A distributed optical fiber intelligent monitoring and early warning system and method for deep foundation pit construction

Abstract

The present application relates to the field of construction engineering structure health monitoring, and particularly relates to a distributed optical fiber intelligent monitoring and early warning system and method for deep foundation pit construction, which comprises the following steps: obtaining an optical fiber strain sequence and differentially extracting a gradient sequence; performing multi-scale space-frequency decomposition to obtain a space-frequency energy matrix; extracting a slowly varying component to construct a background model, calculating a logarithmic domain difference to generate a residual saliency map; statistically quantifying multi-scale gradient direction consistency to quantify edge confidence, and adaptively adjusting a smoothing scale to output refined features; combining feature purity and spatial dispersion to generate a comprehensive atlas through weighted fusion; and constructing an energy function considering spatial adjacent consistency and performing global optimization to solve the problem. The present application effectively improves the weak anomaly detection and early warning accuracy by segmenting physically coherent abnormal areas to trigger early warning.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring in construction engineering, and in particular to a distributed fiber optic intelligent monitoring and early warning system and method for deep foundation pit construction. Background Technology

[0002] In the field of safety and disaster prevention during deep foundation pit construction, distributed fiber optic sensing technology is widely used for physical deformation monitoring. Typically, differential operations are performed on the original strain sequence of the fiber optic cable to extract abrupt change features. However, foundation pits generally experience severe overall deformation caused by excavation and unloading. This high-energy, low-frequency, wide-bandwidth baseline can completely drown out the extremely weak high-frequency local signals induced by early micro-cracks. Simple spatial domain analysis cannot isolate aliased frequency bands, making it highly susceptible to missed hazard detection under strong background interference.

[0003] To suppress environmental interference in monitoring data, a global smoothing filter module with fixed parameters can be used. However, the critical load-bearing nodes of the deep foundation pit support system have a large number of inherent physical structural edges, which are accompanied by strong strain gradients. Using indiscriminate global fixed smoothing, while reducing random noise, will inevitably over-smooth or even completely erase the real weak signals attached to the edges, resulting in serious monitoring blind spots artificially created in high-risk stress concentration areas of the project.

[0004] Furthermore, traditional early warning systems heavily rely on absolute threshold segmentation based on fixed values ​​for warnings. However, engineering damage such as foundation pit cracking inevitably exhibits a coherent and clustered evolutionary pattern in space. Absolute threshold determination severs the physical adjacency between measuring points, making them highly susceptible to being misled by local transient pulses. This results in a large number of isolated and physically meaningless false early warning points, leading to extremely fragmented early warning results. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a distributed fiber optic intelligent monitoring and early warning system and method for deep foundation pit construction.

[0006] In a first aspect, the present invention provides a distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction, which adopts the following technical solution:

[0007] The optical fiber strain distribution sequence during the deep foundation pit construction process is obtained, and the optical fiber strain distribution sequence is subjected to spatial difference processing to extract the strain gradient sequence.

[0008] Multi-scale spatial frequency decomposition is performed on the strain gradient sequence to obtain a spatial frequency energy matrix that characterizes the intensity of different spatial frequency abrupt changes.

[0009] A preliminary environmental background model is constructed by extracting the gradual variation component of the space-frequency energy matrix, and a residual significance map is generated by calculating the logarithmic domain difference between the space-frequency energy matrix and the preliminary environmental background model.

[0010] The gradient direction consistency of the residual significance map at multiple scales is statistically analyzed to calculate the edge confidence. Based on this, the smoothing scale of the preliminary environmental background model is adaptively adjusted to output refined significance features.

[0011] The fusion weight is calculated by combining the information purity and spatial dispersion of the refined saliency features. The spatial dispersion is obtained by extracting the spatial coordinate distribution of the foreground response regions with values ​​greater than the mean in the refined saliency features and calculating their spatial distribution standard deviation. The weighted fusion generates a comprehensive saliency map.

[0012] A system total energy function that takes into account both significant numerical values ​​and spatial adjacency consistency is constructed and global optimization is performed to segment physically coherent abnormal regions to trigger construction safety linkage early warning.

[0013] Preferably, the step of performing multi-scale spatial-frequency decomposition on the strain gradient sequence to obtain a spatial-frequency energy matrix for characterizing the intensity of different spatial frequency abrupt changes includes: using a continuous wavelet with complex expression capabilities as the mother function, performing convolution transformation analysis on the strain gradient sequence on a preset spatial translation sequence and frequency band scale sequence; extracting the absolute amplitude of the complex coefficient values ​​of each transformation node, and generating the spatial-frequency energy matrix with the scale dimension and spatial dimension as analytical axes, so as to isolate the low-frequency overall deformation and high-frequency local damage caused by deep foundation pit construction.

[0014] Preferably, the step of constructing a preliminary environmental background model by extracting the gradual transformation component of the space-frequency energy matrix and calculating the logarithmic domain difference between the space-frequency energy matrix and the preliminary environmental background model to generate a residual significance map includes: using an anisotropic filtering module with a smoothing kernel width in the spatial dimension greater than that in the scale dimension to convolve the space-frequency energy matrix to obtain the preliminary environmental background model; adding a very small positive number to the values ​​of the preliminary environmental background model and the space-frequency energy matrix respectively to prevent logarithmic overflow, performing a logarithmic transformation operation, calculating the transformation difference between the two, and generating the residual significance map by amplifying the high-frequency abrupt signal in the logarithmic domain.

[0015] Preferably, the step of calculating the edge confidence by statistically analyzing the gradient direction consistency of the residual saliency map at multiple scales includes: calculating the two-dimensional spatial gradient direction and the average absolute gradient magnitude of the residual saliency map at multiple adjacent scales in a two-dimensional analytical domain composed of spatial and scale dimensions; counting the number of spatial gradient direction alignment scales that fall within a preset angle tolerance range at each physical spatial location, and determining the edge confidence used to identify the edge of the inherent structure of the foundation pit by multiplying the proportion of the number of alignment scales by the average absolute gradient magnitude.

[0016] Preferably, the step of adaptively adjusting the smoothing scale of the preliminary environmental background model and outputting refined saliency features includes: inputting the edge confidence into the smooth bounded contraction calculation model to obtain the kernel width adjustment coefficient corresponding to the physical spatial location; in the high confidence region determined to be the edge of the structure, dynamically contracting the spatial dimension of the preliminary environmental background model using the kernel width adjustment coefficient to smooth the kernel width in order to reduce the degree of environmental background suppression, and re-extracting the background residual based on the adjusted kernel width to generate the refined saliency features.

[0017] Preferably, the step of calculating the fusion weight by combining the information purity and spatial dispersion of the refined saliency features includes: mapping the refined saliency features at each scale into a probability distribution form, calculating the overall information entropy of its distribution state as the information purity; extracting the spatial coordinate distribution of the foreground response regions with values ​​greater than the mean in the refined saliency features, calculating its spatial distribution standard deviation as the spatial dispersion; selecting the scale that minimizes the product of the overall information entropy and the spatial distribution standard deviation, assigning it the highest numerical weight, and obtaining the fusion weight corresponding to each scale through exponential function normalization.

[0018] Preferably, the construction of the system's total energy function, which takes into account both saliency values ​​and spatial adjacency consistency, specifically includes constructing an anomaly determination data cost term and a spatial continuity penalty cost term: The numerical values ​​of the comprehensive saliency map are converted into the prospective probability of a single physical spatial node being in an anomalous damage state using a mapping function, and the negative logarithmic probabilities of forcibly labeling the corresponding physical spatial node as an anomalous state and a normal state are calculated respectively, serving as the anomaly determination data cost term, so that the numerical penalty for misjudging highly saliency regions as normal increases dramatically; the numerical comparison differences between adjacent physical spatial nodes in the comprehensive saliency map are extracted, and a contrast-sensitive computational feature is constructed. When adjacent physical spatial nodes have similar values ​​but are assigned completely mutually exclusive independent state labels, an exponentially increasing spatial continuity penalty cost term is generated to force the physical distribution coherence of anomalous regions; the system's total energy function is generated by combining the anomaly determination data cost term and the spatial continuity penalty cost term through a balance coefficient.

[0019] Preferably, the global optimization solution, which segments physically coherent abnormal regions to trigger construction safety linkage early warning, includes: transforming the total system energy function, which combines the anomaly judgment data cost term and the spatial continuity penalty cost term, into a network flow capacity allocation model in a directed graph structure; executing a network minimum cut search strategy to find the optimal partition boundary for cutting off extreme energy links; determining the output set of abnormal nodes as the physically coherent abnormal region; and executing a system early warning action.

[0020] Secondly, this invention provides a distributed fiber optic intelligent monitoring and early warning system for deep foundation pit construction, employing the following technical solution:

[0021] A distributed fiber optic intelligent monitoring and early warning system for deep foundation pit construction includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the distributed fiber optic intelligent monitoring and early warning method for deep foundation pit construction as described above.

[0022] The present invention has the following technical effects:

[0023] 1. The strain gradient is projected onto a two-dimensional plane to generate a space-frequency energy matrix, and an anisotropic filter kernel is used to extract a slowly varying background model. Subsequently, feature differences are calculated in the logarithmic domain to generate a residual significance map, thereby improving the significance of weak anomalies under strong disturbances.

[0024] 2. Extract the gradient direction and alignment scale number of the continuous frequency band to construct the edge confidence of the entity probability. Utilize a smooth bounded function to dynamically shrink the spatial filtering kernel width in this region, maintaining noise resistance while preserving minor damage at the edges of critical engineering structures.

[0025] 3. Dynamic weights are generated by jointly calculating information entropy and discreteness. Subsequently, a system total energy function is constructed that integrates the cost term of anomaly detection data and the cost term of spatial continuity penalty. Then, the optimal boundary is searched using a minimum cut topology model, and adjacent physical spatial nodes are assigned unified labels to reduce random noise interference and ensure that the anomaly region has a coherent entity form. Attached Figure Description

[0026] Figure 1 This is a flowchart of a distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to an embodiment of the present invention. Detailed Implementation

[0027] This invention discloses a distributed fiber optic intelligent monitoring and early warning method for deep foundation pit construction, referring to... Figure 1 The process includes steps S1-S6, as detailed below:

[0028] S1: Obtain the optical fiber strain distribution sequence during the deep foundation pit construction process, and perform spatial difference processing on the optical fiber strain distribution sequence to extract the strain gradient sequence.

[0029] Because the construction of deep foundation pits involves a mixture of large-scale, slowly varying strain components caused by the overall deformation of the foundation pit, and small-scale, high-frequency abrupt changes induced by local early damage, and because the energy of the slowly varying components is absolutely dominant, forming extremely strong low-frequency environmental noise, which severely masks weak abnormal signals, the anomaly determination of the original strain distribution sequence after optical fiber acquisition and benchmark correction will face serious baseline drift interference.

[0030] A spatial sequence difference mechanism is introduced for preliminary feature separation. When the foundation pit deforms, the large-scale deformation is a smooth, slowly varying function with a very small first-order spatial rate of change; while the local damage anomaly fluctuates violently in a small space and has an extremely high spatial rate of change.

[0031] Therefore, first-order difference is performed on the spatial sequence for data filtering to suppress low-frequency slowly varying baselines and significantly amplify high-frequency abrupt peaks, thus achieving preliminary signal noise reduction.

[0032] In one embodiment, the fiber strain distribution sequence of sensing fibers deployed along the foundation pit structure is obtained. Let the total number of sensing points be... The set of spatial coordinates is The fiber strain distribution sequence is a one-dimensional discrete vector containing strain values ​​at multiple sampling points.

[0033] To extract high-frequency abrupt change features, a first-order forward difference calculation was performed by traversing the fiber strain distribution sequence. This transformed the original absolute deformation data into a spatial gradient representation reflecting the local deformation intensity of the foundation pit, generating a strain gradient sequence representing the rate of change in physical space. , where the strain gradient sequence The calculation formula is:

[0034]

[0035] In the formula, Represents the first strain gradient in the differential strain gradient sequence. The gradient values ​​at each measuring point are used to represent the severity of sudden changes in local strain at the node.

[0036] and The first in the fiber strain distribution sequence The first one adjacent to the physical Static strain values ​​at each measuring point.

[0037] The physical distance between adjacent physical nodes depends on the hardware spatial resolution hyperparameter of the fiber optic demodulation equipment. Its experience value configuration is The amount of rice can be adjusted by the implementer according to the specific implementation scenario.

[0038] The strain gradient sequence is obtained by performing a first-order forward difference calculation. This can reduce long-term cumulative drift error.

[0039] S2: Perform multi-scale spatial frequency decomposition on the strain gradient sequence to obtain the spatial frequency energy matrix used to characterize the intensity of different spatial frequency abrupt changes.

[0040] After obtaining the strain gradient sequence after first-order difference processing, although the overall deformation background is initially suppressed, it is still deeply mixed with abrupt change modes induced by different damage mechanisms. For example, the spatial frequencies corresponding to microcracks and wide and gentle plastic deformation zones are very different. These modes are superimposed on each other in a single spatial domain, making it impossible to simultaneously and accurately determine the core spatial range of damage occurrence and its local energy distribution state, which seriously affects the accuracy of early risk identification under complex working conditions.

[0041] Since strain abrupt changes induced by different physical damages on optical fibers have different characteristic scales, micro-damage is rich in high-frequency components, and deformation during the development period is mainly of medium and low frequency, multi-scale localization analysis can be used to perform two-dimensional deconstruction of the frequency band space. That is, continuous basis functions with complex expression are used to transform the strain gradient sequence, so as to achieve adaptive matching of various signal characteristics and improve spatial positioning capability.

[0042] In one embodiment, the strain gradient sequence As the initial input feature vector, a continuous wavelet in complex form is selected as the mother function. and in the preset multi-band scale sequence and spatial translation sequences The initial input feature vector is subjected to discretized convolution transformation analysis.

[0043] Calculate the complex domain transform node values ​​at each physical measurement point location at each analysis scale, thereby generating a space-frequency energy matrix containing local signal structure phase information. The specific construction of the discrete convolution extraction calculation model for this matrix is ​​as follows:

[0044]

[0045] In the formula, It encompasses the complex domain mapping intensity of the signal in this frequency band; The scale parameter matrix, representing the bandwidth of the frequency analysis, is inversely proportional to the spatial frequency. This represents the translation parameter matrix corresponding to each physical measurement point.

[0046] Generating function Embedded center frequency hyperparameter To balance the local focusing characteristics in both spatial and frequency dimensions, this hyperparameter... The experience value is configured as 6, which can be adjusted by the implementer according to the specific implementation method.

[0047] To obtain the absolute intensity of high-frequency sudden changes caused by foundation pit anomalies, it is necessary to eliminate the positive and negative oscillations caused by complex phases.

[0048] Extracting the absolute magnitude of the complex coefficients at each transform node yields the space-frequency energy matrix. The root mean square of the sum of the squares of the real and imaginary parts of all elements is taken to generate the space-frequency energy matrix, which characterizes the intensity of different spatial frequency abrupt changes. .

[0049] The specific numerical conversion model for this space-frequency energy extraction feature is constructed as follows:

[0050]

[0051] In the formula, the space-frequency energy matrix It maps the energy intensity level of potential damage release in a specific spatial frequency band.

[0052] and These represent the real and imaginary physical components of the decomposition coefficients, respectively.

[0053] Thus, the low-frequency, wide-amplitude overall deformation caused by normal construction disturbance of the deep foundation pit, and the high-frequency, narrow-amplitude local damage induced by the expansion of micro-cracks, are separated and isolated on the two-dimensional analytical plane of scale and space.

[0054] S3: Construct a background model by extracting the gradual variation component of the space-frequency energy matrix, and generate a residual significance map by calculating the logarithmic domain difference between the space-frequency energy matrix and the background model.

[0055] After obtaining the aforementioned spatial frequency energy matrix characterizing the intensity of multi-band abrupt changes, the weak damage features in the two-dimensional analysis domain are easily submerged by the environmental baseline. As a result, the spatial frequency energy matrix not only includes the high-frequency abrupt change energy caused by local real damage, but also deeply includes the slowly varying energy background induced by the temperature field fluctuations of the foundation pit and normal construction disturbances.

[0056] These gradual change components often dominate in absolute value. If they are not stripped, the low-amplitude signals such as early micro-cracks will have an extremely low signal-to-noise ratio, which will lead to serious underreporting.

[0057] Since the gradually varying background exhibits a low-frequency smoothing trend on the two-dimensional analytical plane, and its rate of change in the spatial dimension is significantly lower than that in the scale dimension, the use of an asymmetric smoothing strategy to extract the background can avoid blurring the damage features.

[0058] Meanwhile, the logarithmic function has the property of compressing high amplitudes and stretching low amplitudes. Calculating the residual in the logarithmic domain can maximize the significance of weak damage signals relative to the environmental background.

[0059] In practical implementation, the aforementioned space-frequency energy matrix containing absolute energy values ​​is used. Based on this, a preliminary environmental background model is constructed to characterize slowly changing environmental disturbances. .

[0060] Call anisotropic 2D smooth convolution kernel For the space-frequency energy matrix Perform global sliding window convolution calculation, where the convolution kernel... Smooth kernel width parameter configured in the spatial dimension It must be strictly greater than the smooth kernel width parameter configured in the scale dimension. .

[0061] The specific formula for calculating the smooth extraction of asymmetric backgrounds is as follows:

[0062]

[0063] In the formula, the matrix This represents the initial physical background characterization that retains only the gradual trend of environmental change after stripping away the details of local mutations.

[0064] core The spatial dimension smoothing kernel width hyperparameter determines the background fitting span along the physical length of the optical fiber. The engineering empirical value is configured as 15 sampling nodes, which can be adjusted by the implementer according to the specific implementation scenario.

[0065] core The scale-smoothing kernel width hyperparameter is used to maintain the details of energy distribution across frequency bands, and its empirical value is configured as 1.5 scale resolution units.

[0066] Thus, by smoothing the asymmetric background, the local damage energy is preserved while isolating the overall interference.

[0067] After successfully extracting the slowly varying background, the process enters the high-contrast visualization stage of anomalous features, where the aforementioned spatial frequency energy matrix is ​​used. Preliminary environmental background model Add a very small positive number to each of the corresponding node values. Then, a natural logarithmic transformation operation is performed.

[0068] By subtracting the background model's logarithmic mapping value from the logarithmic mapping value of the energy matrix, the system ultimately generates a residual significance map for locating damage. .

[0069] The specific numerical transformation model for making the logarithmic domain residual explicit is constructed as follows:

[0070]

[0071] In the formula, the significance diagram of the residuals is shown. The significance of pure damage mutations after filtering out engineering baseline interference is presented intuitively.

[0072] function This represents a nonlinear logarithmic compression mapping.

[0073] To prevent the extremely small positive hyperparameter from causing mathematical calculation overflow due to zero amplitude, and to ensure the absolute stability of logarithmic operations, its empirical value is [value missing]. It can be adjusted by the implementer according to the specific implementation scenario.

[0074] By calculating the significance plot of the residuals This allows the extremely weak early crack anomalies within the deep foundation pit to exhibit a bright peak feature in the two-dimensional map.

[0075] S4: Calculate the gradient direction consistency of residual significance maps at multiple scales to determine the margin confidence, and adaptively adjust the smoothing scale of the background model accordingly to output refined significance features.

[0076] Strong gradient abrupt changes exist at the inherent physical edges of structures such as deep foundation pit support piles. When convolution is performed using a fixed spatial dimension smooth kernel width from the previous steps, such normal structural abrupt changes will be misjudged as environmental background and forcibly suppressed. This will severely weaken or even erase the real early weak damage signals attached to the physical edges, leading to missed reports in critical areas.

[0077] Since the edges of real foundation pit structures exhibit highly consistent gradient orientation characteristics across multiple spatial frequency bands, while randomly distributed environmental noise or spurious undulations show chaotic directional distribution across different scales, the inherent edges can be accurately identified by calculating the degree of cross-scale directional alignment, thereby dynamically shrinking the background smoothing scale of the region.

[0078] In specific implementation, select A saliency subset of consecutive adjacent analytical scales for each physical space node. In a two-dimensional analytic domain consisting of spatial and scale dimensions, the partial derivatives of the residual significance plot along the spatial coordinate axis and along the scale coordinate axis are calculated using the central difference method, and then the arctangent function is applied. Synthesize to obtain the two-dimensional spatial gradient direction of the corresponding node. And the absolute magnitude of the gradient.

[0079] Subsequently, an alignment tolerance hyperparameter at a specific angle is set. And statistically analyze the spatial gradient direction of the node across all selected scales. Total number of scale alignments that strictly fall within the stated tolerance range Among them, tolerance parameter The empirical value is set at 20 degrees, which can be adjusted by the implementer according to the specific implementation scenario.

[0080] By constructing a multi-scale edge feature confidence calculation model, the current physical space nodes are obtained. The physical probability of the existence of a real foundation pit structure edge.

[0081] Calculate the absolute magnitude of the local average gradient across scales and multiply it by the scale alignment percentage index to obtain the edge confidence sequence used to guide adaptive filtering. Marginal confidence sequence The calculation process is as follows:

[0082]

[0083] In the formula, sequence Reflects the nodes The probability of a continuous and stable structural edge existing at that location.

[0084] This represents the total number of scales that are aligned and not affected by random noise.

[0085] constant The total number of adjacent frequency band scales participating in the evaluation is set to an empirical value of 5, which can be adjusted by the implementer according to the specific implementation scenario.

[0086] function This represents the mean magnitude of the absolute magnitude sequence of local multi-scale gradients after performing an arithmetic mean operation, and multiplication ensures that high confidence is generated only at specific locations.

[0087] Obtain the edge confidence sequence Then, this index is input into a nonlinear smoothing bounded function to dynamically adjust the spatial dimension benchmark smoothing kernel width parameter in the preliminary environmental background model. .

[0088] The dynamic spatial filter kernel width at each physical measurement point is calculated using the hyperbolic tangent mapping mechanism. .

[0089] The residuals are then re-extracted based on the dynamic kernel width, ultimately generating refined saliency features that preserve complete edge damage. The specific calculation process is as follows:

[0090]

[0091] In the formula, the dynamic space filter kernel width It achieves smooth suppression and reduction of edge regions of high-confidence structures.

[0092] The fixed global core width baseline physical value set for the preceding steps.

[0093] coefficient The maximum shrinkage ratio hyperparameter of the kernel width limits the lower bound of the smoothing scale. The empirical value is set to 0.4, which can be adjusted by the implementer according to the specific implementation scenario.

[0094] coefficient To adjust the sensitivity hyperparameter for confidence level and control the steepness of the response, an empirical value of 3.5 is set, which can be adjusted by the implementer according to the specific implementation scenario.

[0095] function It is a hyperbolic tangent function, used to ensure that the contraction coefficient is bounded.

[0096] S5: Combine the information purity and spatial dispersion of refined saliency features to calculate the fusion weight, and generate a comprehensive saliency map by weighted fusion.

[0097] Because the response intensity and noise interference level of salient features under different analysis frequency bands to potential damage in the same foundation pit vary significantly, using equal-weighted averaging or a single optimal scale extraction strategy will fail to fully utilize the complementary redundancy information contained in multi-scale spatial signals. This can easily lead to the serious masking or omission of weak early damage responses in weak frequency bands, reducing the robustness of early warning assessment.

[0098] A high-quality foundation pit anomaly response scale should possess two physical characteristics: first, an information purity attribute with extremely high distinguishability between target damage and background noise; and second, a highly clustered attribute in the spatial distribution of the actual structural damage response.

[0099] By calculating the global information entropy and spatial coordinate dispersion of each analysis frequency band, the data quality can be dynamically evaluated and used as weights to construct a high-fidelity saliency map.

[0100] In practice, the first step is to obtain refined saliency features at multiple scales. For each independent analytical scale The calculation of two-dimensional quantitative indicators is initiated in parallel.

[0101] The first step is to calculate the information purity by performing an equidistant discretization mapping on the numerical intervals of the feature matrix. Specifically, the total number of discrete intervals is preset to 256, i.e., Bin=256. The data is evenly divided into 256 equidistant intervals from the minimum to the maximum value, and the frequency of the value falling into each discrete interval is counted to construct a probability distribution vector. .

[0102] Based on this distribution, an overall information entropy characterizing purity is constructed. The lower the value, the more concentrated the anomalous energy and the less interference at that scale. The specific mathematical model for its purely quantitative characteristics is constructed as follows:

[0103]

[0104] In the formula, The significance value appears in the first position. The empirical probability of a quantization interval.

[0105] To prevent the extremely small positive number hyperparameter from causing logarithmic overflow with zero probability, the empirical value is configured as follows: It can be adjusted by the implementer according to the specific implementation scenario.

[0106] The second step is to calculate the spatial dispersion and extract the set of foreground response spatial coordinates that are greater than the mean from the feature matrix at this scale. Calculate the standard deviation of its spatial distribution.

[0107] Subsequently, the standard deviation of this physical coordinate set is calculated to generate a dispersion index characterizing the spatial clustering state. The specific mathematical calculation characteristics of this dispersion evaluation index are constructed as follows:

[0108]

[0109] In the formula, the dispersion index The smaller the physical value, the more the foundation pit damage response exhibits a highly concentrated point-like or block-like morphology in space.

[0110] For the first The spatial coordinates of each foreground response node.

[0111] This is the arithmetic mean of the coordinates of all foreground nodes.

[0112] constant This represents the total number of foreground nodes.

[0113] Combined with the obtained overall information entropy With dispersion index A dynamic weighting evaluation result based on negative exponential mapping is constructed. The scale that minimizes the product of the overall information entropy and the standard deviation of the spatial distribution is assigned the highest numerical weight to calculate the first... Unnormalized fusion weights of scale . The calculation method is as follows:

[0114]

[0115] In the formula, when the overall information entropy With dispersion index The smaller the value, the purer and more clustered the features, and the closer the absolute value of the exponent term is to 0, thus making the unnormalized fusion weights more effective. It approaches the maximum value of 1.

[0116] The hyperparameter of the adjustment factor that controls the intensity of the ratio's influence is set to 1 based on experience, and can be adjusted by the implementer according to the specific implementation scenario.

[0117] Finally, the unnormalized fusion weights at each scale are normalized to obtain the final weights, and the refined saliency features at all scales are spatially weighted and summed according to these weights to output a one-dimensional comprehensive saliency map. The high-fidelity map was constructed.

[0118] S6: Construct an energy function that takes into account the consistency of significant numerical values ​​and spatial adjacency, and perform global optimization to segment physically coherent abnormal regions to trigger construction safety linkage early warning.

[0119] Since the spectrum is essentially a one-dimensional continuous numerical sequence distributed along the physical coordinates of the optical fiber, directly using an absolute threshold to determine the state will ignore the characteristic that the local physical damage of the deep foundation pit presents a coherent clustering pattern in space.

[0120] Such blind judgments based on fragmented spatial relationships are prone to generating a large number of discontinuous false early warning points, which severely reduces the reliability and interpretability of the monitoring and early warning system, making it urgent to introduce spatial constraints.

[0121] The optimal delineation of abnormal areas in a foundation pit must simultaneously meet two criteria: firstly, the status label must closely fit the original significance observation evidence; secondly, adjacent physical measurement points should be forcibly assigned the same safety status label to ensure consistency.

[0122] A unified energy function is constructed that includes a data cost term for data anomaly detection and a spatial continuous penalty cost term. The segmentation is transformed into a graph theory model that seeks the minimum extremum of the global energy link, which can completely eliminate the interference of isolated scattered points.

[0123] In specific implementation, the above comprehensive saliency map is used. Using the observation benchmark, a system total energy function is constructed for state binarization mapping. .

[0124] First, calculate the cost item for the first core component, namely the anomaly detection data. Calling the nonlinear mapping function to plot the spectrum The Middle physical space nodes The absolute significance value is converted into a prospect probability characterizing the potential damage state. Introducing binary state labels , where 1 represents an abnormal state and 0 represents a normal state.

[0125] The specific calculation and construction of the anomaly detection data cost item is as follows:

[0126]

[0127] In the formula, This represents the fitting penalty cost required to determine a specific measurement point node as the current label.

[0128] This maps the statistical probability of a real deep foundation pit structural anomaly occurring at that physical location.

[0129] A minimal positive hyperparameter is used to prevent the system from crashing due to logarithmic operations with zero probability. This is used when a node has a high probability of malfunctioning, i.e. When the value is close to 1 but is forcibly assigned a normal label, This will result in a huge numerical penalty.

[0130] Subsequently, the second core component for calculating the total energy function is the spatially continuous penalty cost term. Extract physically adjacent nodes With neighboring nodes In the comprehensive saliency map The range of numerical differences.

[0131] Based on the contrast-sensitive characteristic, an exponential decay penalty mechanism is constructed. When adjacent physical space nodes have similar values ​​but are assigned completely mutually exclusive independent state labels by the system, a significant smoothing penalty resistance is applied. The specific decay calculation of this spatial continuity penalty cost term is constructed as follows:

[0132]

[0133] In the formula, This represents the system energy cost required to disrupt physical coherence.

[0134] and These represent the absolute saliency values ​​of physically adjacent nodes.

[0135] The global smoothing hyperparameter, used to control contrast sensitivity, is empirically set to 0.7 times the overall standard deviation of the spectrum, and can be adjusted by the implementer according to the specific implementation scenario.

[0136] This is a discrete function indicating the state. When adjacent entity labels are different, the value is activated to 1, and when they are the same, it is assigned a value of 0, which drives the spatial clustering distribution of the pit damage area.

[0137] Finally, the cost item for anomaly detection data will be... With spatial continuous penalty cost Through the balance coefficient Combine to generate the complete system total energy function. :

[0138]

[0139] In the formula, the balance coefficient The relative weights used to adjust data fit and spatial coherence.

[0140] The total energy function The complete mapping is a network flow capacity allocation representation model under a directed graph structure.

[0141] The core algorithm of minimum cut graph theory is executed to find the optimal split boundary in the graph network topology that cuts links to maximize capacity, and outputs the optimal set of labeled nodes that minimizes energy. Among them, minimum cut graph theory search is a well-known technique and will not be elaborated further.

[0142] Extracting sets The system automatically analyzes the actual engineering spatial coordinates and physical span of the physically connected abnormal areas, and instantly generates intervention commands to trigger a safety linkage early warning for deep foundation pit construction.

[0143] This invention also discloses a distributed optical fiber intelligent monitoring and early warning system for deep foundation pit construction, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to the present invention is implemented.

[0144] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0145] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), etc., or any other medium that can be used to store desired information and can be accessed by an application program, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0146] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A distributed fiber optic intelligent monitoring and early warning method for deep foundation pit construction, characterized in that, include: The optical fiber strain distribution sequence during the deep foundation pit construction process is obtained, and the optical fiber strain distribution sequence is subjected to spatial difference processing to extract the strain gradient sequence. Multi-scale spatial frequency decomposition is performed on the strain gradient sequence to obtain a spatial frequency energy matrix that characterizes the intensity of different spatial frequency abrupt changes. A preliminary environmental background model is constructed by extracting the gradual variation component of the space-frequency energy matrix, and a residual significance map is generated by calculating the logarithmic domain difference between the space-frequency energy matrix and the preliminary environmental background model. The gradient direction consistency of the residual significance map at multiple scales is statistically analyzed to calculate the edge confidence. Based on this, the smoothing scale of the preliminary environmental background model is adaptively adjusted to output refined significance features. The fusion weight is calculated by combining the information purity and spatial dispersion of the refined saliency features. The spatial dispersion is obtained by extracting the spatial coordinate distribution of the foreground response regions with values ​​greater than the mean in the refined saliency features and calculating their spatial distribution standard deviation. The weighted fusion generates a comprehensive saliency map. A system total energy function that takes into account both significant numerical values ​​and spatial adjacency consistency is constructed and global optimization is performed to segment physically coherent abnormal regions to trigger construction safety linkage early warning.

2. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 1, characterized in that, The multi-scale spatial frequency decomposition of the strain gradient sequence to obtain a spatial frequency energy matrix characterizing the intensity of different spatial frequency abrupt changes includes: A continuous wavelet with complex number representation capability is used as the mother function to perform convolution transformation analysis on the strain gradient sequence on a preset spatial translation sequence and frequency band scale sequence; The absolute magnitude of the complex coefficients of each transform node is extracted to generate the space-frequency energy matrix with scale and spatial dimensions as analytical axes.

3. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 1, characterized in that, The step of constructing a preliminary environmental background model by extracting the gradual variation component of the space-frequency energy matrix and calculating the logarithmic domain difference between the space-frequency energy matrix and the preliminary environmental background model to generate a residual significance map includes: An anisotropic filtering module with a smoothing kernel width greater than that in the spatial dimension is used to convolve the spatial frequency energy matrix to obtain a preliminary environmental background model. After adding a minimum positive number to the values ​​of the preliminary environmental background model and the space-frequency energy matrix, a logarithmic transformation operation is performed, and the transformation difference between the two is calculated. The residual significance map is generated by amplifying the high-frequency abrupt signal in the logarithmic domain.

4. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 1, characterized in that, The method of calculating marginal confidence by statistically analyzing the gradient direction consistency of the residual significance map across multiple scales includes: In a two-dimensional analytical domain consisting of spatial and scale dimensions, the two-dimensional spatial gradient direction and the average absolute magnitude of the residual significance map at multiple adjacent scales are calculated. The number of spatial gradient direction alignment scales falling within the preset angle tolerance range at each physical spatial location is counted, and the product of the proportion of the number of alignment scales and the absolute magnitude of the average gradient is determined as the edge confidence.

5. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 3, characterized in that, The adaptive adjustment of the smoothing scale of the initial environmental background model to output refined saliency features includes: The edge confidence is input into the smooth bounded contraction calculation model to obtain the kernel width adjustment coefficient for the corresponding physical spatial location; In the high-confidence region identified as the edge of the structure, the spatial dimension of the preliminary environmental background model is dynamically reduced using the kernel width adjustment coefficient to smooth the kernel width, and the background residual is re-extracted based on the adjusted kernel width to generate the refined saliency features.

6. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 1, characterized in that, The calculation of the fusion weight by combining the information purity and spatial dispersion of the refined saliency features includes: The refined saliency features at each scale are mapped into a probability distribution form, and the overall information entropy of its distribution state is calculated as the information purity. Extract the spatial coordinate distribution of the foreground response regions with values ​​greater than the mean from the refined saliency features, and calculate the standard deviation of its spatial distribution as the spatial dispersion. The scale that minimizes the product of the overall information entropy and the standard deviation of the spatial distribution is assigned the highest numerical weight, and the fusion weights corresponding to each scale are obtained through exponential function normalization.

7. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 1, characterized in that, The construction of the system's total energy function, which takes into account both saliency and spatial adjacency consistency, specifically includes constructing an anomaly detection data cost term and a spatial continuity penalty cost term: The numerical values ​​of the comprehensive saliency map are converted into the prospect probability of a single physical space node being in an abnormal damage state using a mapping function, and the negative log probabilities of forcibly marking the corresponding physical space node as an abnormal state and a normal state are calculated respectively, which are used as the anomaly determination data cost term. Extract the numerical comparison differences between adjacent physical space nodes in the comprehensive saliency map, construct contrast-sensitive computational features, and generate the spatial continuous penalty cost term that grows exponentially when adjacent physical space nodes have similar values ​​but are assigned completely mutually exclusive independent state labels. The total energy function of the system is generated by combining the anomaly determination data cost term and the spatial continuity penalty cost term through a balance coefficient.

8. The distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to claim 7, characterized in that, The global optimization solution segments physically coherent abnormal regions to trigger construction safety linkage early warnings, including: The total system energy function, which combines the anomaly determination data cost term and the spatial continuity penalty cost term, is transformed into a network flow capacity allocation model in a directed graph structure. The network minimum cut search strategy is executed to find the optimal partition boundary for cutting off extreme energy links. The set of abnormal nodes output is determined as the physically coherent abnormal region and the system early warning action is executed.

9. A distributed fiber optic intelligent monitoring and early warning system for deep foundation pit construction, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a distributed optical fiber intelligent monitoring and early warning method for deep foundation pit construction according to any one of claims 1-8.

Images