Karst tunnel water gushing monitoring method and system based on multi-modal data fusion

Abstract

The invention relates to the technical field of tunnel engineering safety monitoring, and discloses a karst tunnel water gushing monitoring method and system based on multi-modal data fusion, and the method comprises the steps: collecting multi-modal monitoring data; performing intra-modal characteristic decomposition and model residual construction; constructing a time sequence causal constraint index and a joint feature vector; constructing a water burst risk fusion index; and executing abnormal water burst early warning by adopting a multi-stage early warning judgment mechanism. In the prior art, a scheme of carrying out water burst early warning mainly depending on a single water level or a flow threshold value, and particularly under the condition of strong coupling of karst tunnel surrounding rock structure response and a hydrological process, the technical problem that multi-stage early warning is difficult to accurately carry out on the water burst risk is solved. According to the method, the multi-modal features, the time sequence causal constraint and the self-adaptive weight modulation mechanism are introduced, so that the comprehensive perception of the surrounding rock structure change and hydrological anomaly co-evolution process is realized, and the accuracy of karst tunnel water gushing monitoring is improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, and in particular to a method and system for monitoring water inrush in karst tunnels based on multimodal data fusion. Background Technology

[0002] Currently, karst tunnels, due to the presence of caves, dissolution fissures, and hidden groundwater channels in the surrounding rock, exhibit a complex coupling relationship between their structural stability and groundwater activity. Under the combined influence of multiple factors such as rainfall recharge, groundwater level fluctuations, and stress redistribution in the surrounding rock, water inrush processes in karst tunnels often exhibit characteristics such as suddenness, complex evolution paths, and uncertain impact range. Once an abnormal water inrush occurs, it can easily pose a serious threat to the structural and operational safety of the tunnel.

[0003] Existing technologies for monitoring and early warning of tunnel water inrush mainly rely on changes in drainage ditch water levels or local water levels as the primary basis for monitoring and judgment, with some schemes supplemented by flow rate monitoring or empirical threshold judgment. However, these methods generally suffer from problems such as limited monitoring parameters and insufficient perception of the response of surrounding rock structures. For example, in karst tunnels, seepage from cavities or fissures may not cause a significant rise in drainage water levels in the initial stage, but structural responses or seepage path adjustments may have already occurred within the surrounding rock. In this case, relying solely on water level changes is insufficient to reflect potential water inrush risks in a timely manner, easily leading to delayed early warnings.

[0004] In addition, due to factors such as differences in drainage capacity, changes in cross-sectional conditions, and local blockages, there is often no stable one-to-one correspondence between drainage water level and actual water inrush risk. Early warning methods based solely on water level or flow thresholds are easily affected by short-term disturbances and noise in practical applications, resulting in high false alarm rates and frequent fluctuations in early warning status, which makes it difficult to meet the monitoring needs for the long-term stable operation of karst tunnels.

[0005] Therefore, there is an urgent need for a monitoring method that can comprehensively perceive changes in the surrounding rock structure and the evolution of hydrological processes, and make stable judgments and graded early warnings of water inrush risk, even when the response of the surrounding rock structure in karst tunnels is highly coupled with the groundwater dynamic process, and the monitoring signals are multi-source heterogeneous and have complex temporal relationships, so as to improve the accuracy and reliability of water inrush monitoring and early warning in karst tunnels. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a karst tunnel water inrush monitoring method based on multimodal data fusion. This method aims to solve the technical problem that existing technologies mainly rely on a single water level or flow threshold for water inrush early warning, especially under the condition of strong coupling between the surrounding rock structure response and hydrological process in karst tunnels, making it difficult to accurately provide multi-level early warning of water inrush risk.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention proposes a method for monitoring water inrush in karst tunnels based on multimodal data fusion.

[0008] The method for monitoring water inflow in karst tunnels based on multimodal data fusion includes:

[0009] Step S10: Collect multimodal data of karst tunnels, and perform synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and output karst tunnel monitoring dataset D;

[0010] Step S20: Based on the karst tunnel monitoring dataset D, perform the intra-modal feature extraction task using intra-modal feature decomposition and model residual construction mechanisms, and output the intra-modal feature set. ;

[0011] Step S30: Based on the intra-modal feature set The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. ;

[0012] Step S40: Based on joint feature vector An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. ;

[0013] Step S50: Based on the fusion index of inrush risk A multi-level early warning and judgment mechanism is adopted to perform the task of early warning of abnormal water inrush and output the early warning information of abnormal water inrush.

[0014] Preferably, step S10, which involves collecting multimodal data of karst tunnels and performing synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and outputting the karst tunnel monitoring dataset D, specifically includes:

[0015] Step S101: Collect multimodal data of the karst tunnel at time t within the karst tunnel to be monitored. The multimodal data of the karst tunnel includes the original time series of fiber optic strain of the surrounding rock. Original time series of water flow at the drainage ditch cross-section Original time series of water levels at the drainage ditch cross-section The initial time calibration process was performed on the multimodal data of karst tunnels to form a set of original time series of multimodal data to be reconstructed.

[0016] Step S102: First, obtain a physical evolution sequence model of the surrounding rock structure response, water level change, and water flow change during the water inflow process in the karst tunnel. Based on the physical evolution sequence model, perform cross-correlation analysis on the multimodal original time series set using a cross-correlation analysis method with monotonic causal constraints, and output the first time lag compensation amount of water level relative to surrounding rock strain. The second time lag compensation amount of water flow rate relative to surrounding rock strain Among them, the amount of compensation delayed in the first instance. Second time lag compensation amount The constraints are satisfied: ;

[0017] Step S103: Based on the first time lag compensation amount Second time lag compensation amount The original multimodal time series dataset is subjected to lag compensation mapping using an asymmetric time axis mapping method, and the final output is the karst tunnel monitoring dataset D.

[0018] Preferably, in step S20, based on the karst tunnel monitoring dataset D, an intra-modal feature extraction task is performed using intra-modal feature decomposition and model residual construction mechanisms to output an intra-modal feature set. The steps specifically include:

[0019] Step S201: Based on the karst tunnel monitoring dataset D, perform trend residual decomposition using the STL-based decomposition method to obtain low-frequency trend components. With high-frequency residual components Among them, the low-frequency trend component Used to characterize the slow evolution of the surrounding rock structure over the monitoring timescale; high-frequency residual components Used to characterize the transient disturbance process of the surrounding rock structure within the monitoring time scale;

[0020] Step S202: Determine the time series of the treatment flow rate of the drainage ditch section based on the karst tunnel monitoring dataset D and the preset hydraulic parameter template set. Based on the original time series of water flow at the cross-section of the drainage ditch The flow time series of drainage ditch cross-section treatment Constructing deterministic model residuals using the direct difference method ;

[0021] Step S203: Fuse the deterministic model residuals using a feature-level concatenation method. Low-frequency trend components With high-frequency residual components The final output is the intra-modal feature set. .

[0022] Preferably, in step S30, based on the intra-modal feature set The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. The steps specifically include:

[0023] Step S301: Within the preset sliding time window W, statistically analyze the intramodal feature set. Deterministic model residuals at a given time t Low-frequency trend components With high-frequency residual components The corresponding set of in-window statistics includes the residuals of the deterministic model. Arithmetic mean within the window statistic, low-frequency trend component Maximum in-window statistics and high-frequency residual components Arithmetic mean in-window statistics; and construct a preliminary joint feature vector based on the set of in-window statistics;

[0024] Step S302: Within the same sliding time window W, analyze the low-frequency trend components. With high-frequency residual components The low-frequency cross-correlation peak value was calculated using a cross-correlation analysis method based on Hilbert transform. and high-frequency cross-correlation peak ;

[0025] Step S303: Targeting low-frequency cross-correlation peak values and high-frequency cross-correlation peak Introducing causal inequality constraints, specifically: And based on the causal inequality constraint, a time-series constraint index T is defined. Time-series constrained index T Specifically: ,in, This is an indicator function used to map temporal sequence relationships to discrete temporal constraint indices;

[0026] Step S304: Finally, the low-frequency cross-correlation peak value is calculated. High-frequency cross-correlation peak and time-series constrained index T Defined as an additional constraint quantity, and constructing a joint eigenvector based on the additional constraint quantity and the initial joint eigenvector. .

[0027] Preferably, in step S40, based on the joint feature vector An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. The steps specifically include:

[0028] Step S401: First, process the joint feature vector The Min–Max normalization function is used to perform normalization processing to obtain the normalized joint feature vector. ; and preset the cross-correlation peak value with low frequency respectively. and high-frequency cross-correlation peak The corresponding first weighted gating factor Second weighted gating factor ;

[0029] Step S402: From the joint feature vector Extracting the temporal constraint index T Based on time-series constraint index T For the first weighted gating factor Second weighted gating factor Adaptive modulation is performed to output optimized first weight gate factor and optimized second weight gate factor;

[0030] Step S403: Apply the optimized first weight gating factor and the optimized second weight gating factor, combined with the normalized joint feature vector. A weighted linear fusion method is used to calculate the inrush risk fusion and output the inrush risk fusion index. .

[0031] Preferably, in step S402, based on the time-series constraint index T For the first weighted gating factor Second weighted gating factor The steps for performing adaptive modulation and optimizing the first weighted gating factor and the second weighted gating factor specifically include:

[0032] When the sequence constraint index T When the value is 1, it is determined that the current sliding time window is in a state of "structural response precedes hydrological response" physical causal order, and at the same time, the first weighted gating factor is... Enhanced modulation based on the Sigmoid mapping function is performed, and the optimized first weight gate factor is output.

[0033] When the sequence constraint index T When the value is 0, it is determined that the current sliding time window does not satisfy the physical causal order of "structural response precedes hydrological response", and the second weighted gating factor is adjusted accordingly. Suppression modulation based on the Sigmoid mapping function is performed, and the output is an optimized second weighted gating factor.

[0034] Preferably, in step S50, the fusion index based on water inrush risk is used. The steps for implementing an anomaly warning task using a multi-level early warning judgment mechanism and outputting anomaly warning information specifically include:

[0035] Step S501: Integrate inrush risk indicators Compared with the preset multi-level risk threshold range, the multi-level risk threshold range includes the normal risk threshold range, the first-level warning risk threshold range, the second-level warning risk threshold range, and the third-level warning risk threshold range;

[0036] Step S502: When When the current water inflow is normal, it is determined that the current water inflow is in a normal state; when At that time, the current abnormal water inrush is determined to be in a Level 1 warning state; when At that time, the current abnormal water inrush was determined to be in a level two warning state; when At that time, the current abnormal water inrush was determined to be in a level three warning state; among which, This is the first statistical quantile threshold constructed based on historical normal operating condition data; This is the second statistical quantile threshold constructed based on historical normal operating condition data; The third statistical quantile threshold is constructed based on historical normal operating condition data;

[0037] Specifically, when a transition occurs in the warning state, the corresponding level of abnormal water inrush warning state is triggered only if the water inrush risk fusion index R continuously crosses the corresponding risk threshold range within three or more consecutive adjacent sliding time windows; if the water inrush risk fusion index R falls back to below the corresponding risk threshold range within any sliding time window, the current warning state is maintained and no transition occurs.

[0038] Step S503: Based on the judgment result of the early warning status, the final output is the early warning information of abnormal water inrush.

[0039] This invention also provides a karst tunnel water inrush monitoring system based on multimodal data fusion, comprising:

[0040] The unified time reference synchronous reconstruction module is used to collect multimodal data of karst tunnels and perform synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and outputs karst tunnel monitoring dataset D.

[0041] The intramodal eigenvalue decomposition and residual construction module is used to perform intramodal feature extraction based on the karst tunnel monitoring dataset D using intramodal eigenvalue decomposition and model residual construction mechanisms, and outputs an intramodal feature set. ;

[0042] The temporal causal constraint indexing and joint feature construction module is used for constructing features based on intra-modal feature sets. The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. ;

[0043] An adaptive weighted modulation risk fusion module is used for fusion based on joint feature vectors. An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. ;

[0044] A multi-level early warning judgment and early warning information output module is used to determine the risk of water inrush based on the fusion index. A multi-level early warning and judgment mechanism is adopted to perform the task of early warning of abnormal water inrush and output the early warning information of abnormal water inrush.

[0045] The present invention also provides a karst tunnel water inrush monitoring device based on multimodal data fusion, comprising: a memory, a processor, and a karst tunnel water inrush monitoring program based on multimodal data fusion stored in the memory and executable on the processor. When the karst tunnel water inrush monitoring program based on multimodal data fusion is executed by the processor, a karst tunnel water inrush monitoring method based on multimodal data fusion is implemented.

[0046] The present invention also provides a computer program product, including a karst tunnel water inrush monitoring program based on multimodal data fusion, wherein the karst tunnel water inrush monitoring program based on multimodal data fusion implements the karst tunnel water inrush monitoring method based on multimodal data fusion when executed by a processor.

[0047] The beneficial effects of this invention are as follows: By introducing a multimodal data fusion and temporal causal constraint mechanism, this invention can collaboratively analyze the surrounding rock structure strain information and hydrological monitoring information such as water level and flow rate under a unified time reference, which can effectively depict the real temporal relationship of "structural response - hydrological response - risk evolution" during the water inrush process of karst tunnels.

[0048] This invention constructs a water inrush risk fusion index based on adaptive weight modulation and combines multi-level early warning judgment with continuous time window constraints to achieve hierarchical identification and stable output of water inrush risk levels. This avoids the problem of frequent changes in early warning status in traditional fixed weight or instantaneous threshold judgment methods, making water inrush anomaly early warning more consistent with the gradual evolution characteristics of water inrush risk in karst tunnels, thereby improving the safety assurance capability and engineering practicality during tunnel operation. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the first embodiment of the karst tunnel water inrush monitoring method based on multimodal data fusion of the present invention.

[0051] Figure 2 This is a schematic diagram of the equipment for the karst tunnel water inrush monitoring method based on multimodal data fusion according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the karst tunnel water inrush monitoring method based on multimodal data fusion of the present invention, which proposes the first embodiment of the karst tunnel water inrush monitoring method based on multimodal data fusion of the present invention.

[0054] In the first embodiment, the method for monitoring water inflow in karst tunnels based on multimodal data fusion includes:

[0055] Step S10: Collect multimodal data of karst tunnels, and perform synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and output karst tunnel monitoring dataset D;

[0056] It should be noted that the "unified time reference reconstruction mechanism based on physical response sequence constraints" in this step refers to the following: when synchronizing the multimodal monitoring data of karst tunnels, it is not simply aligned based on the timestamps of each sensor, but rather the physical sequence relationship between the response of the surrounding rock structure and the hydrological response is introduced as a constraint condition to reconstruct the time axis of different modal data. This mechanism includes at least the following: a process of time alignment offset correction of water level and flow signals based on the surrounding rock strain signal, and a process of resampling multimodal data based on a unified time step to form synchronized data tuples, thereby generating a karst tunnel monitoring dataset D with a unified time reference.

[0057] It should be understood that, compared with the direct alignment method based on sensor local timestamps or fixed sampling frequencies commonly used in traditional tunnel water inrush monitoring technology, this step, by introducing physical response sequence constraints, can effectively solve the practical problem in karst tunnels where "the response of the surrounding rock structure occurs first, while changes in water level and flow rate lag behind," reducing false correlations or missed judgments caused by simple time alignment. This significantly improves the consistency of multimodal monitoring data in the time dimension and its effectiveness in risk analysis, enhancing the reliability of subsequent water inrush early warning results.

[0058] For example, in the actual operation of karst tunnels, when the dissolution fissures inside the tunnel surrounding rock expand or the seepage channels are adjusted, the surrounding rock strain monitoring data usually shows continuous but small-amplitude changes in the early stages, while significant changes in drainage ditch water level and flow rate often only gradually appear after several minutes or even longer. If the traditional synchronization method based on the original timestamp is used, it is easy to separate strain changes from subsequent hydrological changes, making it difficult to identify potential water inrush risks in a timely manner. However, through the unified time base reconstruction mechanism in this step, the above-mentioned multimodal data can be aligned to the same physical evolution stage, so that the surrounding rock strain anomalies can be included in the subsequent analysis process before the water level and flow rate change significantly, thus providing effective support for the early identification of water inrush risks.

[0059] Step S20: Based on the karst tunnel monitoring dataset D, perform the intra-modal feature extraction task using intra-modal feature decomposition and model residual construction mechanisms, and output the intra-modal feature set. ;

[0060] It should be noted that the "intramodal feature decomposition and model residual construction mechanism" in this step refers to: for different modal data in the karst tunnel monitoring dataset D, without cross-modal fusion, constructing corresponding feature decomposition models within each modality, and constructing residual information reflecting the degree of abnormal deviation based on the models; wherein, the modes include at least the surrounding rock strain mode, water level mode, and water flow mode, and the feature set within the modality includes at least the trend features, fluctuation features obtained from the decomposition of the original monitoring sequence, and residual features constructed based on physical or empirical models.

[0061] It should be understood that, compared to directly using the original water level, flow rate, or strain values ​​as input for risk assessment, this step, by introducing intramodal eigenvalue decomposition and model residual construction, can effectively eliminate the influence of background trends formed during the long-term stable operation of karst tunnels, and highlight the deviation characteristics caused by changes in surrounding rock structure or abnormal hydrological processes. This significantly improves the ability of subsequent water inrush risk analysis to identify early abnormal signals and reduces the risk of misjudgment caused by background fluctuations or measurement noise.

[0062] Step S30: Based on the intra-modal feature set The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. ;

[0063] It should be noted that the "temporal causal construction mechanism" in this step refers to: based on the obtained set of features within the modalities, not only analyzing whether there is a correlation between different modal features, but also further analyzing their sequential relationship in the time dimension. By performing lag analysis on the time series of each modal feature within a sliding time window, a temporal constraint index is constructed that can characterize whether "features that change first have indicative significance for features that change later". Among them, the temporal sequence constraint index is at least used to characterize the sequential relationship of the response of the surrounding rock structure-related features to the water level or water flow features.

[0064] It is understandable that by introducing a temporal causal construction mechanism, the relationship between different modal features is no longer limited to whether the numerical changes are synchronous, but is further reflected as the evolutionary order of "which modal change occurs first and which modal change occurs later".

[0065] For example, in the monitoring of karst tunnels, when dissolution fissures expand within the surrounding rock, strain-related characteristics typically show continuous changes within an earlier time window, while water level or flow characteristics gradually respond after several time windows. If analysis is based solely on correlation coefficients, the two may be deemed insufficiently correlated due to the small magnitude of numerical changes. However, through the temporal causal construction mechanism in this step, a stable temporal relationship where "strain characteristic changes precede hydrological characteristic changes" can be identified. Based on this, a temporal sequence constraint index is generated, explicitly encoding this sequence relationship into the joint feature vector. This allows the subsequent risk fusion process to prioritize feature combinations with physical directional significance, thereby reflecting potential water inrush risks earlier.

[0066] Step S40: Based on joint feature vector An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. ;

[0067] It should be noted that the "adaptive weight modulation mechanism" in this step refers to the following: when constructing the fusion index for water inrush risk, instead of assigning a preset, unchanging weight to each feature component in the joint feature vector, the contribution weight of each feature component in the risk fusion process is dynamically modulated in conjunction with the temporal sequence constraint index generated in step S30. The weight modulation is used to reflect at least the differences in the ability of surrounding rock structural features and hydrological features to indicate water inrush risk at different evolution stages, thereby forming a risk fusion weight structure that changes over time.

[0068] Understandably, by introducing an adaptive weight modulation mechanism, the inrush risk fusion index can be dynamically adjusted according to the physical significance of different characteristics under the current monitoring state. When the characteristics related to the surrounding rock structure change before the hydrological characteristics, their weight in risk fusion will be increased accordingly. Conversely, when the changes in hydrological characteristics lack clear structural response support, their weight will be suppressed. For example, when the surrounding rock strain characteristics show slow cumulative changes over multiple consecutive time windows, while the water level and flow characteristics have not yet shown obvious anomalies, the traditional fixed weight method often underestimates the risk due to the insignificant changes in hydrological parameters. Through the adaptive weight modulation mechanism, the weight of the surrounding rock structure characteristics at this stage will be dynamically increased, causing the inrush risk fusion index to rise earlier. This allows the potential inrush risk to be reflected before significant changes in water level and flow occur, providing a more sufficient lead time for subsequent early warning judgments.

[0069] Step S50: Based on the fusion index of inrush risk A multi-level early warning and judgment mechanism is adopted to perform the task of early warning of abnormal water inrush and output the early warning information of abnormal water inrush.

[0070] It should be noted that the "multi-level early warning judgment mechanism" in this step refers to the following: based on the obtained water inrush risk fusion index, it does not simply make a binary judgment on water inrush anomalies based on a single threshold, but rather pre-constructs risk threshold intervals corresponding to multiple risk levels, and combines them with the continuity constraint in the time dimension to classify and judge the water inrush risk status; among them, the multi-level early warning includes at least a normal state, a first-level early warning state, a second-level early warning state, and a third-level early warning state, with different early warning states corresponding to different risk severity and early warning response levels.

[0071] It should be understood that, compared with the common instantaneous triggering method based on a single risk indicator and fixed threshold in traditional tunnel water inrush early warning technology, this step, by introducing multi-level risk intervals and time windows to continuously maintain constraints, can effectively avoid the problem of frequent or false triggering of early warnings due to short-term disturbances, sensor noise, or changes in local drainage conditions. It is particularly suitable for complex working conditions in karst tunnels where water inrush risk presents a phased accumulation and sudden enhancement, thereby significantly improving the stability and reliability of early warning results.

[0072] For example, during the operation of a karst tunnel, when the combined effects of surrounding rock structural characteristics and hydrological characteristics cause the fusion index of water inrush risk to gradually increase, the traditional single-threshold method may frequently trigger and deactivate warnings during short-term fluctuations in the risk index, interfering with normal operation and maintenance judgments. However, through the multi-level warning judgment mechanism in this step, the corresponding level of water inrush anomaly warning information will only be triggered when the fusion index of water inrush risk stably crosses the corresponding risk threshold range within multiple consecutive sliding time windows. This makes the warning results more consistent with the actual risk evolution law and provides more sufficient response time for taking targeted prevention and control measures.

[0073] Example 2: Furthermore, the karst tunnel water inrush monitoring system based on multimodal data fusion provided by this invention, employing the karst tunnel water inrush monitoring method based on multimodal data fusion in the above embodiments, can solve the technical problem of karst tunnel water inrush monitoring based on multimodal data fusion. The beneficial effects of the karst tunnel water inrush monitoring system based on multimodal data fusion provided by this invention are the same as those of the karst tunnel water inrush monitoring method based on multimodal data fusion provided in the above embodiments, and other technical features of the karst tunnel water inrush monitoring system based on multimodal data fusion are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0074] Example 3: This invention provides a karst tunnel water inrush monitoring device based on multimodal data fusion. Please refer to... Figure 2The karst tunnel water inrush monitoring device based on multimodal data fusion includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the karst tunnel water inrush monitoring method based on multimodal data fusion described in Embodiment 1 above. The karst tunnel water inrush monitoring device based on multimodal data fusion in this embodiment may include, but is not limited to, data sensing devices, fiber optic strain gauges, water level gauges, electronic water gauges, flow meters; data acquisition and transmission devices; mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. The karst tunnel water inrush monitoring device based on multimodal data fusion is merely an example and should not impose any limitations on the functionality and scope of use of this embodiment. The karst tunnel water inrush monitoring device based on multimodal data fusion may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the karst tunnel water inrush monitoring device based on multimodal data fusion. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the karst tunnel water inflow monitoring equipment based on multimodal data fusion to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows karst tunnel water inflow monitoring equipment based on multimodal data fusion with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0075] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for monitoring water inflow in karst tunnels based on multimodal data fusion. The computer program product provided by this invention can solve the technical problem of monitoring water inflow in karst tunnels based on multimodal data fusion. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for monitoring water inflow in karst tunnels based on multimodal data fusion provided in the above embodiments, and will not be repeated here.

[0076] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0077] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0078] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for monitoring water inrush in karst tunnels based on multimodal data fusion, characterized in that, The methods include: Step S10: Collect multimodal data of karst tunnels, and perform synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and output karst tunnel monitoring dataset D; Step S20: Based on the karst tunnel monitoring dataset D, perform the intra-modal feature extraction task using intra-modal feature decomposition and model residual construction mechanisms, and output the intra-modal feature set. ; Step S30: Based on the intra-modal feature set The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. ; Step S40: Based on joint feature vector An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. ; Step S50: Based on the fusion index of inrush risk A multi-level early warning and judgment mechanism is adopted to perform the task of early warning of abnormal water inrush and output the early warning information of abnormal water inrush.

2. The method for monitoring water inflow in karst tunnels based on multimodal data fusion as described in claim 1, characterized in that, Step S10 involves collecting multimodal data of karst tunnels and performing synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, outputting the karst tunnel monitoring dataset D. Specifically, this includes: Step S101: Collect multimodal data of the karst tunnel at time t within the karst tunnel to be monitored. The multimodal data of the karst tunnel includes the original time series of fiber optic strain of the surrounding rock. Original time series of water flow at the drainage ditch cross-section Original time series of water levels at the drainage ditch cross-section The initial time calibration process was performed on the multimodal data of karst tunnels to form a set of original time series of multimodal data to be reconstructed. Step S102: First, obtain a physical evolution sequence model of the surrounding rock structure response, water level change, and water flow change during the water inflow process in the karst tunnel. Based on the physical evolution sequence model, perform cross-correlation analysis on the multimodal original time series set using a cross-correlation analysis method with monotonic causal constraints, and output the first time lag compensation amount of water level relative to surrounding rock strain. The second time lag compensation amount of water flow rate relative to surrounding rock strain Among them, the amount of compensation delayed in the first instance. Second time lag compensation amount The constraints are satisfied: ; Step S103: Based on the first time lag compensation amount Second time lag compensation amount The original multimodal time series dataset is subjected to lag compensation mapping using an asymmetric time axis mapping method, and the final output is the karst tunnel monitoring dataset D.

3. The method for monitoring water inrush in karst tunnels based on multimodal data fusion as described in claim 2, characterized in that, In step S20, based on the karst tunnel monitoring dataset D, the intra-modal feature extraction task is performed using intra-modal feature decomposition and model residual construction mechanisms, outputting an intra-modal feature set. The steps specifically include: Step S201: Based on the karst tunnel monitoring dataset D, perform trend residual decomposition using the STL-based decomposition method to obtain low-frequency trend components. With high-frequency residual components Among them, the low-frequency trend component Used to characterize the slow evolution of the surrounding rock structure over the monitoring timescale; high-frequency residual components Used to characterize the transient disturbance process of the surrounding rock structure within the monitoring time scale; Step S202: Determine the time series of the treatment flow rate of the drainage ditch section based on the karst tunnel monitoring dataset D and the preset hydraulic parameter template set. Based on the original time series of water flow at the cross-section of the drainage ditch The flow time series of drainage ditch cross-section treatment Constructing deterministic model residuals using the direct difference method ; Step S203: Fuse the deterministic model residuals using a feature-level concatenation method. Low-frequency trend components With high-frequency residual components The final output is the intra-modal feature set. .

4. The method for monitoring water inflow in karst tunnels based on multimodal data fusion as described in claim 1, characterized in that, In step S30, based on the intra-modal feature set The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. The steps specifically include: Step S301: Within the preset sliding time window W, statistically analyze the intramodal feature set. Deterministic model residuals at a given time t Low-frequency trend components With high-frequency residual components The corresponding set of in-window statistics includes the residuals of the deterministic model. Arithmetic mean within the window statistic, low-frequency trend component Maximum in-window statistics and high-frequency residual components Arithmetic mean in-window statistics; and construct a preliminary joint feature vector based on the set of in-window statistics; Step S302: Within the same sliding time window W, analyze the low-frequency trend components. With high-frequency residual components The low-frequency cross-correlation peak value was calculated using a cross-correlation analysis method based on Hilbert transform. and high-frequency cross-correlation peak ; Step S303: Targeting low-frequency cross-correlation peak values and high-frequency cross-correlation peak Introducing causal inequality constraints, specifically: And based on the causal inequality constraint, a time-series constraint index T is defined. Time-series constrained index T Specifically: ,in, This is an indicator function used to map temporal sequence relationships to discrete temporal constraint indices; Step S304: Finally, the low-frequency cross-correlation peak value is calculated. High-frequency cross-correlation peak and time-series constrained index T Defined as an additional constraint quantity, and constructing a joint eigenvector based on the additional constraint quantity and the initial joint eigenvector. .

5. The method for monitoring water inrush in karst tunnels based on multimodal data fusion as described in claim 4, characterized in that, In step S40, based on the joint feature vector An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. The steps specifically include: Step S401: First, process the joint feature vector The Min–Max normalization function is used to perform normalization processing to obtain the normalized joint feature vector. ; and preset the cross-correlation peak value with low frequency respectively. and high-frequency cross-correlation peak The corresponding first weighted gating factor Second weighted gating factor ; Step S402: From the joint feature vector Extracting the temporal constraint index T Based on time-series constraint index T For the first weighted gating factor Second weighted gating factor Adaptive modulation is performed to output optimized first weight gate factor and optimized second weight gate factor; Step S403: Apply the optimized first weight gating factor and the optimized second weight gating factor, combined with the normalized joint feature vector. A weighted linear fusion method is used to calculate the inrush risk fusion and output the inrush risk fusion index. .

6. The method for monitoring water inrush in karst tunnels based on multimodal data fusion as described in claim 5, characterized in that, In step S402, based on the time-series constraint index T For the first weighted gating factor Second weighted gating factor The steps for performing adaptive modulation and optimizing the first weighted gating factor and the second weighted gating factor specifically include: When the sequence constraint index T When the value is 1, it is determined that the current sliding time window is in a state of "structural response precedes hydrological response" physical causal order, and at the same time, the first weighted gating factor is... Enhanced modulation based on the Sigmoid mapping function is performed, and the optimized first weight gate factor is output. When the sequence constraint index T When the value is 0, it is determined that the current sliding time window does not satisfy the physical causal order of "structural response precedes hydrological response", and the second weighted gating factor is adjusted accordingly. Suppression modulation based on the Sigmoid mapping function is performed, and the output is an optimized second weighted gating factor.

7. The method for monitoring water inflow in karst tunnels based on multimodal data fusion as described in claim 1, characterized in that, In step S50, based on the fusion index of water inrush risk The steps for implementing an anomaly warning task using a multi-level early warning judgment mechanism and outputting anomaly warning information specifically include: Step S501: Integrate inrush risk indicators Compared with the preset multi-level risk threshold range, the multi-level risk threshold range includes the normal risk threshold range, the first-level warning risk threshold range, the second-level warning risk threshold range, and the third-level warning risk threshold range; Step S502: When When the current water inflow is normal, it is determined that the current water inflow is in a normal state; when At that time, the current abnormal water inrush is determined to be in a Level 1 warning state; when At that time, the current abnormal water inrush was determined to be in a level two warning state; when At that time, the current abnormal water inrush was determined to be in a level three warning state; among which, This is the first statistical quantile threshold constructed based on historical normal operating condition data; This is the second statistical quantile threshold constructed based on historical normal operating condition data; The third statistical quantile threshold is constructed based on historical normal operating condition data; Specifically, when a transition occurs in the warning state, the corresponding level of abnormal water inrush warning state is triggered only if the water inrush risk fusion index R continuously crosses the corresponding risk threshold range within three or more consecutive adjacent sliding time windows; if the water inrush risk fusion index R falls back to below the corresponding risk threshold range within any sliding time window, the current warning state is maintained and no transition occurs. Step S503: Based on the judgment result of the early warning status, the final output is the early warning information of abnormal water inrush.

8. A karst tunnel water inrush monitoring system based on multimodal data fusion, applied to the karst tunnel water inrush monitoring method based on multimodal data fusion as described in any one of claims 1 to 7, characterized in that, The karst tunnel water inrush monitoring system based on multimodal data fusion includes: The unified time reference synchronous reconstruction module is used to collect multimodal data of karst tunnels and perform synchronous reconstruction processing on the multimodal data of karst tunnels using a unified time reference reconstruction mechanism based on physical response sequence constraints, and outputs karst tunnel monitoring dataset D. The intramodal eigenvalue decomposition and residual construction module is used to perform intramodal feature extraction based on the karst tunnel monitoring dataset D using intramodal eigenvalue decomposition and model residual construction mechanisms, and outputs an intramodal feature set. ; The temporal causal constraint indexing and joint feature construction module is used for constructing features based on intra-modal feature sets. The temporal causal construction mechanism is used to perform the tasks of generating temporal sequence constraint indexes and constructing joint feature vectors, and outputs the joint feature vectors. ; An adaptive weighted modulation risk fusion module is used for fusion based on joint feature vectors. An adaptive weight modulation mechanism is used to construct a fusion index for water inrush risk. ; A multi-level early warning judgment and early warning information output module is used to determine the risk of water inrush based on the fusion index. A multi-level early warning and judgment mechanism is adopted to perform the task of early warning of abnormal water inrush and output the early warning information of abnormal water inrush.

9. A karst tunnel water inrush monitoring device based on multimodal data fusion, characterized in that, The karst tunnel water inrush monitoring device based on multimodal data fusion includes: a memory, a processor, and a karst tunnel water inrush monitoring program based on multimodal data fusion stored in the memory and executable on the processor. When the karst tunnel water inrush monitoring program based on multimodal data fusion is executed by the processor, it implements the karst tunnel water inrush monitoring method based on multimodal data fusion as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a karst tunnel water inrush monitoring program based on multimodal data fusion, which, when executed by a processor, implements the karst tunnel water inrush monitoring method based on multimodal data fusion as described in any one of claims 1 to 7.

Images