention provides a multi-source data fusion deep engineering rockburst early warning system adapted to multiple processes, such as... Figure 2 As shown, it includes a multi-dimensional sensor array, an edge computing gateway, a cloud server, and an early warning terminal that enable real-time data transmission between each other via 5G or fiber optic communication links.

[0074] The multi-dimensional sensor array is used to collect real-time monitoring data on stress, micro-strain, and vibration in deep engineering sites. In this embodiment of the invention, the multi-dimensional sensor array is deployed at the tunnel face and the surrounding rock area behind it, and includes multiple sets of sensors for detecting different rock layer depths and stresses in different radial and axial directions.

[0075] An edge computing gateway is deployed in the equipment room at the entrance of the construction site for preprocessing data collected on-site, real-time identification of construction procedures, local data caching, and encrypted transmission. In this embodiment of the invention, the edge computing gateway has a built-in data preprocessing unit and a construction procedure identification unit. The data preprocessing unit is used to perform real-time noise reduction, standardization, and feature extraction of the data collected on-site. The construction procedure identification unit obtains the current construction procedure information in real time by accessing the construction log system and the construction equipment operation system and completes one-hot encoding. At the same time, the edge computing gateway is responsible for local data caching and encrypted transmission, enabling rapid front-end processing of early warning data, reducing cloud computing pressure, and ensuring the timeliness of early warnings.

[0076] The cloud server serves as the core computing hub of the system, incorporating the multi-source data fusion deep engineering rockburst early warning method adapted to multiple processes provided in Embodiment 1. This method is used for standardized database storage, early warning model training and updating, multi-source data fusion calculation, and execution of early warning decision logic. In this embodiment, the cloud server includes a database module, a model training module, an early warning calculation module, and an iterative optimization module. The database module stores and manages a standardized rockburst early warning database; the model training module performs supervised training, hyperparameter optimization, and accuracy verification of the early warning model; the early warning calculation module incorporates the multi-source data fusion rockburst early warning model adapted to multiple processes trained in Embodiment 1, performing multi-source data fusion calculation, dynamic adaptation of process weights, identification of rockburst precursor features, and output of early warning results; the iterative optimization module receives on-site early warning feedback results and drives periodic retraining and parameter updates of the model.

[0077] The early warning terminal is used for the simultaneous multi-channel dissemination of tiered early warning information and the issuance of emergency commands. In this embodiment of the invention, the early warning terminal includes a tunnel-mounted audible and visual alarm, an emergency broadcast system, an on-site monitoring platform, and a mobile app. The tunnel-mounted audible and visual alarm and emergency broadcast system are used to disseminate pre-disaster early warning signals in real time. The on-site monitoring platform is used by management personnel to view rockburst risk monitoring data and early warning results in real time. The mobile app is used by on-site construction personnel and management personnel to receive early warning information and emergency response commands in real time, realizing the simultaneous multi-channel dissemination of tiered early warning information and ensuring the rapid delivery of early warning commands.

[0078] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A rockburst early warning method for deep engineering that adapts to multi-process, multi-source data fusion, characterized in that: Includes the following steps: S1. Collect multi-source monitoring data and rockburst case data under different construction procedures of deep engineering, archive and divide the datasets according to the preset structure, and build a standardized rockburst early warning special database. S2. The multi-source monitoring data in the standardized rockburst early warning database is subjected to noise reduction, spatiotemporal registration and standardized preprocessing. Based on the preprocessed data, differentiated rockburst precursor features adapted to different construction procedures and rockburst-specific precursor features covering instantaneous and delayed rockbursts are extracted to obtain a standardized multi-source feature dataset. S3. Construct a multi-source data fusion rockburst early warning model adapted to multiple processes. Using the standardized multi-source feature dataset as the core input and the construction process feature parameters as auxiliary input, supervise the training and optimization of the early warning model to obtain the trained early warning model. S4. Real-time collection of multi-source monitoring data and current construction process information from the deep engineering site, inputting the pre-processed data into the trained early warning model, outputting rockburst risk classification early warning results, and transmitting the rockburst risk classification early warning results back to the standardized rockburst early warning database to achieve closed-loop iterative optimization.

2. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 1, characterized in that, The multi-source monitoring data in step S1 includes full-cycle monitoring data in three dimensions: stress, microstrain, and vibration. Stress monitoring data includes the rate of change of dynamic stress in the surrounding rock, stress concentration factor, and stress disturbance amplitude, which are collected in real time by passive wireless stress sensors. Microstrain monitoring data includes time-series values ​​of microstrain across the entire cross section of the surrounding rock, strain accumulation rate, strain spatial distribution characteristics, and local strain anomalies, all acquired through passive wireless stress sensors. Vibration monitoring data includes the frequency, energy, spectral jitter characteristics, source location, and event evolution of rock mass fracture vibration signals acquired through passive wireless stress sensors.

3. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 2, characterized in that, The construction procedures in step S1 include four core procedures for deep tunnel engineering: ventilation and muck removal, hazard removal and support, drilling and charging, and blasting. When collecting the three-dimensional monitoring data corresponding to each procedure, the operating status of the construction equipment, the work area and personnel and equipment distribution information of the corresponding procedure are collected simultaneously, as well as the rock burst occurrence record under the corresponding procedure. The rock burst occurrence record includes the rock burst occurrence time, intensity and frequency.

4. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 1, characterized in that, The rockburst case data in step S1 includes instantaneous rockburst case data throughout the entire tunnel excavation process, as well as surrounding rock full-cycle monitoring data and delayed rockburst occurrence case data within 72 hours after excavation. The standardized rockburst early warning database archives data according to a four-dimensional structure of construction procedures, monitoring dimensions, time series, and rockburst event tags, and divides it into training set, validation set, and test set in a 7:2:1 ratio.

5. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 1, characterized in that, In step S2, wavelet transform algorithm is used to denoise the original multi-source monitoring data, removing interference signals from construction equipment vibration and environmental noise, and completing outlier removal and missing value completion. The discrete expression for wavelet transform denoising is: in Represents wavelet coefficients, Indicates the scale factor. Indicates the translation factor. Represents the original discrete signal. Indicates the discrete-time sampling point index. Describe the wavelet basis functions; The spatiotemporal registration is based on the GPS and Beidou time synchronization system to perform time synchronization calibration on multi-source monitoring data from different sensors, different sampling frequencies and different processes. At the same time, spatial positioning alignment is completed based on the tunnel excavation face mileage to ensure the comparability of multi-source monitoring data in the same spatiotemporal space. The min-max normalization method is used to standardize and preprocess multi-source monitoring data, eliminating dimensional differences between data of different dimensions. The min-max normalization calculation formula is as follows: in This represents the normalized data value. Represents the original data value. and These represent the minimum and maximum values ​​of the data in this dimension, respectively.

6. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 1, characterized in that, The method for extracting the differentiated rockburst precursor features in step S2 is as follows: for the rockburst occurrence patterns of different construction procedures, extract the stress mutation features, micro-strain accumulation features and vibration signal spectrum features under the corresponding procedures, and eliminate the feature interference brought about by the operation itself. The method for extracting precursor features of rockburst is as follows: for instantaneous rockbursts, extract short-term abrupt change precursor features; for delayed rockbursts, extract long-term evolution precursor features.

7. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 4, characterized in that, The multi-source data fusion rockburst early warning model in step S3 includes a multi-source feature fusion backbone module, a construction procedure dynamic adaptation weight module, and a rockburst early warning decision module. The multi-source feature fusion backbone module is constructed based on an attention mechanism and a multi-scale temporal feature extraction network. It is used to automatically learn the feature weights of monitoring data in three dimensions: stress, microstrain, and vibration, and to complete the complementary fusion of multi-dimensional precursor features. The feature fusion weight calculation formula of the multi-source feature fusion backbone module is as follows: in Indicates the first Attention weights for each feature This represents the scoring function. Indicates the first One original feature, Indicates the first One original feature, Represents the total number of features. Indicates the features after fusion; The construction process dynamic adaptation weight module takes the construction process characteristic parameters as input, and the dynamic adjustment formula for the process weight of the construction process dynamic adaptation weight module is as follows: in Indicate process Next The adaptation weights of each feature, This represents the scoring function for the attention mechanism. Represents the basic weights of the features. This represents the Sigmoid activation function. and These are the network training parameters; The rockburst early warning decision module is built on a multi-task learning network and is used to simultaneously identify the precursor features of instantaneous and delayed rockbursts and output rockburst risk prediction results. The supervised training is labeled with the time window period, intensity level and probability of rockburst occurrence. The model hyperparameters are optimized through the validation set and the early warning accuracy of the model is verified through the test set.

8. The method for early warning of rockburst in deep engineering by multi-source data fusion adapted to multiple processes as described in claim 1, characterized in that, The rockburst risk classification and early warning results in step S4 include the rockburst occurrence time window, rockburst intensity level, rockburst occurrence probability and risk level. The rockburst occurrence time window covers the risk duration window within 72 hours after delayed rockburst excavation. The rockburst intensity level is classified into four levels according to industry standards; The probability of a rockburst occurrence is quantified using a Sigmoid function and divided into four risk levels, as shown in the formula: in Indicates the probability of a rock eruption. This represents the output value of the fully connected layer of the model. Represents the Sigmoid activation function, when The risk level is low when the percentage is less than 30%, and ≤30% is considered a risk level. The risk level is medium when the risk is less than 60%, and 60% or less... The risk level is high if the percentage is less than 90%. When the risk level is ≥90%, the risk level is extremely high.

9. A deep engineering rockburst early warning system adapted to multi-process, multi-source data fusion, characterized in that: This includes multi-dimensional sensor arrays, edge computing gateways, cloud servers, and early warning terminals that enable real-time data transmission between each other via 5G or fiber optic communication links; The multi-dimensional sensor array is used to collect real-time monitoring data on three dimensions: stress, micro-strain, and vibration at the deep engineering site. The edge computing gateway is deployed at the engineering site for preprocessing of on-site collected data, real-time identification of construction procedures, local data caching, and encrypted transmission. The cloud server is built with the multi-source data fusion deep engineering rockburst early warning method as described in any one of claims 1-8, which is used for standardized database storage, early warning model training and updating, multi-source data fusion calculation and early warning decision logic execution. The early warning terminal is used for the simultaneous release of tiered early warning information through multiple channels and the issuance of emergency commands.

10. The multi-source data fusion deep engineering rockburst early warning system adapted to multiple processes as described in claim 9, characterized in that, The multi-dimensional sensor array includes multiple sets of sensors for detecting different rock layer depths and stresses in different radial and axial directions; The edge computing gateway has a built-in construction process identification unit, which can obtain the current construction process information in real time by accessing the construction log and the equipment operation system. The cloud server includes a database module, a model training module, an early warning calculation module, and an iterative optimization module, which respectively implement data management, model training, early warning calculation, and closed-loop optimization functions. The early warning terminal includes an audible and visual alarm in the tunnel, an emergency broadcast system, an on-site monitoring platform, and a mobile app.

Images