Group effect-based underground cavern disaster prevention and control method, program product and equipment

Abstract

The disclosure provides a group effect-based underground cavern disaster prevention and control method, program product and equipment, relating to the technical field of engineering data processing. The method is applied to a group of underground caverns of a water conservancy project, comprising: in the case of considering the influence between caverns and not considering the influence between caverns, respectively acquiring a first parameter of a target cavern, and determining a static group effect coefficient; determining a dynamic group effect coefficient according to the transmission between multiple disasters; obtaining a comprehensive group effect coefficient by combining the static group effect coefficient and the dynamic group effect coefficient; acquiring initial monitoring data of the target cavern, and obtaining fused monitoring data through data fusion processing; performing feature extraction on the comprehensive group effect coefficient and the fused monitoring data to obtain a target feature; obtaining an evaluation result of the target cavern based on the target feature, and determining a disaster prevention and control measure according to the evaluation result. The disclosure realizes group effect-oriented underground cavern disaster prevention and control based on intelligent algorithm, and improves the accuracy and effectiveness of prevention and control.

Description

Technical Field

[0001] This disclosure relates to the field of engineering data processing technology, and more specifically, to methods, program products and equipment for disaster prevention and control of underground caverns based on swarm effects. Background Technology

[0002] In large-scale water conservancy projects such as hydropower stations, it is usually necessary to design and construct underground cavern complexes to serve as core functional spaces such as main powerhouses, auxiliary powerhouses, surge tanks, and water diversion tunnels. Underground caverns are prone to disasters such as surrounding rock cracking, block collapse, and water and mud inrush. Therefore, it is necessary to assess the condition of underground caverns in order to prevent and control potential disaster risks.

[0003] In related technologies, most disaster prevention and control schemes for underground caverns adopt the approach of independent prevention and control for single caverns, ignoring the group effect between caverns. This approach cannot adapt to the clustered layout characteristics of underground cavern groups, resulting in cavern assessment results deviating from the actual situation, and corresponding prevention and control measures failing to match the intensity of the group effect. Summary of the Invention

[0004] This disclosure provides a method, device, program product, and electronic equipment for disaster prevention and control of underground caverns based on swarm effects, so as to at least partially solve the technical problem that disaster prevention and control measures for underground caverns cannot match the intensity of swarm effects.

[0005] According to a first aspect of this disclosure, a method for disaster prevention and control of underground caverns based on swarm effects is provided, applied to underground cavern groups in water conservancy projects, wherein the underground cavern group includes a target cavern; the method includes: obtaining first parameters of the target cavern under two conditions, considering inter-cave influence and not considering inter-cave influence, and determining a static swarm effect coefficient based on the first parameters under the two conditions; determining a dynamic swarm effect coefficient based on the transmission between various disasters that may occur in the underground cavern group; obtaining a comprehensive swarm effect coefficient by combining the static swarm effect coefficient and the dynamic swarm effect coefficient; obtaining initial monitoring data of the target cavern and obtaining fused monitoring data through data fusion processing; extracting features from the comprehensive swarm effect coefficient and the fused monitoring data to obtain target features; obtaining an evaluation result of the target cavern based on the target features; and determining disaster prevention and control measures based on the evaluation result.

[0006] According to a second aspect of this disclosure, a disaster prevention and control device for underground caverns based on swarm effects is provided, applied to underground cavern groups in water conservancy projects, wherein the underground cavern group includes a target cavern. The device includes: a static swarm effect processing module configured to acquire first parameters of the target cavern under two conditions: considering inter-cave influence and not considering inter-cave influence, and to determine a static swarm effect coefficient based on the first parameters under the two conditions; a dynamic swarm effect processing module configured to determine a dynamic swarm effect coefficient based on the transmission between various possible disasters in the underground cavern group; a comprehensive swarm effect processing module configured to obtain a comprehensive swarm effect coefficient by combining the static swarm effect coefficient and the dynamic swarm effect coefficient; a monitoring data processing module configured to acquire initial monitoring data of the target cavern and obtain fused monitoring data through data fusion processing; a feature processing module configured to extract features from the comprehensive swarm effect coefficient and the fused monitoring data to obtain target features; and a prevention and control measure determination module configured to obtain an evaluation result of the target cavern based on the target features and to determine disaster prevention and control measures based on the evaluation result.

[0007] According to a third aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of the first aspect described above and possible implementations thereof.

[0008] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of the first aspect and possible implementations thereof by executing the executable instructions.

[0009] The technical solution disclosed herein has the following beneficial effects:

[0010] On the one hand, an intelligent algorithm-based disaster prevention and control scheme for underground caverns, guided by swarm effects, is provided. This breaks away from the approach of independent prevention and control for single caverns in related technologies, integrating swarm effects throughout the entire disaster prevention and control process for underground caverns. This adapts to the clustered layout characteristics of underground cavern groups, ensuring that cavern assessment results closely approximate actual conditions, and that corresponding prevention and control measures match the intensity of swarm effects. This improves the accuracy and effectiveness of prevention and control. On the other hand, by quantifying and fusing static and dynamic swarm effects to obtain a comprehensive swarm effect coefficient, and extracting features from this coefficient along with monitoring data, the assessment results for the target cavern and the corresponding disaster prevention and control measures are determined based on the obtained target features. This improves the accuracy and effectiveness of prevention and control, and is adaptable to engineering scenarios with large cavern groups, complex structures, and numerous disaster factors, meeting the disaster prevention and control needs of large-scale water conservancy projects. Attached Figure Description

[0011] Figure 1A flowchart illustrating a method for preventing and controlling underground cavern disasters based on swarm effects is shown in this disclosure.

[0012] Figure 2 This diagram illustrates a flowchart of an embodiment of the present disclosure for obtaining fused monitoring data;

[0013] Figure 3 A flowchart illustrating one method of obtaining a target feature according to an embodiment of this disclosure is shown;

[0014] Figure 4 This diagram illustrates a model architecture according to an embodiment of the present disclosure.

[0015] Figure 5 This diagram illustrates a process architecture according to an embodiment of the present disclosure;

[0016] Figure 6 A comparison chart of the loss values ​​of the improved model and the original model in the embodiments of this disclosure is shown;

[0017] Figure 7 A schematic diagram of a group effect-based underground cavern disaster prevention and control device is shown in this embodiment of the present disclosure;

[0018] Figure 8 A schematic diagram of an electronic device according to one embodiment of the present disclosure is shown. Detailed Implementation

[0019] Exemplary embodiments of this disclosure will be described more fully below with reference to the accompanying drawings.

[0020] The accompanying drawings are schematic illustrations of this disclosure and are not necessarily drawn to scale. Some block diagrams shown in the drawings may be functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in hardware modules or integrated circuits, or in networks, processors, or microcontrollers. Implementations can be carried out in various forms and should not be construed as limited to the examples set forth herein. The features, structures, or characteristics described in this disclosure can be combined in any suitable manner in one or more implementations. In the following description, numerous specific details are provided to give a thorough description of one embodiment of this disclosure. However, those skilled in the art will recognize that one or more specific details may be omitted when implementing the technical solutions provided in an embodiment of this disclosure, or other methods, components, apparatuses, steps, etc., may be used to replace one or more specific details.

[0021] Traditional solutions in related technologies employ fixed-threshold disaster monitoring, issuing disaster warnings when relevant data from underground caverns exceeds the threshold. This approach is only suitable for simple disaster prevention and control in small to medium-sized caverns or single caverns, and is ill-suited for large cavern clusters, complex structures, and situations with multiple disaster factors. With the development of artificial intelligence, deep learning solutions have emerged in related technologies. These solutions use deep learning models to process underground cavern monitoring data to assess cavern disaster risks. While the assessment accuracy is higher than traditional solutions, they still do not consider the impact of cluster effects between caverns, making them unsuitable for the clustered layout characteristics of large cavern clusters. This results in prevention and control measures that cannot match the intensity of the cluster effect, leading to insufficient precision and effectiveness.

[0022] Furthermore, traditional solutions can only issue warnings when a disaster has already occurred or is imminent, lacking the ability to identify and predict the evolution of disaster chains in their early stages. While deep learning solutions can achieve disaster prediction to some extent, they are not deeply integrated with tiered prevention and control measures, making it difficult to directly guide on-site prevention and control implementation. This results in delays in the implementation of prevention and control measures and makes it difficult to effectively prevent the transmission and spread of disaster chains.

[0023] In view of one or more of the above-mentioned problems, this disclosure provides a method for preventing and controlling underground cavern disasters based on group effects, applicable to underground cavern groups in water conservancy projects (such as hydropower stations, dams, etc.). The underground cavern group can be a cavern group in steady-state operation after excavation is completed, or it can be a cavern group during the excavation process; this disclosure does not limit it in this way.

[0024] Figure 1 An exemplary process for disaster prevention and control in underground caverns is shown, including the following steps:

[0025] S110, under the two cases of considering the inter-cavity influence and not considering the inter-cavity influence, the first parameter of the target cavern is obtained respectively, and the static group effect coefficient is determined based on the first parameter under the two cases.

[0026] S120, the dynamic group effect coefficient is determined based on the transmission between various disasters that may occur in the underground cavern group;

[0027] S130, by combining the static group effect coefficient and the dynamic group effect coefficient, the comprehensive group effect coefficient is obtained;

[0028] S140: Acquire initial monitoring data of the target chamber, and obtain fused monitoring data through data fusion processing;

[0029] S150, feature extraction is performed on the comprehensive group effect coefficient and fused monitoring data to obtain target features;

[0030] S160: Based on the target characteristics, the assessment results of the target cavern are obtained, and disaster prevention and control measures are determined based on the assessment results.

[0031] Based on the above methods, on the one hand, an intelligent algorithm-based disaster prevention and control scheme for underground caverns is provided, breaking away from the independent prevention and control approach for single caverns in related technologies. It integrates the group effect throughout the entire disaster prevention and control process for underground caverns, adapting to the clustered layout characteristics of underground cavern groups. This ensures that cavern assessment results closely approximate actual conditions, and that corresponding prevention and control measures match the intensity of the group effect, thereby improving the accuracy and effectiveness of prevention and control. On the other hand, by quantifying and fusing static and dynamic group effects to obtain a comprehensive group effect coefficient, and extracting features from this coefficient along with monitoring data, the assessment results for the target cavern and the corresponding disaster prevention and control measures are determined based on the obtained target features. This improves the accuracy and effectiveness of prevention and control, making it suitable for engineering scenarios with large cavern groups, complex structures, and numerous disaster factors, thus meeting the disaster prevention and control needs of large-scale water conservancy projects.

[0032] The following describes, in conjunction with one or more embodiments and related accompanying drawings, Figure 1 Each step in the process will be explained in detail.

[0033] refer to Figure 1 In step S110, the first parameters of the target cavern are obtained under both cases, considering the influence between caverns and not considering the influence between caverns, and the static group effect coefficient is determined based on the first parameters under the two cases.

[0034] The target cavern is the cavern that requires risk assessment and disaster prevention, and can be any cavern in the underground cavern group. If prevention and control are required for each cavern in the underground cavern group, each cavern can be used as the target cavern, and the method of this disclosed embodiment can be used for assessment to obtain the assessment results and corresponding prevention and control measures for each cavern. For example, the underground cavern group of a large hydropower station includes eight caverns: main powerhouse, auxiliary powerhouse, surge tank, water diversion tunnel, main transformer tunnel, busbar tunnel, tailrace tunnel, and access tunnel, arranged in a cluster. Any cavern in this cluster can be used as the target cavern.

[0035] The group effect refers to the effect generated by the interaction between different caverns within an underground cavern group as a whole. In step S110, considering the influence between caverns means considering the group effect; not considering the influence between caverns means not considering the group effect. The group effect of underground caverns can be divided into static group effect and dynamic group effect. The static group effect refers to the group effect under steady-state operation (such as after the underground cavern group has completed excavation and entered the steady-state operation stage), while the dynamic group effect refers to the effect that changes in real time as the disaster evolves after an initial disaster occurs in the underground cavern group.

[0036] In this embodiment, the static group effect is quantified by the influence of the group effect on the first parameter of the target cavern. The first parameter refers to one or more parameters that may be affected by the group effect, such as mechanical parameters and deformation parameters. The first parameter of the target cavern is obtained in both cases, considering the inter-cave influence and not considering the inter-cave influence. By comparing the first parameters in the two cases, the degree of influence of the group effect on the first parameter is determined and quantified as a static group effect coefficient. For example, the ratio or difference of the first parameter in the two cases can be calculated as the static group effect coefficient.

[0037] In one implementation, the first parameter includes the maximum principal stress around the tunnel and the tunnel convergence. The determination of the static group effect coefficient based on the first parameter in both cases includes the following steps:

[0038] Calculate the ratio of the maximum principal stress around the target cavern in the two cases, and the ratio of the cavern convergence in the two cases. Based on the two ratios, obtain the static group effect coefficient.

[0039] Among them, the maximum principal stress around the cavern refers to the largest principal stress component in the redistributed stress field around the cavern after excavation, reflecting the stress concentration degree of the surrounding rock mass. Cavern convergence refers to the displacement or deformation of the cavern perimeter (usually the cavern wall) into the cavern space, reflecting the degree of deformation and contraction of the surrounding rock mass. Both are important parameters characterizing the static mechanical state of the cavern. By combining these two parameters to determine the static group effect coefficient, the influence of group effects on the static state of the target cavern can be reflected from both mechanical and deformation dimensions. This makes the quantification of the static group effect coefficient more closely match the actual static coupling characteristics of underground cavern groups, improving the accuracy and reliability of the static group effect coefficient and providing a high-quality data foundation for subsequent calculations of the comprehensive group effect coefficient.

[0040] In one implementation, the maximum principal stress and convergence of the target cavern, considering inter-cave influences, can be determined based on the periphery stress and cavern convergence detected by sensors. The sensors are installed in the actual environment of an underground cavern complex, and the detected periphery stress and cavern convergence are naturally data under the influence of group effects. The raw data can be preprocessed, including noise removal and data smoothing, and then the maximum principal stress and cavern convergence of the target cavern can be obtained through mechanical analysis and deformation calculations. For example, a periphery stress field can be established based on the preprocessed periphery stress data, and the maximum principal stress can be determined based on this field.

[0041] In one implementation, a single-cavity scenario of the target cavern can be simulated. Based on the simulation results, the maximum principal stress around the target cavern and the cavern convergence can be calculated without considering the influence between caverns. For example, geotechnical engineering simulation software can be used to numerically simulate a scenario where the target cavern exists independently as a single cavity. The parameter settings of the simulation model are consistent with the actual engineering geological parameters, geometric dimensions, and excavation technology of the target cavern. After running the simulation model, the maximum principal stress around the target cavern and the cavern convergence can be calculated based on the stress and deformation data output by the simulation, without considering group effects. This method can eliminate the influence of other caverns and realize parameter calculation in a single-cavity scenario.

[0042] The ratio of the maximum principal stress around the target cavern in the two cases and the ratio of the cavern convergence in the two cases are calculated. Both ratios are dimensionless values. The two ratios are combined by adding, multiplying, weighting and other methods to obtain the static group effect coefficient.

[0043] In one implementation, the static group effect coefficient can be determined using the following formula:

[0044] (1)

[0045] in, λ This represents the static group effect coefficient (or more specifically, the static group effect coupling coefficient, etc.). α , β Let be the weighting coefficient, satisfying α + β =1; σ group To determine the maximum principal stress around the target cavern under the influence of inter-cavity factors, σ single To determine the maximum principal stress around the target cavern without considering the influence between caverns, u group To determine the convergence of the target chamber considering the influence between chambers, u single This represents the convergence of the target cavern without considering the influence between caverns.

[0046] In one implementation, the weighting coefficients in formula (1) can be set empirically. α , β For example, in stress-dominated cavern scenarios α =0.6, β =0.4, in a cavern scene with active deformation. α =0.4, β =0.6. Different weighting coefficients can be used at different stages of the project.

[0047] The static group effect coefficient is calculated using formula (1). λ This indicates the intensity of the static influence of other chambers on the target chamber. λ =1 indicates no static group effect. λ >1 indicates the presence of a coupled and amplified static group effect. λ A value less than 1 indicates the presence of a weak, offsetting static group effect. This allows for a clear determination of the type and strength of the static group effect, providing an intuitive and accurate quantitative indicator for the subsequent calculation of the comprehensive group effect coefficient.

[0048] Continue to refer to Figure 1 In step S120, the dynamic group effect coefficient is determined based on the transmission between various disasters that may occur in the underground cavern group.

[0049] This involves using historical data from underground cavern groups or similar cavern groups to determine potential hazards, establishing the evolutionary transmission relationships between various hazards, identifying the structure and transmission characteristics of hazard chains, and quantifying the transmission intensity achieved through swarm effects after a hazard occurs to obtain dynamic swarm effect coefficients. For example, if an underground cavern group may experience four hazards, the hazard chains can be determined. L for D 1→ D 2→ D 3→ D 4 represents the disaster transmission relationship of "surrounding rock cracking → block collapse → water inrush and mudslide → structural instability". The dynamic group effect coefficient is determined according to each type of transmission.

[0050] In one implementation, the dynamic group effect coefficient includes a transmission coefficient between different disasters. The transmission coefficient characterizes the intensity of a disaster of one type being transmitted to another type through a group effect. The dynamic group effect coefficient is determined based on the transmissibility between various potential disasters within an underground cavern group, including:

[0051] The transmission probabilities between different disasters are obtained under two scenarios: considering the influence between caverns and not considering the influence between caverns.

[0052] The transmission coefficient between different disasters is determined based on the ratio of the transmission probabilities between different disasters under the two conditions.

[0053] This can be achieved by acquiring historical data for both cavern clusters and single cavern scenarios, statistically analyzing the temporal relationships of different disasters in the historical data, and obtaining the inter-disaster transmission probabilities considering and not considering inter-cave cluster influences. Alternatively, historical data for cavern cluster scenarios can be acquired, and the temporal relationships of different disasters can be statistically analyzed to obtain the inter-disaster transmission probability considering inter-cave cluster influences; software simulations of single cavern scenarios can be performed to obtain simulated data of disasters occurring in a single cavern, and the inter-disaster transmission probability without considering inter-cave cluster influences can be obtained by statistically analyzing the temporal relationships of different disasters. Then, based on the ratio of the transmission probabilities between different disasters in the two scenarios, the transmission coefficient between different disasters can be determined. For example, the transmission coefficient between different disasters can be obtained using the following formula:

[0054] (2)

[0055] in, Indicates from the first i Disaster transmission to the first j The transmission coefficient of a disaster type is used to reflect the transmission of the first type of disaster. i Disasters of this type are transmitted to the first through swarm effects. j The intensity of the disaster; When considering the impact between caverns, the first i Class of disasters to the first j The transmission probability of a disaster (i.e., considering the influence between caverns, the probability of transmission of the first type of disaster) i The first disaster after the occurrence of the disaster j (Probability of occurrence of disasters) When the influence between caverns is not considered, the first i Class of disasters to the first j The transmission probability of a disaster of class III (i.e., without considering the influence between caverns, the probability of transmission of the third type of disaster) i The first disaster after the occurrence of the disaster j (Probability of occurrence of disasters). Typically... ≥1, The larger the value, the stronger the transmission and amplification effect between disasters, that is, the stronger the dynamic group effect.

[0056] In one implementation, disaster transmission pairs with an evolutionary transmission relationship can be identified based on the disaster chain, and the transmission coefficient can be calculated for each disaster transmission pair using formula (2). For example, based on the disaster chain... D 1→ D 2→ D 3→ D 4. Identify three disaster transmission pairs, including D 1→ D 2. D 2→ D 3. D 3→ D4. Calculate the transmission coefficient for each disaster transmission pair. , , .

[0057] The transmission coefficient is determined by the ratio of the disaster transmission probability under the two cases of considering and not considering the influence between caverns. This coefficient serves as the dynamic group effect coefficient, which can closely reflect the actual characteristics of the evolution of disasters in underground cavern groups, accurately quantify the amplification of disaster transmission by the group effect, improve the accuracy of the dynamic group effect coefficient calculation, and provide reliable dynamic characteristic data for the subsequent fusion calculation of the comprehensive group effect coefficient.

[0058] Continue to refer to Figure 1 In step S130, the combined group effect coefficient is obtained by combining the static group effect coefficient and the dynamic group effect coefficient.

[0059] The static group effect coefficient and the dynamic group effect coefficient can be combined by adding, multiplying, or weighting to obtain the comprehensive group effect coefficient. The comprehensive group effect coefficient takes into account both the static coupling foundational influence of the cavern group in steady state and the dynamic transmission influence of disaster evolution, and can characterize the overall impact of group effects on the evolution of the disaster chain, thus achieving a comprehensive quantitative characterization of group effects.

[0060] In one implementation, the overall group effect coefficient can be determined using the following formula:

[0061] (3)

[0062] in, Γ The coefficient representing the overall group effect can satisfy 0 < Γ ≤2, Γ Characterize the overall impact of group effects on the evolution of disaster chains. λ Represents the static group effect coefficient. n This represents the number of possible disaster types in the underground cavern complex. i Indicates the first i disasters, Indicates from the first i Disaster transmission to the first i The transmission coefficient of Class +1 disasters is determined by the disaster chain. The average transmission coefficient of different disaster transmission pairs is calculated using formula (3), and then multiplied by the static group effect coefficient to obtain the comprehensive group effect coefficient. This comprehensive group effect coefficient has high accuracy and reliability, and can intuitively reflect the overall impact intensity of the group effect, providing a core basis for subsequent cavern assessment and prevention and control measures.

[0063] Continue to refer to Figure 1 In step S140, the initial monitoring data of the target cavern is obtained, and the fused monitoring data is obtained after data fusion processing.

[0064] The initial monitoring data can be data collected by sensors. For example, different types of sensors can be deployed at key monitoring sections such as the entrances, middle sections, and intersections of underground cavern complexes to collect initial monitoring data for each cavern. In one implementation, the following four categories and eight items of monitoring data can be collected:

[0065] Mechanical parameters: stress around the tunnel σ Surrounding rock strain ε ;

[0066] Deformation parameters: Cavern convergence u Surrounding rock settlement s ;

[0067] Seepage parameters: seepage flow rate q seepage pressure p ;

[0068] Environmental parameters: temperature T ,humidity ω .

[0069] The initial monitoring data is preprocessed before data fusion. Data fusion methods include, but are not limited to, one or more of the following: single-item fusion, which fuses data according to each sub-indicator (such as tunnel perimeter stress, surrounding rock strain, tunnel convergence, etc.), for example, fusing tunnel perimeter stress collected by different sensors, fusing surrounding rock strain collected by different sensors, etc.; category fusion, which fuses data according to each physical category (such as mechanics, deformation, seepage, environment), for example, fusing tunnel perimeter stress and surrounding rock strain collected by different sensors, fusing tunnel convergence and surrounding rock settlement collected by different sensors, etc.; global fusion, which fuses all types of monitoring data into one category of data, such as fusing the above eight types of monitoring data into one dimension of data.

[0070] In one implementation, reference Figure 2 As shown, the process of obtaining initial monitoring data from the target cavern and then performing data fusion processing to obtain fused monitoring data includes the following steps:

[0071] S210 obtains initial monitoring data through multiple different types of sensors;

[0072] S220: Remove outliers from the initial monitoring data and perform standardization processing to obtain standardized monitoring data;

[0073] S230: Based on the distance between each sensor and the center of the target chamber, the standardized monitoring data is weighted and fused to obtain fused monitoring data.

[0074] After obtaining different types of initial monitoring data, noise and outlier data can be removed using outlier detection rules to eliminate data interference. In one implementation, removing outliers from the initial monitoring data includes the following steps:

[0075] For each type of initial monitoring data, the range of numerical fluctuation is determined based on the average value and standard deviation of the initial monitoring data within the first time window;

[0076] The second time window is determined with the acquisition time of each value point in the initial monitoring data as the center, and the reference value of the initial monitoring data within the second time window is determined.

[0077] Values ​​that are outside the range of numerical fluctuations and whose absolute difference from the reference value is greater than the trend consistency threshold will be removed.

[0078] The first time series window is a basic time series window suitable for statistical data indicators such as average and standard deviation, and can be a day, a week, or a month. The second time series window is a time series window used to measure the degree of deviation of each value from its neighboring values, and can be 12 hours or three days. In one implementation, the length of the first time series window is greater than the length of the second time series window.

[0079] The numerical fluctuation range is determined based on the average and standard deviation of the initial monitoring data within the first time window, representing a reasonable range of fluctuation. For example, this range can be expressed as [avg-3std, avg+3std], where avg represents the average of the initial monitoring data within the first time window, and std represents the standard deviation. The numerical fluctuation range is used to measure whether the numerical points in the initial monitoring data deviate excessively from the overall level. Statistical analysis is performed on the initial monitoring data within the second time window to determine reference values, such as the average or median within the second time window. The trend consistency threshold indicates the reasonable deviation of the values ​​within the second time window from the reference values. If the absolute difference between a value and the reference value is greater than the trend consistency threshold, it indicates that the value has deviated excessively from the overall level of its neighboring values, thus determining whether the numerical points in the initial monitoring data deviate from the time-series trend. Based on these two judgment criteria, numerical points outside the numerical fluctuation range and whose absolute difference from the reference value is greater than the trend consistency threshold are removed. In other words, if a value falls within the range of fluctuation, or its absolute difference from the reference value does not exceed the trend consistency threshold, then that value is retained. For example, the process of removing outliers can be referenced using the following formula:

[0080] (4)

[0081] in, xi This represents the first in a certain type of initial monitoring data. i A numerical value (the unit varies depending on the parameter type). x valid This represents the valid data after removing outliers, and... x i The units are consistent; avg represents the average value of the initial monitoring data of this type within the first time window, and std represents the standard deviation of the initial monitoring data of this type within the first time window; Indicates the second timing window (including the first) i Before and after the number k Reference values ​​(such as the average value) within a range of adjacent values ​​are used to characterize the time-series trend of the data; The trend consistency threshold can be set empirically for each type of monitoring data. For example, the trend consistency threshold for tunnel perimeter stress is 2 MPa, and the trend consistency threshold for surrounding rock strain is 50 × 10⁻⁶ MPa. -6 The trend consistency threshold for cavern convergence is 0.8 mm, and the trend consistency threshold for seepage flow is 3 L / min, etc., to determine whether the data deviates from the time series trend. Through formula (4), the monitoring data within the numerical fluctuation range is retained. On this basis, by introducing the trend consistency verification condition (i.e., the absolute difference between the numerical value and the reference value is not greater than the trend consistency threshold), the monitoring data that meets the trend consistency is retained. This avoids the mistaken removal of effective fluctuation data (such as fluctuation data caused by the instantaneous excavation) and matches the characteristics of strong temporal continuity of underground cavern data and instantaneous noise caused by excavation vibration.

[0082] Besides identifying outliers, other preprocessing steps, such as data deduplication, can be performed on the initial monitoring data. Then, the processed initial monitoring data is standardized to eliminate differences in dimensions and numerical scales between different data types, resulting in standardized monitoring data. In one implementation, considering the large differences in dimensions and significant variations in numerical values ​​among different types of monitoring data from underground caverns, an offset is introduced to optimize the standardization formula, avoiding standardization distortion caused by data distribution offsets. The specific formula can be found below:

[0083] (5)

[0084] in, x norm This represents standardized monitoring data that satisfies 0 ≤ x norm ≤1; x valid This indicates valid monitoring data after preprocessing (such as removing outliers and deduplicating data). x minThis represents the minimum value of this type of monitoring data (which can be initial monitoring data or valid monitoring data). x max This indicates the maximum value of this type of monitoring data, which can be the minimum or maximum value over the entire monitoring period; ε This is an offset used to avoid when... x max= x min When the monitoring data tends to stabilize, the denominator is 0, which is suitable for the characteristics of steady-state monitoring data in underground caverns and reduces the impact of data distribution offset on the standardization results. For example, the offset can be determined empirically, or an adaptive offset can be determined based on the numerical scale or magnitude of the monitoring data. ε It can be 10 -6 .

[0085] After standardization, the standardized monitoring data from each sensor is weighted and fused based on their distance from the center of the target chamber to obtain fused monitoring data. The weights used for weighted fusion are related to this distance; for example, a negative correlation function can be used to determine the weights corresponding to different distances, with closer distances resulting in larger weights. In one implementation, the standardized monitoring data can be weighted and fused using the following formula to obtain the fused monitoring data:

[0086] ;

[0087] (6)

[0088] in, x fusion This indicates the integration of monitoring data. k Indicates the first k One sensor, m This indicates the number of sensors of the same type on the same monitoring section. w k1 Indicates the first k The spatial correlation weights of each sensor can satisfy . w k2 Indicates the first k The parameter sensitivity weights of each sensor can satisfy In one implementation, parameter sensitivity weights can be set based on experience and the degree to which different types of monitoring data are affected by swarm effects. For example, the parameter sensitivity weight for mechanical parameters can be set to 0.3, the parameter sensitivity weight for deformation parameters can be set to 0.3, the parameter sensitivity weight for seepage parameters can be set to 0.25, and the parameter sensitivity weight for environmental parameters can be set to 0.15. This adapts to the characteristics of underground caverns where swarm effects are dominated by mechanics and deformation, and supplemented by seepage.x norm,k Indicates the first k Standardized monitoring data from each sensor dis k Indicates the first k The distance between each sensor and the center of the target chamber.

[0089] It should be noted that this disclosure does not limit the dimensions of the fused monitoring data. For example, by fusing the eight categories of monitoring data separately using a single-item fusion method, eight-dimensional fused monitoring data can be obtained. Alternatively, by fusing by major categories, four-dimensional fused monitoring data can be obtained. Alternatively, by fusing globally, one-dimensional fused monitoring data can be obtained. Alternatively, multiple methods of single-item fusion, major category fusion, and global fusion can be combined, such as by fusing major category fusion and global fusion, to obtain five-dimensional fused monitoring data (where four dimensions are the result of major category fusion and one dimension is the result of global fusion).

[0090] The above weighted fusion method matches the characteristics of strong spatial correlation and significant group effect of underground cavern groups. By introducing spatial correlation weight and parameter sensitivity weight to replace the single precision weight, the fused monitoring data not only reflects the spatial correlation of the group effect, but also highlights the dominant role of the core parameters.

[0091] Continue to refer to Figure 1 In step S150, feature extraction is performed on the integrated group effect coefficient and the fused monitoring data to obtain the target features.

[0092] For example, the integrated group effect coefficient and the fused monitoring data can be treated as a whole and processed using a machine learning model with feature extraction capabilities. Feature extraction can capture core features such as spatial coupling and temporal correlation in the data, and achieve deep fusion of group effect features and multi-source monitoring data features to obtain target features that simultaneously contain group effect features and monitoring data features.

[0093] In one implementation, reference Figure 3 As shown, the above-mentioned feature extraction of the integrated group effect coefficient and fused monitoring data to obtain target features includes the following steps S310 to S360:

[0094] S310, the integrated group effect coefficient and the fused monitoring data are combined to obtain the first sequence.

[0095] For example, integrating monitoring data x fusion With the overall group effect coefficient Γ The data is spliced ​​together to obtain 9-dimensional data (8 dimensions representing 8 types of monitoring data, and 1 dimension representing the comprehensive group effect coefficient). The 9-dimensional data from different times are arranged in chronological order to form the first sequence. X =[x 1, x 2, … , x len ],in len Given the time series length, the overall dimension of the first sequence can be represented as ( len ,9). When training a machine learning model for feature extraction and embedding processing, the dimension of the first sequence of samples can be expanded to ( N , len ,9), N The number of samples. Each element in the first sequence. x i All data are 9-dimensional and can be represented as x i = [ x i1 , x i2 , … , x i9 The input format of the first sequence can fully take into account the temporal correlation of multi-source parameters and the characterization of group effects, and is suitable for subsequent feature extraction and temporal modeling requirements.

[0096] S320, perform grouped convolution processing on the first sequence to obtain multiple groups of first intermediate features.

[0097] The first sequence can be divided into multiple groups based on the data type or characteristics in the first sequence. Convolution processing is performed on each group to extract the local features of each group and obtain multiple groups of first intermediate features.

[0098] In one implementation, the fused monitoring data includes mechanical parameters, deformation parameters, seepage parameters, and environmental parameters. The above-described grouped convolutional processing of the first sequence yields multiple sets of first intermediate features, including:

[0099] The data in the first sequence is divided into three groups, and each group is convolved to obtain the first intermediate feature corresponding to each group. Among them, mechanical parameters, deformation parameters, and comprehensive group effect coefficient belong to the first group, seepage parameters belong to the second group, and environmental parameters belong to the third group.

[0100] For example, the overall dimension of the first sequence is ( len(9), where each element is 9-dimensional data, namely, tunnel perimeter stress, surrounding rock strain, tunnel convergence, surrounding rock settlement, seepage flow, seepage pressure, temperature, humidity, and comprehensive group effect coefficient. Tunnel perimeter stress, surrounding rock strain, tunnel convergence, surrounding rock settlement, and comprehensive group effect coefficient are divided into the first group; seepage flow and seepage pressure are divided into the second group; and temperature and humidity are divided into the third group. Convolution processing is performed within each group. Alternatively, the overall dimension of the first sequence is ( len The dataset (5) contains 5-dimensional data, including mechanical parameters (data fused from tunnel perimeter stress and surrounding rock strain), deformation parameters (data fused from tunnel convergence and surrounding rock settlement), seepage parameters (data fused from seepage flow and seepage pressure), environmental parameters (data fused from temperature and humidity), and a comprehensive group effect coefficient. The mechanical parameters and comprehensive group effect coefficient are grouped into the first group, the seepage parameters into the second group, and the environmental parameters into the third group. After grouping, convolution processing is performed within each group to obtain multiple sets of first intermediate features.

[0101] The feature extraction method of grouped convolution is in line with the actual engineering situation of underground cavern groups. In particular, the division method of the above three groups is in line with the characteristics of group effects dominated by mechanics and deformation, thereby improving the extraction accuracy of important features and laying the foundation for the extraction and fusion of deeper features in the future.

[0102] S330 performs depthwise separable convolution on multiple sets of first intermediate features to obtain second intermediate features.

[0103] In this approach, the grouping dimension of multiple sets of first intermediate features can be used as a channel, or the temporal dimension can be used as a channel, for depthwise separable convolution. Specifically, convolution is first performed within each channel, and then pointwise convolution is performed between channels. Depthwise separable convolution can reduce the model's parameter size and computational cost, while enabling the extraction of deeper features to obtain the second intermediate features.

[0104] S340, by activating the second intermediate feature, the third intermediate feature is obtained.

[0105] This disclosure does not limit the specific method of activating the second intermediate feature. For example, activation functions such as ReLU (Rectified Linear Unit), sigmoid (a sigmoid function), and tanh (hyperbolic tangent function) can be used for activation.

[0106] In one implementation, multiple scales of convolution kernels are used to perform depthwise separable convolution processing on multiple sets of first intermediate features to obtain second intermediate features at multiple scales. The second intermediate features at multiple scales are then merged, and the merged result is activated to obtain third intermediate features. In this way, by using convolution processing at multiple scales, group effect information of different spatial ranges in the first intermediate features is captured, thereby improving the comprehensiveness and sufficiency of feature extraction.

[0107] For example, considering the significant spatial coupling characteristic of cavern cluster effects, a lightweight CNN (Convolutional Neural Network) feature extraction layer is designed and employed for feature extraction. Traditional convolutions are replaced with GroupConv and DepthSepConv to enhance the extraction of spatial coupling features of cluster effects while maintaining a lightweight approach. Three convolutional kernels of different scales (e.g., 3×3, 5×5, 7×7) are used in the DepthSepConv processing to capture cluster effects at different spatial ranges (e.g., close-range cavern coupling and long-range cavern collaboration). Refer to the following formula:

[0108] ;

[0109] ;

[0110] (7)

[0111] in, X This represents the first input sequence, whose dimension can be ( len ,9), and input it into a CNN for feature extraction. f conv,group The first intermediate feature can have dimensions of ( len ,9), g =3 indicates that the convolution is performed in 3 groups; f conv,sep For the second intermediate feature, the dimension can be ( len ,9); f conv The final extracted spatial coupling features of the group effect; f conv,3×3 , f conv,5×5 , f conv,7×7 The second intermediate features corresponding to the outputs of the convolution kernels at three different scales represent the group effect features in different spatial ranges. W group , W sep These represent the weights of grouped convolution and depthwise separable convolution, respectively. bgroup , b sep These represent the biases of grouped convolution and depthwise separable convolution, respectively. b conv This indicates the convolution bias, used to adjust the feature baseline and avoid systematic errors; concat This indicates a feature splicing operation that merges second intermediate features at different spatial scales; ReLU Using an activation function can prevent gradient vanishing and enhance the nonlinear representation of features.

[0112] Based on the convolution process shown in formula (7), grouped convolution matches the heterogeneity of multi-source parameters, depthwise separable convolution can achieve lightweighting, and multi-scale convolution kernels capture group effects of different ranges, greatly improving the targeting and efficiency of feature extraction, and providing high-quality spatial feature support for subsequent adaptive embedding of group effect features.

[0113] S350 uses embedded weights to weight the third intermediate feature with the comprehensive group effect coefficient to obtain the fused feature.

[0114] After obtaining the third intermediate feature, the comprehensive group effect coefficient can be directly embedded into the third intermediate feature through a weighted method to obtain the fused feature. The weights used in this weighting process are embedding weights, which can be determined through machine learning training or manually set based on experience; this disclosure does not limit this. In one embodiment, the comprehensive group effect coefficient can be numerically adjusted and normalized, and the embedding weights can be determined based on the normalization result; the dimensions of the third intermediate feature and the comprehensive group effect coefficient are unified, and the embedding weights are used to weight the dimension-unified third intermediate feature and the comprehensive group effect coefficient to obtain the fused feature. For example, the following formula is referenced:

[0115] ;

[0116] (8)

[0117] Among them, w Γ This represents adaptive embedding weights, which dynamically adjust the embedding weights of group effect features using the sigmoid activation function to achieve adaptive matching between the group effect strength and the model input. This connects the spatial features extracted by the CNN layers; the stronger the group effect (…), the better. Γ The larger the value, the greater the weight, highlighting the influence of the group effect; k , b Γ To adaptively adjust parameters, they can be set based on experience or determined through training. For example... k =0.8, b Γ =0.1; Γ expandTo unify the dimensions of the comprehensive group effect coefficient after dimensional expansion with those of the third intermediate feature, for example, a dimension is added to the comprehensive group effect coefficient, and a preset value or the original value of the comprehensive group effect coefficient is filled in to achieve dimensional unification. f embed This is the fusion feature of the third intermediate feature and the comprehensive group effect coefficient. Through formula (8), the adaptive fusion of group effect features and spatial features can be achieved, adapting to the dual features of static coupling and dynamic transmission of group effects, and providing comprehensive feature support for subsequent time series modeling.

[0118] S360 uses a time-series model to process the fused features to obtain the target features.

[0119] Among them, time-series models include, but are not limited to, LSTM (Long Short Term Memory) and GRU (Gated Recurrent Unit), which can further mine the time-series information in the fused features to obtain the target features.

[0120] For example, a lightweight LSTM is used as the time series model. It employs a lightweight decoupling design (decoupling the gating weight matrix, reducing parameters by 40%), and introduces a time series attention mechanism to focus on key time steps in the disaster chain, connecting with the output of the previous embedding layer. f embed This adapts to the temporal evolution characteristics of disaster chains. Refer to the following formula:

[0121] ;

[0122] ;

[0123] (9)

[0124] in, c t express t The LSTM cell state at each time step is used to store the temporal feature information of the current time step, characterizing the temporal cumulative features of the evolution of the disaster chain; c t-1 express t-1 LSTM cell state at any given time; f t express t The output of the LSTM forget gate at each moment; i t express t At what moment does the LSTM input gate output? f embed,t The output of the adaptive embedding layer representing group effect features t The fusion characteristics of moments; WC The cell state update weight matrix represents the cell state. b C This represents the cell state update bias vector; h t for t LSTM hidden layer output at time; o t express t The LSTM output gate outputs at any given time; h t express t The core temporal features after attention weighting at each moment, i.e., the target features; W a Represents the temporal attention weight matrix; dim k Indicates the hidden layer dimension; tanh( ) is the hyperbolic tangent function, used to perform nonlinear transformations on input features; softmax( The activation function is used to convert the output of the hidden layer at each time step. h t Weight normalization.

[0125] Based on the above feature extraction scheme, adaptive matching and deep fusion of group effect features and spatial features were achieved. In addition, information such as parameter type grouping and time series was combined to ensure that comprehensive, complete and sufficient target features were extracted, providing a solid foundation for subsequent cavern assessment and determination of prevention and control measures.

[0126] Continue to refer to Figure 1 In step S160, the assessment results of the target cavern are obtained based on the target characteristics, and disaster prevention and control measures are determined based on the assessment results.

[0127] This process involves conducting a comprehensive assessment and disaster analysis of the target cavern based on its characteristics, resulting in an assessment of the target cavern's current condition and potential disaster risks. Corresponding disaster prevention and control measures are then matched to the assessment results to achieve targeted disaster prevention and control for the target cavern.

[0128] In one implementation, the assessment results for the target cavern include the predicted probability of a disaster and the predicted time of the disaster. The assessment results for the target cavern obtained based on the target characteristics include:

[0129] By using fully connected layers and activation functions to process the target features, the probability of disaster occurrence in the target cavern can be obtained.

[0130] By using a linear regression layer to process the target features, the predicted occurrence time of disasters in the target cavern can be obtained.

[0131] For example, a fully connected layer and a softmax activation function are used to output the probability of disaster occurrence. P z A linear regression layer is used to output the predicted time of disaster occurrence. t z Please refer to the following formula:

[0132] (10)

[0133] in, P z Indicates the first z The predicted probability of occurrence of disasters; t z Indicates the first z The predicted occurrence time of such disasters; z This indicates the sequence number of a disaster in a disaster chain, and can refer to the disaster chain itself. L ={ D 1→ D 2→…→ D n The first in} z disasters such as z =1 corresponds to the initial disaster D 1 (e.g., cracking of the surrounding rock). z =2,3,…, n Corresponding to subsequent secondary disasters D 2. D 3、…、 D n (such as block collapse, water inrush and mudslide, structural instability); h t Represent target features; W p , W t These represent the weights of the fully connected layer and the linear regression layer, respectively. b p , b t These represent the biases of the fully connected layer and the linear regression layer, respectively; the output disaster prediction probability and disaster prediction time provide accurate basis for the next step of intelligent prevention and control, and realize the seamless connection between cavern disaster prediction and prevention and control measures.

[0134] For example, refer to Figure 4As shown, a lightweight CNN-LSTM model for feature extraction can be constructed, and an output layer (including the aforementioned fully connected layer, softmax activation function, and linear regression layer) can be added to form a complete model (which can be called the group effect feature embedding-lightweight CNN-LSTM model). After inputting the first sequence into the input layer of this complete model, convolutional feature extraction is performed. Specifically, grouped convolutions and depthwise separable convolutions are performed in the lightweight CNN feature extraction layer, and features corresponding to different scales are concatenated and activated to obtain the third intermediate feature. Then, feature embedding is performed. Specifically, in the group effect feature adaptive embedding layer, the embedding weights are determined by adaptive matching and the sigmoid activation function. The third intermediate feature and the comprehensive group effect coefficient are weighted using the embedding weights to obtain the fused feature. Next, temporal state propagation is performed. Specifically, in the lightweight uncoupled LSTM temporal layer, the target feature is obtained based on the temporal attention mechanism and temporal state propagation. Finally, temporal modeling output is performed. Specifically, the target feature is input into the output layer, the disaster prediction probability is output through the fully connected layer and the softmax activation function, and the disaster prediction time is output through the linear regression layer.

[0135] Figure 4 The complete model shown balances prediction accuracy and engineering real-time performance, enabling early identification of disaster chains (such as...). D 1) - To evolutionary trends (e.g.) D 2- D n Accurate prediction of "time of occurrence".

[0136] In one implementation, the AdamW optimizer can be used, with weight decay (e.g., a weight decay coefficient of 1e-5) to avoid model overfitting and improve model generalization ability. The number of iterations is set to 100 rounds, and an early stopping strategy is adopted (training stops if the validation set loss does not decrease for 10 consecutive rounds) to ensure that the model training effect and lightweight performance are balanced, providing reliable model support for accurate prediction of subsequent disaster chains.

[0137] In one implementation, the assessment results for the target cavern include the predicted probability of a disaster and the predicted time of occurrence of the disaster. The determination of disaster prevention and control measures based on the assessment results includes:

[0138] Based on the comprehensive group effect coefficient, the probability of disaster occurrence, and the predicted time of disaster occurrence, corresponding disaster prevention and control measures are determined.

[0139] In addition to the assessment results, a comprehensive group effect coefficient is added, and the corresponding disaster prevention and control measures are determined based on these three parameters. For example, the prevention and control level is determined based on the comprehensive group effect coefficient, the predicted probability of disaster occurrence, and the predicted time of disaster occurrence, and then the corresponding disaster prevention and control measures are determined. The determination of the prevention and control level is based on the comprehensive group effect coefficient. Γ Based on the probability of disaster occurrence P z (This could be based on the probability of the most severe disaster occurring, i.e., the highest probability among various disasters predicted for the target cavern) and the predicted time of disaster occurrence. t z (This could be the predicted time of occurrence corresponding to the highest probability of a disaster) To supplement this, a three-dimensional intelligent judgment standard is formed to replace the single-dimensional judgment, ensuring more accurate classification. Based on the actual situation of underground cavern complex engineering prevention and control, it is divided into 4 levels of prevention and control, as shown in Table 1:

[0140] Table 1

[0141]

[0142] In one implementation, after implementing disaster prevention and control measures, the measures are evaluated based on the changes in the comprehensive group effect coefficient, the changes in the predicted probability of disaster occurrence, and the response time. Specifically, after implementing the measures, the comprehensive group effect coefficient and the predicted probability of disaster occurrence for the target cavern can be re-determined, and the changes are compared to the values ​​before the implementation of the measures. The response time refers to the time span from obtaining the above evaluation results to the implementation of the prevention and control measures. The disaster prevention and control measures are evaluated based on the changes in the comprehensive group effect coefficient, the changes in the predicted probability of disaster occurrence, and the response time. Generally, the higher the change (reduction) in the comprehensive group effect coefficient, the higher the change (reduction) in the predicted probability of disaster occurrence, and the shorter the response time, the better the disaster prevention and control measures are considered.

[0143] For example, constructing the group effect weakening rate The rate of reduction in the probability of disasters Prevention and control response time τ Three core indicators are used to comprehensively evaluate the effectiveness of prevention and control. Among them, Γ before , Γ after These represent the comprehensive group effect coefficients before and after the implementation of disaster prevention and control measures; P before , P after These represent the predicted probability of disaster occurrence before and after the implementation of disaster prevention and control measures; τ The response time for epidemic prevention and control is measured in minutes. An evaluation standard for epidemic prevention and control effectiveness is constructed based on three indicators, allowing for a tiered evaluation of the effectiveness. See Table 2 for details.

[0144] Table 2

[0145]

[0146] In one implementation, if the evaluation result of disaster prevention and control measures is "unqualified", the prevention and control measures can be adjusted according to the evolution characteristics of the swarm effect, such as increasing the reinforcement intensity, increasing the monitoring frequency, and optimizing the prevention and control priority.

[0147] Figure 5 A flowchart of the implementation method of this disclosure is shown. Specifically, for the target cavern, the static coupling coefficient (i.e., static group effect coefficient) and dynamic transmission coefficient (i.e., dynamic group effect coefficient) are determined through swarm effect quantification analysis, and the comprehensive group effect coefficient is obtained by combining the two. Multi-source monitoring data of the target cavern are collected, and data preprocessing and standardization are performed using numerical fluctuation range (avg±std) and trend consistency verification. Data fusion is performed based on spatial correlation weights and parameter sensitivity weights. The fused monitoring data and the comprehensive group effect coefficient are input into a swarm effect feature embedding-lightweight CNN-LSTM model to obtain the disaster prediction probability and disaster prediction time. Based on these two results and combined with the comprehensive group effect coefficient, disaster prevention and control measures are determined through a three-dimensional intelligent prevention and control judgment standard. Disaster prevention and control measures are formulated and implemented. The prevention and control effect is evaluated based on the swarm effect weakening rate, the disaster occurrence probability reduction rate, and the prevention and control response time. If the evaluation is unsatisfactory, the disaster prevention and control measures are re-determined; if the evaluation is satisfactory, the disaster chain prevention and control is terminated.

[0148] The following example illustrates the implementation of this disclosure.

[0149] An underground cavern complex of a large hydropower station was selected (comprising eight caverns: main powerhouse, auxiliary powerhouse, surge tank, water diversion tunnel, main transformer tunnel, busbar tunnel, tailrace tunnel, and access tunnel, arranged in a cluster). The caverns are buried at a depth of 200-350m, with granite as the main surrounding rock and localized fault fracture zones. The main powerhouse dimensions (length × width × height) are 180m × 25m × 50m. The dimensions of the other caverns are adapted to their functions. The excavation progress is 60%, indicating a mid-stage excavation phase with significant cluster effects. Sixty multi-source sensors were deployed, covering key sections of the eight caverns, collecting eight parameters across four categories: stress, deformation, seepage, and environmental factors. Over 50,000 data points were collected for model training, cluster effect quantification, and prevention and control verification. Historical monitoring data shows that the cavern complex has experienced two minor rock cracking events, indicating a risk of secondary disaster transmission due to the cluster effect.

[0150] First, a quantitative analysis of the group effect of the cavern clusters is performed. Using formula (1), since the group effect is stress-dominated in the current engineering stage, the weighting coefficients in formula (1) can be determined. α =0.6, β =0.4, calculate the static group effect coefficient. λ Dynamic transmission coefficients are calculated through historical disaster statistics. Number of disaster chains n=4, D 1→ D 2→ D 3→ D 4 represents the following steps respectively: surrounding rock cracking → block collapse → water and mud inrush → structural instability. The dynamic transmission coefficient is a unified value for each link in the disaster chain, including... = 1.82 。 =1.65 ， =1.58 average conductivity =1.68. Finally, combining formula (3), the comprehensive group effect coefficient of the 8 caverns is obtained. Γ Some of the results are shown in Table 3.

[0151] Table 3

[0152]

[0153] Next, the monitoring data were preprocessed and fused. Outliers were removed using a consistency check of numerical fluctuation range and trend, with a removal rate of 3.2%. These outliers mainly included anomalies caused by sudden stress changes due to excavation vibration and sensor malfunction data. After processing using a constructed adaptive standardization formula, a dual-weight fusion method (spatial correlation weight and parameter sensitivity weight) was used to obtain the fused data. The fused mechanical parameter values ​​were obtained from the tunnel perimeter stress and surrounding rock strain, while the fused deformation parameter values ​​were obtained from the tunnel convergence and surrounding rock settlement. Key monitoring sections of the main powerhouse and surge tank (two core tunnels) were selected, and the fusion results of some core parameters at three typical time steps are shown in Table 4.

[0154] Table 4

[0155]

[0156] Then, the machine learning model was trained and disaster chain prediction was performed. Specifically, the preprocessed data of over 40,000 records were divided into training, validation, and test sets in a 7:2:1 ratio, and the original lightweight CNN-LSTM model and the improved model were trained respectively (see reference). Figure 4 The model improvements include a lightweight CNN feature extraction layer, a group effect adaptive embedding layer, and a lightweight decoupled LSTM temporal layer. Training parameters were set as follows: AdamW optimizer (weight decay coefficient 1e-5), learning rate 0.001, 100 iterations, and an early stopping strategy (stopping after 10 consecutive iterations if the validation set loss does not decrease). Model comparison results are shown in Table 5.

[0157] Table 5

[0158]

[0159] According to the comparison results of the four core performance indicators in Table 5, the real-time performance and prediction accuracy of the improved model have been significantly improved, indicating that the improvement measures in this implementation are effective. The improved model is more in line with the actual engineering needs of predicting the disaster chain of underground cavern groups in large hydropower stations, and can provide strong support for the efficient implementation of subsequent graded prevention and control measures, thus having higher engineering application value. Figure 6 The comparison of the loss values ​​of the original model and the improved model is shown. Compared with the original model, the loss value of the improved model is significantly reduced.

[0160] The results of the disaster chain prediction for the main plant using the improved model are as follows:

[0161] ① Early identification results: D 1 (Cracks in the surrounding rock) have occurred;

[0162] ② The predicted probability of occurrence of each disaster: P 1 (i.e., disaster) D The predicted probability of 1 occurring is 98.5%. P 2 = 45.3% P 3 = 18.7% P 4 = 8.2%;

[0163] ③Predicted occurrence time of each disaster: t 1 (i.e., disaster) D 1. Predicted occurrence time = (Previously occurred, current time) t 2 = 18.2h t 3=42.5h t 4 = 78.3h.

[0164] Based on the above calculation results and Table 1, the selection of prevention and control indicators follows the principle of choosing the highest level, and is determined to be Level III prevention and control, including emergency intelligent monitoring of the main plant, advanced grouting reinforcement, evacuation of personnel and equipment, and Level III early warning, etc.

[0165] In summary, the embodiments disclosed herein address the shortcomings of existing disaster chain prevention technologies for underground cavern groups in large-scale water conservancy projects. By combining swarm effect quantification, multi-source perception fusion, and lightweight deep learning technologies, a complete intelligent prevention and control system of "perception-analysis-prediction-prevention" is constructed. Its beneficial effects closely align with the needs of theoretical research and engineering applications, as detailed below:

[0166] The group effect of underground cavern groups is quantified into static group effect coefficients and dynamic group effect coefficients, achieving a breakthrough from "qualitative description" to "quantitative modeling" of group effects. Furthermore, the comprehensive group effect coefficient is embedded into the entire process of data fusion, disaster prediction, and hierarchical prevention and control, changing the problems of independent prevention and control of single caverns and insufficient consideration of group effects in related technologies. This enables precise prevention and control guided by group effects and effectively solves the problem of insufficient targeting of prevention and control of caverns with different group effect intensities.

[0167] By optimizing the lightweight CNN-LSTM model and improving the feature extraction, swarm effect embedding, and temporal modeling structure based on the characteristics of cavern cluster monitoring data, while taking into account the model inference speed, it is possible to achieve early identification of disaster chains, accurate prediction of disaster occurrence probability and estimated time, and advance prediction of disaster chain evolution trends, providing sufficient time for proactive prevention and control, and effectively curbing the transmission and spread of disaster chains.

[0168] Establish a four-dimensional hierarchical prevention and control standard based on "cluster effect intensity + disaster probability + occurrence time + disaster type", and formulate prevention and control measures that are intelligently adapted to each level, disaster type, and cluster effect intensity. This will not only prevent cost waste caused by excessive investment in prevention and control, but also avoid risks and hidden dangers caused by insufficient investment in prevention and control, thus achieving optimal control of prevention and control costs.

[0169] This disclosure also provides a cluster-effect-based underground cavern disaster prevention and control device, applied to underground cavern groups in water conservancy projects, wherein the underground cavern group includes a target cavern. (Reference) Figure 7 As shown, the underground cavern disaster prevention and control device 700 includes:

[0170] The static group effect processing module 710 is configured to obtain the first parameter of the target cavern under two conditions: considering the influence between caverns and not considering the influence between caverns, and to determine the static group effect coefficient based on the first parameter under the two conditions.

[0171] The dynamic group effect processing module 720 is configured to determine the dynamic group effect coefficient based on the transmission between various disasters that may occur in the underground cavern group;

[0172] The integrated group effect processing module 730 is configured to obtain the integrated group effect coefficient by combining the static group effect coefficient and the dynamic group effect coefficient;

[0173] The monitoring data processing module 740 is configured to acquire the initial monitoring data of the target cavern and obtain fused monitoring data through data fusion processing;

[0174] The feature processing module 750 is configured to extract features from the integrated group effect coefficient and the fused monitoring data to obtain target features;

[0175] The prevention and control measures determination module 760 is configured to obtain the assessment result of the target cavern based on the target characteristics, and determine the disaster prevention and control measures based on the assessment result.

[0176] In one embodiment, the first parameter includes the maximum principal stress around the tunnel and the tunnel convergence; determining the static group effect coefficient based on the first parameter in the two cases includes: calculating the ratio of the maximum principal stress around the tunnel in the two cases and the ratio of the tunnel convergence in the two cases, and obtaining the static group effect coefficient based on the two ratios.

[0177] In one implementation, obtaining the static group effect coefficient based on the two ratios includes: determining the static group effect coefficient using the following formula:

[0178] ;

[0179] in, λ This represents the static group effect coefficient. α , β Let be the weighting coefficient, satisfying α + β =1; σ group To determine the maximum principal stress around the target cavern considering the influence between caverns, σ single The maximum principal stress around the target cavern is determined without considering the influence between caverns. u group To determine the cavern convergence of the target cavern considering the influence between caverns, u single This refers to the cavern convergence of the target cavern without considering the influence between caverns.

[0180] In one embodiment, obtaining the first parameters of the target cavern under both the case of considering the inter-cave influence and the case of not considering the inter-cave influence includes: determining the maximum principal stress and the cavern convergence of the target cavern under the case of considering the inter-cave influence based on the cavern periphery stress and cavern convergence detected by the sensor; performing a simulation of a single-cave scenario of the target cavern, and calculating the maximum principal stress and the cavern convergence of the target cavern under the case of not considering the inter-cave influence based on the simulation results.

[0181] In one embodiment, the dynamic group effect coefficient includes the transmission coefficient between different disasters; determining the dynamic group effect coefficient based on the transmission between various disasters that may occur in the underground cavern group includes: obtaining the transmission probability between different disasters under two conditions, considering the influence between caverns and not considering the influence between caverns; and determining the transmission coefficient between different disasters based on the ratio of the transmission probability between different disasters under the two conditions.

[0182] In one implementation, obtaining the comprehensive group effect coefficient by combining the static group effect coefficient and the dynamic group effect coefficient includes: determining the comprehensive group effect coefficient using the following formula:

[0183] ;

[0184] in, Γ This represents the combined group effect coefficient. λ This represents the static group effect coefficient. n The number of possible disaster types in the aforementioned underground cavern complex. i Indicates the first i disasters, Indicates from the first i Disaster transmission to the first i The transmission coefficient of +1 type disasters, the ordinal number of each disaster is determined by the disaster chain.

[0185] In one embodiment, the step of acquiring initial monitoring data of the target cavern and obtaining fused monitoring data through data fusion processing includes: obtaining the initial monitoring data through multiple different types of sensors; removing outliers from the initial monitoring data and performing standardization processing to obtain standardized monitoring data; and weighting and fusing the standardized monitoring data according to the distance of each sensor from the center of the target cavern to obtain the fused monitoring data.

[0186] In one implementation, removing outliers from the initial monitoring data includes: for each type of initial monitoring data, determining the numerical fluctuation range based on the average value and standard deviation of the initial monitoring data within a first time window; determining a second time window centered on the collection time of each numerical point in the initial monitoring data, and determining a reference value for the initial monitoring data within the second time window; and removing numerical points that are outside the numerical fluctuation range and whose absolute difference from the reference value is greater than a trend consistency threshold.

[0187] In one embodiment, the step of weighting and fusing the standardized monitoring data based on the distances of each sensor to the center of the target chamber to obtain the fused monitoring data includes: weighting and fusing the standardized monitoring data using the following formula to obtain the fused monitoring data:

[0188] ;

[0189] ;

[0190] in, x fusion This refers to the fused monitoring data. k Indicates the firstk One sensor, m This indicates the number of sensors of the same type on the same monitoring section. w k1 Indicates the first k Spatial correlation weights of individual sensors w k2 Indicates the first k The parameter sensitivity weights of each sensor x norm,k Indicates the first k Standardized monitoring data from each sensor dis k Indicates the first k The distance between each sensor and the center of the target chamber.

[0191] In one embodiment, the step of extracting features from the integrated group effect coefficient and the fused monitoring data to obtain target features includes: merging the integrated group effect coefficient and the fused monitoring data to obtain a first sequence; performing grouped convolution processing on the first sequence to obtain multiple sets of first intermediate features; performing depthwise separable convolution processing on the multiple sets of first intermediate features to obtain second intermediate features; performing activation processing on the second intermediate features to obtain third intermediate features; weighting the third intermediate features with the integrated group effect coefficient using embedding weights to obtain fused features; and processing the fused features using a time-series model to obtain the target features.

[0192] In one embodiment, the fused monitoring data includes mechanical parameters, deformation parameters, seepage parameters, and environmental parameters; the step of performing grouped convolution processing on the first sequence to obtain multiple groups of first intermediate features includes: dividing the data in the first sequence into three groups, performing convolution processing on each group to obtain the first intermediate features corresponding to each group; wherein, the mechanical parameters, deformation parameters, and the comprehensive group effect coefficient belong to the first group, the seepage parameters belong to the second group, and the environmental parameters belong to the third group.

[0193] In one embodiment, the step of performing depthwise separable convolution processing on the multiple sets of first intermediate features to obtain second intermediate features includes: performing depthwise separable convolution processing on the multiple sets of first intermediate features using convolution kernels of multiple scales respectively to obtain second intermediate features of multiple scales; the step of obtaining third intermediate features by activating the second intermediate features includes: merging the second intermediate features of multiple scales and activating the merged result to obtain the third intermediate feature.

[0194] In one implementation, the step of using embedding weights to weight the third intermediate feature and the comprehensive group effect coefficient to obtain a fused feature includes: adjusting and normalizing the comprehensive group effect coefficient numerically, determining the embedding weights based on the normalization result; unifying the dimensions of the third intermediate feature and the comprehensive group effect coefficient, and using the embedding weights to weight the dimension-unified third intermediate feature and the comprehensive group effect coefficient to obtain the fused feature.

[0195] In one embodiment, the evaluation result of the target cavern includes the disaster prediction probability and the disaster prediction time; obtaining the evaluation result of the target cavern based on the target features includes: processing the target features using a fully connected layer and an activation function to obtain the disaster prediction probability of the target cavern; and processing the target features using a linear regression layer to obtain the disaster prediction time of the target cavern.

[0196] In one embodiment, the assessment result of the target cavern includes the disaster prediction probability and the disaster prediction time; determining disaster prevention and control measures based on the assessment result includes: determining corresponding disaster prevention and control measures based on the comprehensive group effect coefficient, the disaster prediction probability, and the disaster prediction time; the device is further configured to: after implementing the disaster prevention and control measures, evaluate the disaster prevention and control measures based on the change in the comprehensive group effect coefficient, the change in the disaster prediction probability, and the prevention and control response time.

[0197] The specific details of each part of the above-mentioned device have been described in detail in the method section of the implementation plan. For any undisclosed details, please refer to the implementation plan of the method section, and therefore will not be repeated here.

[0198] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0199] This disclosure also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the method steps of various exemplary embodiments of this disclosure.

[0200] In one implementation, the computer program product can be a tangible product, such as a computer-readable storage medium storing a computer program. The readable storage medium can be based on electrical, magnetic, optical, electromagnetic, infrared, or other signals, and includes, but is not limited to: Random Access Memory (RAM), Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory, Hard Disk Drive (HDD), Solid State Disk (SSD), etc. For example, the computer program product can be a non-volatile storage medium storing a computer program, such as read-only memory, NAND flash memory, etc.

[0201] In one implementation, the computer program product can be an intangible product. For example, the computer program product can be a virtual digital product, such as an executable file or installation package containing a computer program.

[0202] Computer program code can be written in one or more programming languages. Examples of programming languages ​​include C, Java, and C++. Program code can execute entirely on the user's computing device, partially on the user's computing device, or as a standalone software package. It can also execute partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computing device (e.g., via an internet connection provided by a mobile network operator).

[0203] Computer programs can be carried or transmitted via signals such as electrical, magnetic, optical, electromagnetic, and infrared rays. Electronic devices can convert the signals carrying computer programs into digital signals, thereby running the computer programs. When a computer program runs on an electronic device, its code is used to cause the electronic device to execute (more specifically, to be executed by the processor of the electronic device) the method steps of various embodiments of this disclosure, such as... Figure 1 The method and steps.

[0204] Implementing the above methods and steps through computer programs achieves the following technical effects: Firstly, it provides a cluster-effect-oriented disaster prevention and control scheme for underground caverns based on intelligent algorithms. This breaks away from the approach of independent prevention and control for single caverns in related technologies, integrating the cluster effect throughout the entire disaster prevention and control process for underground caverns. This adapts to the clustered layout characteristics of underground cavern groups, ensuring that cavern assessment results closely approximate actual conditions, and that corresponding prevention and control measures match the intensity of the cluster effect. This improves the accuracy and effectiveness of prevention and control. Secondly, by quantifying and fusing static and dynamic cluster effects to obtain a comprehensive cluster effect coefficient, and extracting features from it and monitoring data, the assessment results of the target cavern and corresponding disaster prevention and control measures are determined based on the obtained target features. This improves the accuracy and effectiveness of prevention and control, and can adapt to engineering scenarios with large cavern groups, complex structures, and numerous disaster factors, meeting the disaster prevention and control needs of large-scale water conservancy projects.

[0205] This disclosure also provides an electronic device. The electronic device includes a processor and a memory. The memory stores executable instructions for the processor, such as computer programs. The processor executes the executable instructions to perform the method steps of various exemplary embodiments of this disclosure.

[0206] The following is for reference. Figure 8 The electronic device is illustrated by way of a general-purpose computing device. It should be understood that... Figure 8 The electronic device 800 shown is merely an example and should not be construed as limiting the functionality or scope of this disclosure.

[0207] like Figure 8 As shown, the electronic device 800 may include: a processor 810, a memory 820, a bus 830, an I / O (input / output) interface 840, and a network adapter 850.

[0208] The memory 820 may include volatile memory, such as RAM 821 and cache unit 822, and may also include non-volatile memory, such as ROM 823. The memory 820 may also include one or more program modules 824, including but not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. For example, program module 824 may include the modules described above.

[0209] The processor 810 may include one or more processing units, such as an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor, and / or an NPU (Neural-Network Processing Unit).

[0210] The processor 810 can be used to execute executable instructions stored in the memory 820 to perform method steps of various embodiments of this disclosure, such as... Figure 1 The method and steps.

[0211] By executing the above method steps through processor 810, the following technical effects are achieved: Firstly, an intelligent algorithm-based disaster prevention and control scheme for underground caverns, guided by swarm effects, is provided. This breaks away from the approach of independent prevention and control for single caverns in related technologies, integrating swarm effects throughout the entire disaster prevention and control process for underground caverns. This adapts to the clustered layout characteristics of underground cavern groups, ensuring that cavern assessment results closely approximate actual conditions, and that corresponding prevention and control measures match the intensity of swarm effects. This improves the accuracy and effectiveness of prevention and control. Secondly, by quantifying and fusing static and dynamic swarm effects to obtain a comprehensive swarm effect coefficient, and extracting features from it and monitoring data, the assessment results of the target cavern and corresponding disaster prevention and control measures are determined based on the obtained target features. This improves the accuracy and effectiveness of prevention and control, and can adapt to engineering scenarios with large cavern groups, complex structures, and numerous disaster factors, meeting the disaster prevention and control needs of large-scale water conservancy projects.

[0212] Bus 830 is used to connect different components of electronic device 800 and may include data bus, address bus and control bus.

[0213] Electronic device 800 can communicate with one or more external devices 900 (such as keyboard, mouse, external controller, etc.) through I / O interface 840.

[0214] Electronic device 800 can communicate with one or more networks via network adapter 850. For example, network adapter 850 can provide mobile communication solutions such as 3G / 4G / 5G, or wireless communication solutions such as wireless LAN, Bluetooth, and near-field communication. Network adapter 850 can communicate with other modules of electronic device 800 via bus 830.

[0215] In one embodiment, the electronic device 800 further includes a display for displaying a graphical user interface.

[0216] although Figure 8 As not shown in the diagram, other hardware and / or software modules may also be configured in the electronic device 800, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (Redundant Arrays of Independent Disks) systems, tape drives, and data backup storage systems.

[0217] As can be seen from the above, the technical solutions disclosed herein can be implemented as methods, apparatus, systems, computer program products, storage media, electronic devices, etc. Those skilled in the art will understand that various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects. Exemplarily, these three forms can be referred to as "circuit," "module," and "system," respectively.

[0218] It should be understood that this disclosure is not limited to the specific methods, steps, or structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. Those skilled in the art will readily conceive of other embodiments based on the specific implementations provided in this disclosure. Therefore, the specific implementations provided in this disclosure are merely exemplary, and the scope and spirit of this disclosure are indicated by the claims, and should cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary technical means in the art not disclosed in this disclosure.

Claims

1. A method for preventing and controlling underground cavern disasters based on swarm effects, characterized in that, An underground cavern complex used in water conservancy projects, wherein the underground cavern complex includes a target cavern; the method includes: In both cases, considering the inter-cavity influence and not considering the inter-cavity influence, the first parameter of the target cavern is obtained respectively, and the static group effect coefficient is determined based on the first parameter in the two cases. The dynamic group effect coefficient is determined based on the transmission between various disasters that may occur in the underground cavern group; By combining the static group effect coefficient and the dynamic group effect coefficient, a comprehensive group effect coefficient is obtained; Initial monitoring data of the target cavern is acquired, and fused monitoring data is obtained through data fusion processing. Feature extraction is performed on the integrated group effect coefficient and the fused monitoring data to obtain target features; Based on the target characteristics, an assessment result of the target cavern is obtained, and disaster prevention and control measures are determined based on the assessment result; The step of extracting features from the integrated group effect coefficient and the fused monitoring data to obtain target features includes: merging the integrated group effect coefficient and the fused monitoring data to obtain a first sequence; performing grouped convolution processing on the first sequence to obtain multiple groups of first intermediate features; performing depthwise separable convolution processing on the multiple groups of first intermediate features to obtain second intermediate features; performing activation processing on the second intermediate features to obtain third intermediate features; using embedding weights to weight the third intermediate features with the integrated group effect coefficient to obtain fused features; and processing the fused features using a time-series model to obtain the target features.

2. The method according to claim 1, characterized in that, The first parameter includes the maximum principal stress around the tunnel and the tunnel convergence; determining the static group effect coefficient based on the first parameter in the two cases includes: Calculate the ratio of the maximum principal stress around the target cavern in the two cases, and the ratio of the cavern convergence in the two cases, and obtain the static group effect coefficient based on the two ratios.

3. The method according to claim 2, characterized in that, The process of obtaining the static group effect coefficient based on the two ratios includes: The static group effect coefficient is determined using the following formula: ； in, λ This represents the static group effect coefficient. α , β Let be the weighting coefficient, satisfying α + β =1; σ group To determine the maximum principal stress around the target cavern considering the influence between caverns, σ single The maximum principal stress around the target cavern is determined without considering the influence between caverns. u group To determine the cavern convergence of the target cavern considering the influence between caverns, u single This refers to the cavern convergence of the target cavern without considering the influence between caverns.

4. The method according to claim 2, characterized in that, The process of obtaining the first parameters of the target cavern under both the case of considering inter-cave chamber influence and the case of not considering inter-cave chamber influence includes: Based on the periphery stress and cavern convergence of the target cavern detected by the sensor, the maximum principal stress and cavern convergence of the target cavern are determined, taking into account the influence between caverns. A simulation was performed on a single-cavity scenario of the target cavern, and the maximum principal stress around the cavern and the cavern convergence were calculated based on the simulation results, without considering the influence between caverns.

5. The method according to claim 1, characterized in that, The dynamic group effect coefficient includes the transmission coefficient between different disasters; determining the dynamic group effect coefficient based on the transmission between various disasters that may occur in the underground cavern group includes: The transmission probabilities between different disasters are obtained under two scenarios: considering the influence between caverns and not considering the influence between caverns. The transmission coefficient between different disasters is determined based on the ratio of the transmission probabilities between different disasters under the two conditions.

6. The method according to claim 5, characterized in that, The process of obtaining the comprehensive group effect coefficient by combining the static group effect coefficient and the dynamic group effect coefficient includes: The combined group effect coefficient is determined using the following formula: ； in, Γ This represents the combined group effect coefficient. λ This represents the static group effect coefficient. n The number of possible disaster types in the aforementioned underground cavern complex. i Indicates the first i disasters, Indicates from the first i Disaster transmission to the first i The transmission coefficient of +1 type disasters, the ordinal number of each disaster is determined by the disaster chain.

7. The method according to claim 1, characterized in that, The process of acquiring initial monitoring data of the target cavern and obtaining fused monitoring data through data fusion processing includes: The initial monitoring data is obtained by detecting multiple different types of sensors; Outliers are removed from the initial monitoring data and the data is standardized to obtain standardized monitoring data. The standardized monitoring data is weighted and fused based on the distance between each sensor and the center of the target chamber to obtain the fused monitoring data.

8. The method according to claim 7, characterized in that, The step of removing outliers from the initial monitoring data includes: For each type of initial monitoring data, the range of numerical fluctuation is determined based on the average value and standard deviation of the initial monitoring data within the first time window; A second time window is determined with the acquisition time of each numerical point in the initial monitoring data as the center, and a reference value of the initial monitoring data within the second time window is determined. Values ​​that are outside the range of the numerical fluctuations and whose absolute difference from the reference value is greater than the trend consistency threshold will be removed.

9. The method according to claim 7, characterized in that, The standardized monitoring data is weighted and fused based on the distances of each sensor to the center of the target chamber to obtain the fused monitoring data, including: The standardized monitoring data is weighted and fused using the following formula to obtain the fused monitoring data: ； ； in, x fusion This refers to the fused monitoring data. k Indicates the first k One sensor, m This indicates the number of sensors of the same type on the same monitoring section. w k1 Indicates the first k Spatial correlation weights of individual sensors w k2 Indicates the first k The parameter sensitivity weights of each sensor x norm,k Indicates the first k Standardized monitoring data from each sensor dis k Indicates the first k The distance between each sensor and the center of the target chamber.

10. The method according to any one of claims 1 to 9, characterized in that, The fused monitoring data includes mechanical parameters, deformation parameters, seepage parameters, and environmental parameters; the first sequence is subjected to grouped convolution processing to obtain multiple sets of first intermediate features, including: The data in the first sequence is divided into three groups, and each group is convolved to obtain the first intermediate feature corresponding to each group; wherein, the mechanical parameters, deformation parameters, and the comprehensive group effect coefficient belong to the first group, the seepage parameters belong to the second group, and the environmental parameters belong to the third group.

11. The method according to any one of claims 1 to 9, characterized in that, The step of performing depthwise separable convolution processing on the multiple sets of first intermediate features to obtain second intermediate features includes: Multiple scales of convolution kernels are used to perform depthwise separable convolution processing on the multiple sets of first intermediate features to obtain second intermediate features of multiple scales; The process of activating the second intermediate feature to obtain the third intermediate feature includes: The second intermediate features at the multiple scales are merged, and the merged result is activated to obtain the third intermediate feature.

12. The method according to any one of claims 1 to 9, characterized in that, The method of weighting the third intermediate feature with the comprehensive group effect coefficient using embedding weights to obtain the fused feature includes: The comprehensive group effect coefficient is numerically adjusted and normalized, and the embedding weight is determined based on the normalization result; The third intermediate feature and the comprehensive group effect coefficient are dimensionally unified, and the embedded weights are used to weight the dimensionally unified third intermediate feature and the comprehensive group effect coefficient to obtain the fused feature.

13. The method according to any one of claims 1 to 9, characterized in that, The assessment results of the target cavern include the predicted probability of disaster occurrence and the predicted time of disaster occurrence; the assessment results of the target cavern obtained based on the target characteristics include: The target features are processed using a fully connected layer and an activation function to obtain the predicted probability of disaster occurrence for the target cavern. The target features are processed using a linear regression layer to obtain the predicted time of disaster occurrence for the target cavern.

14. The method according to any one of claims 1 to 9, characterized in that, The assessment results for the target cavern include the predicted probability of disaster occurrence and the predicted time of disaster occurrence; the determination of disaster prevention and control measures based on the assessment results includes: Based on the comprehensive group effect coefficient, the predicted probability of disaster occurrence, and the predicted time of disaster occurrence, the corresponding disaster prevention and control measures are determined. The method further includes: After implementing the disaster prevention and control measures, the measures are evaluated based on the change in the comprehensive group effect coefficient, the change in the predicted probability of disaster occurrence, and the response time.

15. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 14.

16. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1 to 14 by executing the executable instructions.

Images