Method for locating leakage channel of earth-rockfill dam based on rectified flow and inversion of convective-diffusive heat transfer

Abstract

The present application provides a kind of based on Rectified Flow and convection-diffusion heat transfer inversion earth-rock dam leakage channel positioning method;Method obtains along dam temperature profile sequence and water level, air temperature, sunshine state, and establishes mileage-engineering coordinate mapping;Rectified Flow network based on state information is used to generate no leakage reference temperature distribution;With the difference between measured and reference, construct and baseline correction local temperature deviation;Along the mileage, candidate position is generated by extracting abnormal section, and convection-diffusion heat transfer inversion is used to impose local convection enhancement at candidate, generate candidate temperature deviation response, select optimal position with spatial profile and time trend consistency score, and determine whether leakage or not with absolute threshold and difference threshold;Optimal mileage is mapped to dam engineering coordinate;The present application combines data-driven and mechanism model, suppresses working condition and global drift interference, improves positioning accuracy and reliability, and is suitable for earth-rock dam leakage monitoring and operation early warning.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, and in particular to a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion. Background Technology

[0002] Seepage in earth-rock dams is a critical safety hazard, as seepage channels alter local heat and mass transfer within the dam body. Distributed fiber optic thermometry (DTS), embedded along the dam body, can acquire high spatiotemporal resolution temperature profiles to support seepage identification. However, dam body temperature is simultaneously affected by reservoir water level fluctuations, dam area air temperature variations, and dam surface sunlight / shading conditions, resulting in a highly non-stationary and spatially heterogeneous temperature field, posing challenges to temperature-based seepage location. Engineering applications also need to consider the difference between fiber optic mileage and engineering coordinates to ensure that the location results can be directly used for maintenance and remediation.

[0003] In existing technologies, traditional monitoring methods include seepage pressure, seepage flow, infrared thermography, and electrical resistivity tomography, used for macroscopic identification of leakage risks. DTS-based solutions often employ threshold alarms, time trend analysis, spatial gradients, or clustering to identify abnormal segments. Some methods also use statistical regression with air temperature and water level as independent variables to construct background temperature and perform differencing. Another type of method is based on numerical models of the convection-diffusion equations, which perform forward or inverse modeling to fit observations by setting possible leakage locations and parameters. Still other methods combine statistical characteristics with limited physical priors to form semi-empirical anomaly detection and location processes, which have been applied in engineering.

[0004] The above methods have shortcomings under complex working conditions: First, it is difficult to obtain a stable and comparable "leakage-free benchmark" that varies with water level, temperature and sunshine conditions, which may easily lead to misjudging environmental factors as anomalies; second, the physical inversion lacks candidate-based constraints and space-time consistency verification, resulting in non-unique positioning and limited efficiency; third, it lacks standardized mapping and thresholding decisions from fiber optic mileage to engineering coordinates, making it difficult to directly produce executable leakage location or channel-free determination.

[0005] Therefore, a method for locating seepage channels in earth-rock dams that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion. The core technical problem to be solved by this application is: under the operating scenario of significant fluctuations in reservoir water level, air temperature and sunshine, how to effectively isolate the influence of operating conditions from the DTS temperature profile to obtain a "leakage-free benchmark", and complete candidate verification and uniqueness determination based on the convection-diffusion mechanism, so as to give the location of usable seepage channels in engineering coordinates.

[0007] The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion according to embodiments of the present invention includes:

[0008] S1. Obtain the temperature profile sequence formed by distributed optical fiber temperature measurement, obtain the water level, air temperature and sunshine status information formed by reservoir water level change, air temperature and sunshine status and sunshine shadow distribution, and obtain the correspondence between optical fiber mileage coordinates and dam body engineering coordinates.

[0009] S2. Based on the temperature profile sequence and water level, air temperature and sunshine status information, the RectifiedFlow reference temperature generation network is called to generate a leak-free reference temperature distribution. The RectifiedFlow reference temperature generation network includes a state information extraction network and a RectifiedFlow vector field network. The state information extraction network generates a state condition vector from the water level, air temperature and sunshine status information. Under the action of the state condition vector, the RectifiedFlow vector field network performs continuous rectification transformation on the temperature profile sequence according to the number of rectifications to obtain the leak-free reference temperature distribution.

[0010] S3. Calculate the local temperature deviation distribution based on the temperature profile sequence and the non-leaking reference temperature distribution, and determine the local temperature deviation distribution as the inversion observation;

[0011] S4. Generate a list of candidate leakage locations based on the local temperature deviation distribution. For each candidate leakage location, call the convection-diffusion heat transfer inversion module to parameterize the candidate leakage location as the location of local convection enhancement caused by the leakage channel and generate the candidate temperature deviation response. Calculate the matching score between the candidate temperature deviation response and the local temperature deviation distribution to form a matching score table, and determine the optimal leakage location based on the matching score table.

[0012] S5. Based on the matching scoring table, the optimal leakage location is determined by a threshold. If the threshold is met, the leakage channel location is obtained; otherwise, the no-channel determination result is obtained.

[0013] S6. Based on the correspondence between fiber optic mileage coordinates and dam body engineering coordinates, map the location of the seepage channel to the dam body engineering coordinate positioning result, or associate the no-channel determination result with the dam body engineering coordinate range.

[0014] Optionally, S1 is as follows:

[0015] Temperature profiles corresponding to each time sampling point are obtained by distributed optical fiber temperature measurement along the earth-rock dam, and the temperature profiles are stacked in time order to form a temperature profile sequence using the optical fiber mileage coordinate as the spatial sampling axis.

[0016] The reservoir water level changes corresponding to the time sampling points are obtained to form a water level sequence, and the air temperature status corresponding to the time sampling points is obtained to form an air temperature sequence, so that the number of time sampling points of the water level sequence and the air temperature sequence are consistent with the temperature profile sequence.

[0017] Spatial sampling of the solar and shadow states on the dam surface is performed along the fiber optic mileage coordinates to form a solar and shadow distribution sequence, so that the number of mileage sampling points in the solar and shadow distribution sequence is consistent with the number of mileage sampling points in the temperature profile sequence, and the water level sequence, air temperature sequence, and solar and shadow distribution sequence are combined to form water level, air temperature, and solar status information.

[0018] The correspondence between the fiber optic mileage coordinates obtained from fiber optic laying measurements and the dam engineering coordinate control points is constructed, and the correspondence is used as the input for mapping the location of seepage channels.

[0019] Optionally, S2 is as follows:

[0020] The temperature profile sequence and water level, air temperature and sunshine status information are used as inputs to the RectifiedFlow baseline temperature generation network, and the temperature profile sequence is determined as the current temperature distribution.

[0021] The water level sequence and temperature sequence in the water level, temperature and sunshine status information are input into the time feature extraction subnetwork, and one-dimensional convolution is performed on the water level sequence and temperature sequence to extract the time status feature vector;

[0022] The solar shading distribution sequence in the water level, air temperature and solar shading status information is input into the spatial feature extraction subnetwork. One-dimensional convolution is performed on the solar shading distribution sequence to extract the spatial state feature vector. The temporal state feature vector and the spatial state feature vector are concatenated and input into the fusion fully connected layer. The state condition vector is obtained by passing through the first fully connected layer containing sixty-four neurons and the second fully connected layer containing thirty-two neurons.

[0023] The current temperature distribution and state condition vector are input into the large-scale feature branch of the RectifiedFlow vector field network. The encoder-decoder structure is used to aggregate large-scale temperature gradient features along the fiber optic mileage coordinates. The encoder sequentially passes through three layers of one-dimensional convolution with sixteen, thirty-two, and sixty-four channels and performs two downsampling operations. The decoder performs two upsampling operations and sequentially passes through two layers of one-dimensional convolution with thirty-two and sixteen channels to restore the original fiber optic mileage coordinate resolution.

[0024] The current temperature distribution is input into the local feature branch of the RectifiedFlow vector field network, and local temperature fluctuation features are extracted sequentially through two layers of one-dimensional convolution with sixteen channels.

[0025] The state condition vector is input into the condition injection layer and mapped to a channel bias consistent with the number of channels in each convolutional layer through a fully connected layer. The channel bias is then added to the output of each convolutional layer in the large-scale feature branch and the local feature branch to form fused features.

[0026] The fused feature input-output convolutional layer generates a continuous rectification transformation direction consistent with the dimension of the current temperature distribution. The current temperature distribution is updated multiple times according to the number of rectifications. Each update calls the RectifiedFlow vector field network to obtain the continuous rectification transformation direction and completes one rectification. Finally, the updated current temperature distribution is determined as the leak-free baseline temperature distribution.

[0027] Optionally, in the process of generating a continuous rectification transformation direction consistent with the dimension of the current temperature distribution from the input-output convolutional layer of the fused features, updating the current temperature distribution multiple times according to the number of rectifications, and calling the RectifiedFlow vector field network to obtain the continuous rectification transformation direction and complete one rectification cycle each time, a rule function is set to control the rectification update, wherein the rule function is specifically:

[0028] ;

[0029] in, For the first The current temperature distribution of the next rectification iteration. To complete the first Updated temperature distribution after secondary rectification The rectifier iteration number is the value ranging from zero to... , For the number of rectifications, Update step size based on base This is the minimum step size ratio. The Sigmoid function is used to process input real numbers. The calculation method is to calculate first. The exponent value is then divided by one and added to the exponent value to obtain a value between zero and one. This is the state condition vector. These are gated weight vectors obtained through offline training, used in conjunction with the state condition vector. Multiplying them together yields the gating intermediate value. It is a gating bias scalar and is obtained through offline training. For continuous rectification and direction change, the multiplication is used to apply the scalar update magnitude to At each temperature location, complete the process from arrive After the update, Determined as a leak-free reference temperature distribution Leak-free reference temperature distribution With temperature profile sequence The dimensions are consistent and serve as the baseline input for subsequent steps to calculate the local temperature deviation distribution.

[0030] Optionally, S3 specifically refers to:

[0031] The leak-free reference temperature distribution is resampled according to the time sampling points of the temperature profile sequence and the fiber optic mileage coordinates, so that the leak-free reference temperature distribution and the temperature profile sequence correspond one-to-one on the time axis and the spatial axis.

[0032] For each time sampling point and each fiber optic mileage coordinate sampling point, calculate the local temperature deviation value of the temperature value corresponding to the temperature value of the temperature profile sequence relative to the temperature value corresponding to the leak-free reference temperature distribution, and organize all local temperature deviation values ​​into a local temperature deviation distribution according to the time sampling point and fiber optic mileage coordinate.

[0033] For each time sampling point, the full profile deviation from the baseline is calculated based on the local temperature deviation distribution along the fiber optic mileage coordinates, and the local temperature deviation distribution is used to perform baseline correction on the full profile deviation from the baseline to suppress the overall offset;

[0034] The local temperature deviation distribution after baseline correction is determined as the inversion observation of the convection-diffusion heat transport inversion module, and the number of time sampling points and fiber optic mileage coordinate sampling points of the inversion observation are kept consistent with those of the temperature profile sequence.

[0035] Optionally, S4 specifically refers to:

[0036] Using the local temperature deviation distribution as input, the time dimension deviation intensity index is extracted for each mileage sampling point along the fiber optic mileage coordinate, and an abnormal mileage set is formed by filtering according to the preset intensity threshold.

[0037] The abnormal mileage set is merged into abnormal segments according to mileage continuity, and the center mileage of each abnormal segment is determined as a candidate leakage location. The candidate leakage locations are arranged into a candidate leakage location list in mileage order.

[0038] For each candidate leakage location in the candidate leakage location list, the candidate leakage location is parameterized as the location of local convection enhancement caused by the leakage channel, and the local convection enhancement location, together with the preset convection enhancement intensity parameter, the preset influence range parameter, and the preset diffusion parameter, are used as the input parameters of the convection diffusion heat transfer inversion module.

[0039] In the convection-diffusion heat transfer inversion module, the leak-free reference temperature distribution is used as the background temperature field. The convection-diffusion heat transfer process is driven based on the location of local convection enhancement, and candidate temperature deviation responses with the same dimension as the local temperature deviation distribution are generated.

[0040] The spatial profile morphology consistency calculation and temporal evolution trend consistency calculation are performed between the candidate temperature deviation response and the local temperature deviation distribution. The consistency results are then fused based on preset weights to obtain the matching score of the candidate leakage location.

[0041] Repeat the process of generating candidate temperature deviation responses and calculating matching scores for each candidate leak location in the candidate leak location list to form a matching score table containing the correspondence between candidate leak locations and matching scores;

[0042] The candidate leakage location with the highest matching score is selected as the optimal leakage location based on the matching score table.

[0043] Optionally, the spatial profile morphology consistency and temporal evolution trend consistency of the candidate temperature deviation response and the local temperature deviation distribution are calculated, and the consistency results are fused using a matching function based on preset weights to obtain the matching score of the candidate leakage location, wherein the matching function is specifically:

[0044] ;

[0045] in, For the first Matching scores for each candidate leak location Weighting for consistency of spatial profile morphology. As a weight for consistency of time evolution trend, The number of time sampling points, This represents the number of sampling points for fiber optic mileage coordinates. For local temperature deviations at the sampling time Mileage The local temperature deviation at that location, For the first The candidate temperature deviation response at sampling time Mileage The candidate temperature at that location deviates from the response value. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time The local temperature deviation distribution will be lowered. The aggregated value is obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. Sampling time Next, the candidate temperature deviation response The aggregated value obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. For all sampling times Find the average of the time values. For all sampling times Find the average of the time values.

[0046] Optional, S5 specifically includes:

[0047] Extract the candidate leakage location with the highest matching score from the matching score table as the optimal leakage location, and extract the candidate leakage location with the second highest matching score as the suboptimal leakage location.

[0048] Read the optimal matching score corresponding to the optimal leakage location and the second-best matching score corresponding to the second-best leakage location, and calculate the score difference between the optimal matching score and the second-best matching score.

[0049] The optimal matching score is compared with the absolute threshold, and the score difference is compared with the difference threshold. The absolute threshold is used to constrain the consistency strength of the mechanism between the candidate temperature deviation response and the local temperature deviation distribution, and the difference threshold is used to constrain the uniqueness of the location of the optimal leakage location relative to the second-best leakage location.

[0050] The optimal leakage location is determined as the leakage channel location when the optimal matching score meets the absolute threshold and the score difference meets the difference threshold; otherwise, a no-channel determination result is obtained.

[0051] Optionally, step S6 specifically includes:

[0052] Read the correspondence between fiber optic mileage coordinates and dam body engineering coordinates, represent the correspondence as a segmented mapping composed of multiple control points, and perform linear interpolation on the mileage segments between control points to obtain the dam body engineering coordinates corresponding to any fiber optic mileage coordinate.

[0053] When the location of the seepage channel is obtained, the fiber optic mileage coordinates corresponding to the location of the seepage channel are input into the segmented mapping to obtain the dam body engineering coordinate positioning result, and the dam body engineering coordinate positioning result is output as the positioning coordinates of the seepage channel.

[0054] When the no-channel determination result is obtained, the start and end points of the fiber optic mileage coordinates are input into the segmented mapping to obtain the dam body engineering coordinate range, and the dam body engineering coordinate range is associated with the no-channel determination result.

[0055] The beneficial effects of this invention are:

[0056] (1) This proposal presents an improved method for generating a leak-free reference temperature. It employs a combination of Rectified Flow conditional vector fields and state information extraction: temporal state features are extracted by performing temporal convolution on water level and air temperature, and spatial state features are extracted by performing spatial convolution on sunshine / shade. These features are then fused into a state conditional vector through a fully connected layer and injected into a vector field network with large-scale and local branches. Conditionalization is achieved through channel bias. In the continuous rectified transformation, a gating step size and iteration schedule driven by the state conditional vector are introduced to gradually push the current temperature distribution towards a leak-free reference temperature distribution. Compared to methods relying solely on statistical regression or empirical smoothing, this improvement can stably isolate environmental drivers under significant operating condition fluctuations, providing a comparable background field across time periods and a reliable benchmark for subsequent mechanism inversion.

[0057] (2) This proposal puts forward a novel candidate verification and uniqueness determination method, which constructs a parameterized model of local convection enhancement based on the convection-diffusion heat transfer mechanism: Convection enhancement is applied to the candidate mileage action zone and combined with diffusion update and zero flux boundary to generate candidate temperature deviation response of the same dimension as observation; the normalized correlation of the time-by-time spatial profile and the normalized correlation of the time series aggregated in the action zone are fused according to weight by a matching function to obtain a matching score, and absolute threshold and difference threshold derived from the historical quantile of no leakage are introduced on the entire candidate set to determine the consistency of mechanism and the uniqueness of location. Unlike the approach of direct global inversion or single threshold alarm, the method constrains candidates with physical consistency, reduces the interference caused by environmental noise and multiple solutions, and produces verifiable location conclusions.

[0058] (3) This proposal puts forward an integrated leakage location method that connects data acquisition, modeling and output. It combines the time-dimensional deviation intensity index (weighted average absolute deviation and maximum absolute deviation) with mileage continuity to generate candidates, and uses baseline correction based on the median of time to suppress overall drift. It also uses the non-leakage reference temperature as the background field to drive the inversion. At the engineering application level, the location results are accurately converted from fiber optic mileage to dam engineering coordinates through piecewise linear mapping of mileage to engineering coordinates based on control points. Compared with the scheme that lacks standardized mapping and decision-making process, the overall idea forms a closed-loop output from data to coordinates in complex working conditions, which is convenient for maintenance and operation verification. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0061] Figure 2 This is a flowchart illustrating the data acquisition and construction of water level, air temperature, and sunshine status information for a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0062] Figure 3 This is a flowchart of the RectifiedFlow reference temperature generation network for a method for locating seepage channels in earth-rock dams based on RectifiedFlow and convection-diffusion heat transfer inversion proposed in this invention.

[0063] Figure 4 This is a flowchart of the calculation and baseline correction of local temperature deviation distribution in a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0064] Figure 5 This is a flowchart of the candidate seepage location generation and convection-diffusion heat transfer inversion matching process for a seepage channel location method for earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0065] Figure 6 This is a flowchart illustrating the threshold determination and location uniqueness of a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0066] Figure 7 This is a flowchart illustrating the segmented mapping relationship between fiber optic mileage coordinates and dam body engineering coordinates in a method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0067] Figure 8 This is a schematic diagram of a three-dimensional cross-sectional perspective of the dam body and the location points of the seepage channels in an earth-rock dam based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention.

[0068] Figure 9 This is a biplot comparing the presence and absence of leakage on the same profile of a seepage channel location method for earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion proposed in this invention. Detailed Implementation

[0069] In Example 1, reference Figures 1 to 9 A method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion, comprising:

[0070] S1. Obtain the temperature profile sequence formed by distributed optical fiber temperature measurement, obtain the water level, air temperature and sunshine status information formed by reservoir water level change, air temperature and sunshine status and sunshine shadow distribution, and obtain the correspondence between optical fiber mileage coordinates and dam body engineering coordinates.

[0071] S2. Based on the temperature profile sequence and water level, air temperature and sunshine status information, the RectifiedFlow reference temperature generation network is called to generate a leak-free reference temperature distribution. The RectifiedFlow reference temperature generation network includes a state information extraction network and a RectifiedFlow vector field network. The state information extraction network generates a state condition vector from the water level, air temperature and sunshine status information. Under the action of the state condition vector, the RectifiedFlow vector field network performs continuous rectification transformation on the temperature profile sequence according to the number of rectifications to obtain the leak-free reference temperature distribution.

[0072] S3. Calculate the local temperature deviation distribution based on the temperature profile sequence and the non-leaking reference temperature distribution, and determine the local temperature deviation distribution as the inversion observation;

[0073] S4. Generate a list of candidate leakage locations based on the local temperature deviation distribution. For each candidate leakage location, call the convection-diffusion heat transfer inversion module to parameterize the candidate leakage location as the location of local convection enhancement caused by the leakage channel and generate the candidate temperature deviation response. Calculate the matching score between the candidate temperature deviation response and the local temperature deviation distribution to form a matching score table, and determine the optimal leakage location based on the matching score table.

[0074] S5. Based on the matching scoring table, the optimal leakage location is determined by a threshold. If the threshold is met, the leakage channel location is obtained; otherwise, the no-channel determination result is obtained.

[0075] S6. Based on the correspondence between fiber optic mileage coordinates and dam body engineering coordinates, map the location of the seepage channel to the dam body engineering coordinate positioning result, or associate the no-channel determination result with the dam body engineering coordinate range.

[0076] In this embodiment, step S1 specifically includes:

[0077] Distributed optical fiber temperature measurement fibers are laid along the predetermined monitoring route inside the earth-rock dam. A set of spatial sampling points is set along the fiber optic mileage coordinates, and the mileage value corresponding to each spatial sampling point is recorded as follows. ,in For each mileage sampling point, a set of time sampling points is set up, and the sampling time corresponding to each time sampling point is recorded as . ,in The time sampling point number is used as the time sampling point number. The subsequent RectifiedFlow reference temperature generation network receives the temperature profile sequence in the form of a two-dimensional tensor. Therefore, the number of mileage sampling points and the number of time sampling points are fixed during the acquisition phase to ensure that the temperature profile sequence and the water level, air temperature and sunshine status information can be directly aligned in dimension.

[0078] A distributed fiber optic temperature measurement host is used at each sampling time. Trigger a full-line temperature measurement and read the sampling points at each mileage along the fiber optic mileage coordinates. The temperature value at that location, and record the temperature value as For the same sampling time The obtained temperature set for the entire mileage constitutes a temperature profile, and the temperature data at each sampling time is then analyzed. Temperature profiles are stacked in chronological order to form a temperature profile sequence. Temperature profile sequence The first dimension index corresponds to the time sampling point number. The second dimension index corresponds to the mileage sampling point sequence number. By using a fixed set of mileage sampling points, the temperature profile sequence is guaranteed. The fiber optic mileage coordinates have a consistent spatial resolution, thus providing a stable input for the subsequent RectifiedFlow vector field network that fits a large range of temperature gradients on a spatial scale.

[0079] Reading and sampling time The corresponding reservoir water level observation value is denoted as... And according to the time sampling point number Arranged to form a water level sequence Reading and sampling time The corresponding temperature observation values ​​in the dam area are denoted as follows: And according to the time sampling point number Arranged to form a temperature sequence When the reservoir water level changes and the original sampling time of the air temperature are... When there is inconsistency, the original observations are mapped to the nearest sampling time. , making the water level sequence With temperature series Number of time sampling points and temperature profile sequence The number of time sampling points is consistent, thus enabling the time feature extraction subnetwork of the state information extraction network to process the water level sequence. With temperature series Perform a one-dimensional convolution and output a temporal state feature vector;

[0080] Spatial sampling of the solar radiation and shadow conditions on the dam surface was performed along the fiber optic mileage coordinates to form a solar radiation and shadow distribution sequence. Specifically, firstly, a geometric model of the dam surface under the engineering coordinates of the dam body is established. The geometric model of the dam surface is discretized into a three-dimensional mesh or a three-dimensional polyline patch, and sampling points are set for each mileage. The corresponding spatial locations on the dam surface were determined, and then each sampling time was... The direction of solar incidence is determined by the geographical location of the dam area and the sampling time. Determine, and then sample at each mileage point. At sampling time Perform ray occlusion determination: Taking the dam surface point corresponding to the sampling point at this mileage as the ray origin, and the opposite direction of the sun's incident direction as the ray direction, search for the intersection points of the ray with the geometric model of the dam surface and the geometric model of the surrounding occlusion bodies within the preset maximum determination distance. If an intersection point exists and the intersection point is not the ray origin, then the sampling time is recorded. The sampling point at the next mileage is determined to be either shaded or sunny. After determining all sampling times, each sampling point at that mileage is then... The number of time sampling points that were identified as sunshine durations was calculated to the number of time sampling points that participated in the identification process, and this ratio was used as the sunshine shadow distribution sequence. At mileage sampling points The value at the location ,in A value of zero indicates that the entire scene is in shadow. To indicate that the entire process is under sunlight, the water level sequence is... Temperature sequence Sunlight shadow distribution sequence Composition of water level, air temperature, and sunshine status information and make the solar shadow distribution sequence Mileage sampling points and temperature profile sequence The number of mileage sampling points is consistent, so that the spatial feature extraction subnetwork of the state information extraction network can extract the solar shadow distribution sequence. Perform a one-dimensional convolution and output a spatial state feature vector;

[0081] Multiple control point pairs were obtained based on fiber optic cable laying measurements. Each control point pair contains fiber optic mileage coordinates and dam body engineering coordinates for the same physical location. The fiber optic mileage coordinates are denoted as... The coordinate values ​​of the dam body are denoted as ,in For control point serial numbers, according to After sorting from smallest to largest, establish the correspondence between fiber optic mileage coordinates and dam body engineering coordinates. When it is necessary to map the location of the seepage channel from fiber optic mileage coordinates to dam body engineering coordinates, input the mileage value of the seepage channel location. In the correspondence Mid-positioning meets Less than or equal to and Less than or equal to The interval, the dam body engineering coordinate values ​​at the endpoints of the interval. and Perform linear interpolation according to the mileage ratio to obtain the mileage value. The corresponding dam body engineering coordinate positioning results provide a directly callable input for subsequent steps to output the location of the seepage channel as dam body engineering coordinates.

[0082] In this embodiment, step S2 specifically includes:

[0083] The RectifiedFlow baseline temperature generation network is trained offline and its parameters are fixed before deployment. During online runtime, it only performs forward inference, denoting the temperature profile sequence as... ,in Based on the time sampling point number With fiber optic mileage coordinate sampling point number index, Indicates the sampling time Mileage Temperature value at that location, For the first The sampling time corresponding to each time sampling point For the first The mileage values ​​corresponding to each fiber optic mileage coordinate sampling point are recorded as follows: ,in Indicates the sampling time The reservoir water level observations are recorded as follows: The temperature sequence is denoted as... ,in Indicates the sampling time The observed temperature values ​​are used to denote the solar radiation shadow distribution sequence as follows: ,in Indicates mileage The value of the shadow at the location will be taken. , , Composition of water level, air temperature, and sunshine status information ;

[0084] Temperature profile sequence Information on water level, air temperature, and sunshine conditions Input the RectifiedFlow baseline temperature generation network and output the temperature profile sequence. Directly assign the current temperature distribution ,in and The dimensions are consistent, with the number of time sampling points as the first dimension and the number of fiber optic mileage coordinate sampling points as the second dimension, serving as the initial state for subsequent continuous rectification transformation;

[0085] Water level sequence With temperature series After aligning the time sampling point numbers, a dual-channel time input is formed. This dual-channel time input is then fed into a time feature extraction sub-network. The time feature extraction sub-network performs two layers of one-dimensional convolution along the time axis, with a kernel length of three and a stride of one, so that the output at each time step simultaneously integrates the current time step and the adjacent time steps. and Global average pooling is performed along the time axis on the output of the second-layer one-dimensional convolution to obtain the time-state feature vector. Time-state feature vector Each dimension is a pooled numerical feature component, used to characterize the common change pattern of reservoir water level and air temperature.

[0086] Sunlight shadow distribution sequence As a single-channel spatial input, the spatial feature extraction subnetwork performs two layers of one-dimensional convolution along the fiber optic mileage coordinates. The convolution kernel has a length of three and a stride of one, so that the output of each mileage position simultaneously fuses the output of that mileage position with that of its adjacent mileage positions. The spatial state feature vector is obtained by performing global average pooling along the mileage axis on the output of the second-layer one-dimensional convolution. Spatial state feature vector Each dimension is a pooled numerical feature component, used to characterize the spatial differences in the influence of sunlight and shadow on the dam surface. Subsequently, the temporal state feature vector is... With spatial state feature vector A fusion vector is formed by concatenating the features. This fusion vector is then input into a fusion fully connected layer, which passes sequentially through the first and second fully connected layers. The first fully connected layer has 64 neurons, and the second fully connected layer has 32 neurons, resulting in a state condition vector. State condition vector Used to inject water level, air temperature, and sunshine status information into the RectifiedFlow vector field network;

[0087] Current temperature distribution With state condition vector Input the large-scale feature branch of the RectifiedFlow vector field network, where This is the rectification iteration number, representing the current temperature distribution. The process is frame-by-frame based on time sampling points. The mileage temperature sequence corresponding to each time sampling point is used as the input of a one-dimensional convolution. The convolution operation aggregates a large-scale temperature gradient along the fiber optic mileage coordinates. The encoder sequentially uses three layers of one-dimensional convolution with sixteen, thirty-two, and sixty-four channels to extract features and reduces the mileage resolution by downsampling twice. The downsampling is achieved by one-dimensional convolution with a stride of two. The decoder performs two upsampling operations on the encoder output features to restore the mileage resolution. The upsampling is achieved by linear interpolation and then sequentially uses two layers of one-dimensional convolution with thirty-two and sixteen channels to obtain large-scale feature output, thus forming a large-scale temperature gradient feature consistent with the original fiber optic mileage coordinate resolution.

[0088] Current temperature distribution The local feature branch of the input RectifiedFlow vector field network is also subjected to one-dimensional convolution along the fiber optic mileage coordinates frame by frame according to the time sampling points. The local feature branch is successively passed through two layers of one-dimensional convolution with sixteen channels to extract local temperature fluctuation features, so that the local feature branch remains sensitive to local disturbance morphology, while maintaining the same division of labor with the large-scale feature branch in spatial scale.

[0089] The state condition vector The input conditional injection layer generates a set of channel biases for each convolutional layer through a fully connected mapping. The channel biases are denoted as... ,in The convolutional layer number. The length is consistent with the number of output channels of the convolutional layer. For each output feature of the convolutional layer in the large-scale feature branch and the local feature branch, the corresponding channel bias is broadcast and summed along the channel dimension. This allows the convolutional features to vary with water level, air temperature, and solar radiation status information. Simultaneously, the conditional injection layer generates gating parameters for rectification and updating, which are used to transform the state condition vector... The coefficients are converted into scalar update coefficients, so that the continuous rectification transformation has a consistent update rhythm under different operating conditions. The large-scale features and local features after the channel bias injection are fused to obtain the fused features.

[0090] The fused features are input to the output convolutional layer, and the output convolutional layer maps the fused features to a single-channel continuously rectified transform direction. This continuously rectified transform direction is denoted as... ,in With the current temperature distribution With consistent dimensions, the number of rectifications is set to [value]. Set the basic update step size to Set the minimum step size ratio to For each rectification iteration, a state condition vector is used. Generate the updated gating coefficients and jointly determine the update magnitude with the iteration progress, and complete a rectified update according to the following rule function:

[0091] ;

[0092] in, For the first The current temperature distribution of the next rectification iteration. To complete the first Updated temperature distribution after secondary rectification The rectifier iteration number is the value ranging from zero to... , For the number of rectifications, Update step size based on base This is the minimum step size ratio. The Sigmoid function is used to process input real numbers. The calculation method is to calculate first. The exponent value is then divided by one and added to the exponent value to obtain a value between zero and one. This is the state condition vector. These are gated weight vectors obtained through offline training, used in conjunction with the state condition vector. Multiplying them together yields the gating intermediate value. It is a gating bias scalar and is obtained through offline training. For continuous rectification and direction change, the multiplication is used to apply the scalar update magnitude to At each temperature location, complete the process from arrive After the update, Determined as a leak-free reference temperature distribution Leak-free reference temperature distribution With temperature profile sequence The dimensions are consistent and serve as the baseline input for subsequent steps to calculate the local temperature deviation distribution.

[0093] In this embodiment, step S3 specifically includes:

[0094] Step S3 uses a temperature profile sequence and leak-free reference temperature distribution As input, where Indicates the sampling time Mileage Temperature value at that location, The leak-free reference temperature distribution output in step S2. For the first The sampling time corresponding to each time sampling point For the first The mileage value corresponding to each fiber optic mileage coordinate sampling point. The time sampling point number, The sampling point number is the fiber optic mileage coordinate. The output of step S3 is the local temperature deviation distribution, and the local temperature deviation distribution is determined as the inversion observation of the convection-diffusion heat transfer inversion module.

[0095] Leak-free reference temperature distribution According to temperature profile sequence The time sampling points and fiber optic mileage coordinate sampling points are resampled. Specifically, if there is no leakage, the reference temperature distribution is resampled. Time sampling point set and If inconsistent, then for each mileage sampling point Along the time axis respectively Perform one-dimensional linear interpolation so that the interpolation result is consistent across all sampling times. All obtained corresponding reference temperature values, if there is no leakage, the reference temperature distribution The set of mileage sampling points and If inconsistent, then for each sampling time Pair along fiber optic mileage coordinates Perform one-dimensional linear interpolation so that the interpolation result is consistent across all mileages. All obtained corresponding reference temperature values, and the leak-free reference temperature distribution after resampling is still recorded as . And make it consistent with the temperature profile sequence. There is a one-to-one correspondence between the time axis and the space axis;

[0096] For each sampling time With each mile Calculate the local temperature deviation and record it as . Local temperature deviation Temperature values ​​corresponding to temperature profile sequences Subtract the temperature value corresponding to the leak-free reference temperature distribution The results show that all local temperature deviations are sorted by time sampling point number. with mileage sampling point number The tissue exhibits localized temperature deviations. ,in With temperature profile sequence Consistent dimensions;

[0097] For each sampling time Based on local temperature deviation distribution Calculate the full profile deviation from the baseline along the fiber optic mileage coordinates, and denote the full profile deviation from the baseline as . The entire profile deviates from the baseline. The calculation method is as follows: take the sampling time. Corresponding to all local temperature deviations Sort the values ​​in ascending order. When the number of mileage sampling points is odd, take the value in the middle position after sorting. When the number of mileage sampling points is even, the average of the two middle values ​​after sorting is taken as the average. Subsequently, the sampling time was... Every mile Perform baseline correction to adjust local temperature deviations. Updated to This causes the local temperature to deviate from the distribution. At each sampling time, the deviation of the entire profile from the baseline is suppressed;

[0098] The local temperature deviation distribution after baseline correction will be completed. The inversion observations were identified as belonging to the convection-diffusion heat transport inversion module, and the inversion observations were kept consistent with the temperature profile sequence in terms of the number of time sampling points and the number of fiber optic mileage coordinate sampling points. This consistency enables the convection-diffusion heat transfer inversion module to take inversion observations as input and generate candidate temperature deviation responses and calculate matching scores under the constraint of the candidate leak location list.

[0099] In this embodiment, step S4 specifically includes:

[0100] In this embodiment, step S4 involves the aforementioned local temperature deviation distribution. As the sole inversion observation, and combined with the leak-free reference temperature distribution Generate candidate temperature deviation responses to locate leakage channels and local temperature deviation distributions. Depend on constitute, Indicates the sampling time Mileage The local temperature deviation at that location, For the first The sampling time corresponding to each time sampling point For the first The mileage value corresponding to each fiber optic mileage coordinate sampling point. The time sampling point number, The mileage coordinate sampling point number is the fiber optic mileage coordinate sampling point number, and the mileage difference between adjacent mileage sampling points is defined as follows: , From adjacent The difference is obtained, and the candidate leakage locations are ranked according to the candidate number. Index, will the first The mileage value of each candidate leakage location is denoted as: The preset intensity threshold is denoted as The preset convection enhancement intensity parameter is denoted as The preset influence range parameter is denoted as The preset diffusion parameters are denoted as The consistency weight of spatial profile morphology is denoted as The consistency weight of the time evolution trend is denoted as ;

[0101] Local temperature deviation distribution As input, sample each mileage point along the fiber optic mileage coordinates. Extract the time-dimensional deviation intensity index for a fixed mileage sampling point sequence. Extracting time series The time series is taken as an absolute value, and the time average is calculated. Simultaneously, the maximum value of the absolute value series is calculated, and a preset average weighting coefficient is set. , The time dimension deviation intensity index is calculated as a real number between zero and one. Multiply by the stated time average and Multiply by the sum of the maximum values, and then compare the time dimension deviation intensity index with the preset intensity threshold. The comparison will satisfy the time dimension deviation intensity index as not less than The mileage sampling point number is added to the abnormal mileage set;

[0102] Abnormal mileage sets are merged into abnormal segments based on mileage continuity, and then sorted by mileage sampling point number. If the mileage interval between adjacent abnormal mileage sampling points is no greater than [missing value], then [missing value]. If the interval is greater than 100 km / h, it is determined to be the same abnormal segment and will continue to expand. The current abnormal segment ends and the next abnormal segment begins. For each abnormal segment, the starting mileage and ending mileage are read, and the average of the starting mileage and ending mileage is determined as the segment center mileage. This segment center mileage is then used as the mileage value for candidate leakage locations. All candidate leakage locations are arranged in ascending order of mileage to obtain a list of candidate leakage locations;

[0103] For each candidate leak location in the candidate leak location list, the candidate leak location is parameterized as a location of local convection enhancement caused by the leak channel, and the parameter set of the local convection enhancement location is defined as follows: and will Set to the mileage value of candidate leakage locations Preset convection enhancement intensity parameters Preset influence range parameters Preset diffusion parameters Composed in sequence, As input parameters to the convection-diffusion heat transfer inversion module, the location degree of freedom is limited to the mileage value of the candidate leakage location. ;

[0104] In the convection-diffusion heat transfer inversion module, a leak-free reference temperature distribution is used. As a background temperature field, based on the convective-diffusion heat transfer process driven by the location of local convection enhancement, candidate temperature deviation responses are generated for the first... A set of mileage sampling points was determined based on candidate leakage locations to identify a local convection enhancement zone. This set consisted of points satisfying the mileage requirement. Falling into the range arrive The mileage sampling points constitute the initial value of the candidate temperature deviation response at all sampling times and all mileage sampling points, and are recursively updated according to the time sampling point sequence: at each sampling time... Apply a preset convection enhancement intensity parameter to the effective region. The determined local convection enhancement disturbance is used to perform diffusion propagation updates on candidate temperature deviation responses across the entire mileage. The diffusion propagation update calculates the diffusion increment based on the difference in candidate temperature deviation responses between adjacent mileage sampling points. The diffusion increment is compared with preset diffusion parameters. Proportional to and with mileage sampling interval Relatedly, zero-flux boundary processing is used at the mileage endpoints, allowing endpoint diffusion updates to be completed using the difference between sampling points of adjacent mileages. After recursion at all sampling times, the distribution deviating from the local temperature is obtained. Candidate temperature deviation responses with consistent dimensions will be the first The candidate temperature deviation responses are denoted as ,in Indicates the sampling time Mileage The candidate temperature at that location deviates from the response value;

[0105] Deviate candidate temperature from response Deviation from local temperature distribution Spatial profile consistency calculation and temporal evolution trend consistency calculation are performed, and a matching score for the candidate leakage location is obtained by fusion based on preset weights. The spatial profile consistency calculation adopts normalized correlation calculation at each sampling time, and the temporal evolution trend consistency calculation adopts normalized correlation calculation after aggregating the mileage of the local convection enhancement zone to obtain two time series. The matching score is calculated according to the matching function.

[0106] ;

[0107] in, For the first Matching scores for each candidate leak location Weighting for consistency of spatial profile morphology. As a weight for consistency of time evolution trend, The number of time sampling points, This represents the number of sampling points for fiber optic mileage coordinates. For local temperature deviations at the sampling time Mileage The local temperature deviation at that location, For the first The candidate temperature deviation response at sampling time Mileage The candidate temperature at that location deviates from the response value. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time The local temperature deviation distribution will be lowered. The aggregated value is obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. Sampling time Next, the candidate temperature deviation response The aggregated value obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. For all sampling times Find the average of the time values. For all sampling times Find the average of the time values ​​obtained by averaging.

[0108] For each candidate leak location in the candidate leak location list, the process of generating candidate temperature deviation responses and calculating matching scores is repeated, with each candidate number... Build parameter set And generate candidate temperature deviation responses Then, the matching score is obtained according to the matching score calculation formula above. Mileage values ​​of candidate leakage locations Matching score Organized into a matching scoring table according to a one-to-one correspondence;

[0109] Based on the matching score table, the candidate leakage location with the highest matching score is selected as the optimal leakage location. The candidate number corresponding to the highest matching score is then retrieved from the matching score table. and the candidate number Corresponding candidate leakage location mileage value The optimal leakage location was determined, providing input for subsequent threshold determination of the optimal leakage location based on the matching scoring table.

[0110] In this embodiment, step S5 specifically includes:

[0111] Step S5 takes the matching score table output in step S4 as input. The matching score table consists of multiple sets of candidate leakage locations and their corresponding matching scores. The candidate numbers are denoted as follows: , will the The mileage value of each candidate leakage location is denoted as: , will the The matching score corresponding to each candidate leakage location is denoted as . The number of candidate leakage locations matched in the scoring table is recorded as follows: The pre-stored absolute threshold is denoted as The difference threshold is denoted as The absolute threshold is used to constrain the consistency strength of the mechanism between the candidate temperature deviation response and the local temperature deviation distribution, while the difference threshold is used to constrain the uniqueness of the location of the optimal leakage location relative to the suboptimal leakage location.

[0112] The optimal leak location is selected from the candidate leak location with the highest matching score in the matching score table, and the second-highest matching score is selected as the suboptimal leak location. Specifically, the process involves iterating through the table. arrive Record the current maximum matching score and its corresponding candidate number, and denote the candidate number corresponding to the maximum matching score as... The optimal leakage location is denoted as After extracting the optimal leakage location, the candidate sequence number is... Remove candidates from the set being compared, iterate through the remaining candidate numbers again and record the maximum matching score, then denote the candidate number corresponding to the second-highest matching score as... The suboptimal leakage location is denoted as When multiple candidate leakage locations have the same matching score, the candidate leakage location with the smaller mileage value is determined as the higher-ranked candidate leakage location to ensure that the extraction results of the optimal leakage location and the second-best leakage location are unique.

[0113] Read the optimal matching score corresponding to the optimal leakage location and record it as . ,in Taken from Read the suboptimal matching score corresponding to the suboptimal leakage location and record it as . ,in Taken from Calculate the difference between the best matching score and the second-best matching score and record it as . ,in The value is taken as the optimal matching score Subtract the second-best match score ,when At that time, the suboptimal match score will be determined. The score is fixed at zero, and the score difference is calculated accordingly. ;

[0114] Optimal matching score With absolute threshold Compare and assign score differences With difference threshold Comparison, absolute threshold With difference threshold The threshold configuration is obtained and written through system calibration: a pre-defined period of no-leakage historical operation is selected. During this period, multiple sets of matching score tables are generated in a manner consistent with online operation. The optimal matching score of each set of matching score tables is formed into an optimal matching score sequence, and the score difference of each set of matching score tables is formed into a score difference sequence. The optimal matching score sequence and the score difference sequence are sorted in ascending order of value, and the value located at the pre-defined quantile position is taken as the absolute threshold. With difference threshold The preset quantile position is fixed as a threshold configuration parameter, so that the threshold can be obtained directly from historical no-leakage data.

[0115] In the best matching score Satisfying not less than the absolute threshold And the score difference Satisfy not less than the difference threshold At that time, the optimal leakage location will be determined. If the location is confirmed as a leakage path, otherwise a "no path" result is output. The leakage path location or "no path" result is then compared with the corresponding... , , , , Record them together so that they can be directly called upon later for mapping the location of leakage channels and for operational verification.

[0116] In this embodiment, step S6 specifically includes:

[0117] In this embodiment, step S6 is used to convert the location of the seepage channel output in step S5 from fiber optic mileage coordinates to dam engineering coordinates, and to mark the fiber optic mileage coordinates as follows: , The coordinates of the dam body are defined as the cumulative mileage along the fiber optic cable laying direction. , It consists of three coordinate components, denoted as ,in For the lateral coordinate components of the dam body engineering coordinates, For the longitudinal coordinate components of the dam body engineering coordinates, For the elevation coordinate components of the dam body engineering coordinates;

[0118] The correspondence between fiber optic mileage coordinates and dam body engineering coordinates is read, and the correspondence is represented as a segmented mapping consisting of multiple control points. The control points are then sorted according to their control point numbers. Index, will the first Each control point is denoted as ,in The fiber optic mileage coordinates of the control point. These are the engineering coordinate values ​​of the dam body corresponding to the control points. As the control point number, all control points are assigned according to... Sort the control points from smallest to largest, and require adjacent control points to satisfy... To form a range of adjacent control points The constructed segmented mapping, for any input fiber optic mileage coordinate First locate the satisfaction For the mileage section, calculate the interpolation coefficients. interpolation coefficients Depend on and After determining the ratio, linear interpolation is performed on the three coordinate components respectively to obtain... The three components are respectively and weighted sum, and weighted sum, and The weighted sum, where the weights are determined by... and Composition, if fiber optic mileage coordinates are input Less than the minimum control point mileage Then Set as If fiber optic mileage coordinates are input Greater than the maximum control point mileage Then Set as ,in This represents the total number of control points.

[0119] When the location of the leakage channel is obtained in step S5, the fiber optic mileage coordinates corresponding to the location of the leakage channel are used as the input mileage value, and the location of the leakage channel is recorded as... ,in Take the location of the leakage channel output from step S5, and... Input the segmented mapping to obtain the dam body engineering coordinate positioning results. ,in The coordinate positioning results of the dam body project The output is the location coordinates of the leakage channel, and its relationship to the location of the leakage channel. Output them together to preserve the correspondence between the mileage domain and the engineering coordinate domain;

[0120] When the no-channel determination result is obtained in step S5, the start and end points of the fiber optic mileage coordinates are read, and the start and end points are respectively input into the segmented mapping to obtain the dam body engineering coordinate range. The start point of the fiber optic mileage coordinates is recorded as... The endpoint of the fiber optic mileage coordinates is denoted as ,in and To represent the minimum and maximum mileage values ​​for fiber optic mileage coordinates, Input segmentation mapping ,Will Input segmentation mapping ,Will The coordinate range of the dam body is determined, and the coordinate range of the dam body is associated with the no-passage determination result and output to characterize the spatial range of the dam body covered by the no-passage determination result.

[0121] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion, characterized in that, include: S1. Obtain the temperature profile sequence formed by distributed optical fiber temperature measurement, obtain the water level, air temperature and sunshine status information formed by reservoir water level change, air temperature and sunshine status and sunshine shadow distribution, and obtain the correspondence between optical fiber mileage coordinates and dam body engineering coordinates. S2. Based on the temperature profile sequence and water level, air temperature and sunshine status information, the RectifiedFlow reference temperature generation network is called to generate a leak-free reference temperature distribution. The RectifiedFlow reference temperature generation network includes a state information extraction network and a RectifiedFlow vector field network. The state information extraction network generates a state condition vector from the water level, air temperature and sunshine status information. Under the action of the state condition vector, the RectifiedFlow vector field network performs continuous rectification transformation on the temperature profile sequence according to the number of rectifications to obtain the leak-free reference temperature distribution. S3. Calculate the local temperature deviation distribution based on the temperature profile sequence and the non-leaking reference temperature distribution, and determine the local temperature deviation distribution as the inversion observation; S4. Generate a list of candidate leakage locations based on the local temperature deviation distribution. For each candidate leakage location, call the convection-diffusion heat transfer inversion module to parameterize the candidate leakage location as the location of local convection enhancement caused by the leakage channel and generate the candidate temperature deviation response. Calculate the matching score between the candidate temperature deviation response and the local temperature deviation distribution to form a matching score table, and determine the optimal leakage location based on the matching score table. S5. Based on the matching scoring table, the optimal leakage location is determined by a threshold. If the threshold is met, the leakage channel location is obtained; otherwise, the no-channel determination result is obtained. S6. Based on the correspondence between fiber optic mileage coordinates and dam body engineering coordinates, map the location of the seepage channel to the dam body engineering coordinate positioning result, or associate the no-channel determination result with the dam body engineering coordinate range.

2. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, S1 specifically refers to: Temperature profiles corresponding to each time sampling point are obtained by distributed optical fiber temperature measurement along the earth-rock dam, and the temperature profiles are stacked in time order to form a temperature profile sequence using the optical fiber mileage coordinate as the spatial sampling axis. The reservoir water level changes corresponding to the time sampling points are obtained to form a water level sequence, and the air temperature status corresponding to the time sampling points is obtained to form an air temperature sequence, so that the number of time sampling points of the water level sequence and the air temperature sequence are consistent with the temperature profile sequence. Spatial sampling of the solar and shadow states on the dam surface is performed along the fiber optic mileage coordinates to form a solar and shadow distribution sequence, so that the number of mileage sampling points in the solar and shadow distribution sequence is consistent with the number of mileage sampling points in the temperature profile sequence, and the water level sequence, air temperature sequence, and solar and shadow distribution sequence are combined to form water level, air temperature, and solar status information. The correspondence between the fiber optic mileage coordinates obtained from fiber optic laying measurements and the dam engineering coordinate control points is constructed, and the correspondence is used as the input for mapping the location of seepage channels.

3. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, S2 specifically refers to: The temperature profile sequence and water level, air temperature and sunshine status information are used as inputs to the RectifiedFlow baseline temperature generation network, and the temperature profile sequence is determined as the current temperature distribution. The water level sequence and temperature sequence in the water level, temperature and sunshine status information are input into the time feature extraction subnetwork, and one-dimensional convolution is performed on the water level sequence and temperature sequence to extract the time status feature vector; The solar shading distribution sequence in the water level, air temperature and solar shading status information is input into the spatial feature extraction subnetwork. One-dimensional convolution is performed on the solar shading distribution sequence to extract the spatial state feature vector. The temporal state feature vector and the spatial state feature vector are concatenated and input into the fusion fully connected layer. The state condition vector is obtained by passing through the first fully connected layer containing sixty-four neurons and the second fully connected layer containing thirty-two neurons. The current temperature distribution and state condition vector are input into the large-scale feature branch of the RectifiedFlow vector field network. The encoder-decoder structure is used to aggregate large-scale temperature gradient features along the fiber optic mileage coordinates. The encoder sequentially passes through three layers of one-dimensional convolution with sixteen, thirty-two, and sixty-four channels and performs two downsampling operations. The decoder performs two upsampling operations and sequentially passes through two layers of one-dimensional convolution with thirty-two and sixteen channels to restore the original fiber optic mileage coordinate resolution. The current temperature distribution is input into the local feature branch of the RectifiedFlow vector field network, and local temperature fluctuation features are extracted sequentially through two layers of one-dimensional convolution with sixteen channels. The state condition vector is input into the condition injection layer and mapped to a channel bias consistent with the number of channels in each convolutional layer through a fully connected layer. The channel bias is then added to the output of each convolutional layer in the large-scale feature branch and the local feature branch to form fused features. The fused feature input-output convolutional layer generates a continuous rectification transformation direction consistent with the dimension of the current temperature distribution. The current temperature distribution is updated multiple times according to the number of rectifications. Each update calls the RectifiedFlow vector field network to obtain the continuous rectification transformation direction and completes one rectification. Finally, the updated current temperature distribution is determined as the leak-free baseline temperature distribution.

4. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 3, characterized in that, In the process of generating a continuous rectification transformation direction consistent with the dimension of the current temperature distribution using the fused feature input-output convolutional layer, updating the current temperature distribution multiple times according to the number of rectifications, and calling the RectifiedFlow vector field network to obtain the continuous rectification transformation direction and complete one rectification cycle each time, a rule function is set to control the rectification update. Specifically, the rule function is as follows: ； in, For the first The current temperature distribution of the next rectification iteration. To complete the first Updated temperature distribution after secondary rectification The rectifier iteration number is the value ranging from zero to... , For the number of rectifications, Update step size based on base This is the minimum step size ratio. The Sigmoid function is used to process input real numbers. The calculation method is to calculate first. The exponent value is then divided by one and added to the exponent value to obtain a value between zero and one. This is the state condition vector. These are gated weight vectors obtained through offline training, used in conjunction with the state condition vector. Multiplying them together yields the gating intermediate value. It is a gating bias scalar and is obtained through offline training. For continuous rectification and direction change, the multiplication is used to apply the scalar update magnitude to At each temperature location, complete the process from arrive After the update, Determined as a leak-free reference temperature distribution Leak-free reference temperature distribution With temperature profile sequence The dimensions are consistent and serve as the baseline input for subsequent steps to calculate the local temperature deviation distribution.

5. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, S3 specifically refers to: The leak-free reference temperature distribution is resampled according to the time sampling points of the temperature profile sequence and the fiber optic mileage coordinates, so that the leak-free reference temperature distribution and the temperature profile sequence correspond one-to-one on the time axis and the spatial axis. For each time sampling point and each fiber optic mileage coordinate sampling point, calculate the local temperature deviation value of the temperature value corresponding to the temperature value of the temperature profile sequence relative to the temperature value corresponding to the leak-free reference temperature distribution, and organize all local temperature deviation values ​​into a local temperature deviation distribution according to the time sampling point and fiber optic mileage coordinate. For each time sampling point, the full profile deviation from the baseline is calculated based on the local temperature deviation distribution along the fiber optic mileage coordinates, and the local temperature deviation distribution is used to perform baseline correction on the full profile deviation from the baseline to suppress the overall offset; The local temperature deviation distribution after baseline correction is determined as the inversion observation of the convection-diffusion heat transport inversion module, and the number of time sampling points and fiber optic mileage coordinate sampling points of the inversion observation are kept consistent with those of the temperature profile sequence.

6. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, S4 specifically refers to: Using the local temperature deviation distribution as input, the time dimension deviation intensity index is extracted for each mileage sampling point along the fiber optic mileage coordinate, and an abnormal mileage set is formed by filtering according to the preset intensity threshold. The abnormal mileage set is merged into abnormal segments according to mileage continuity, and the center mileage of each abnormal segment is determined as a candidate leakage location. The candidate leakage locations are arranged into a candidate leakage location list in mileage order. For each candidate leakage location in the candidate leakage location list, the candidate leakage location is parameterized as the location of local convection enhancement caused by the leakage channel, and the local convection enhancement location, together with the preset convection enhancement intensity parameter, the preset influence range parameter, and the preset diffusion parameter, are used as the input parameters of the convection diffusion heat transfer inversion module. In the convection-diffusion heat transfer inversion module, the leak-free reference temperature distribution is used as the background temperature field. The convection-diffusion heat transfer process is driven based on the location of local convection enhancement, and candidate temperature deviation responses with the same dimension as the local temperature deviation distribution are generated. The spatial profile morphology consistency calculation and temporal evolution trend consistency calculation are performed between the candidate temperature deviation response and the local temperature deviation distribution. The consistency results are then fused based on preset weights to obtain the matching score of the candidate leakage location. Repeat the process of generating candidate temperature deviation responses and calculating matching scores for each candidate leak location in the candidate leak location list to form a matching score table containing the correspondence between candidate leak locations and matching scores; The candidate leakage location with the highest matching score is selected as the optimal leakage location based on the matching score table.

7. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 6, characterized in that, The spatial profile morphology consistency and temporal evolution trend consistency of the candidate temperature deviation response and the local temperature deviation distribution are calculated. Based on preset weights, a matching function is used to fuse the consistency results to obtain a matching score for the candidate leakage location. The matching function is as follows: ； in, For the first Matching scores for each candidate leak location Weighting for consistency of spatial profile morphology. As a weight for consistency of time evolution trend, The number of time sampling points, This represents the number of sampling points for fiber optic mileage coordinates. For local temperature deviations at the sampling time Mileage The local temperature deviation at that location, For the first The candidate temperature deviation response at sampling time Mileage The candidate temperature at that location deviates from the response value. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time Next, sample all mileage points. Calculate the mean value of the obtained profile. Sampling time The local temperature deviation distribution will be lowered. The aggregated value is obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. Sampling time Next, the candidate temperature deviation response The aggregated value obtained by summing along the mileage within the set of mileage sampling points covered by the local convection enhancement zone. For all sampling times Find the average of the time values. For all sampling times Find the average of the time values.

8. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, S5 specifically refers to: Extract the candidate leakage location with the highest matching score from the matching score table as the optimal leakage location, and extract the candidate leakage location with the second highest matching score as the suboptimal leakage location. Read the optimal matching score corresponding to the optimal leakage location and the second-best matching score corresponding to the second-best leakage location, and calculate the score difference between the optimal matching score and the second-best matching score. The optimal matching score is compared with the absolute threshold, and the score difference is compared with the difference threshold. The absolute threshold is used to constrain the consistency strength of the mechanism between the candidate temperature deviation response and the local temperature deviation distribution, and the difference threshold is used to constrain the uniqueness of the location of the optimal leakage location relative to the second-best leakage location. The optimal leakage location is determined as the leakage channel location when the optimal matching score meets the absolute threshold and the score difference meets the difference threshold; otherwise, a no-channel determination result is obtained.

9. The method for locating seepage channels in earth-rock dams based on Rectified Flow and convection-diffusion heat transfer inversion as described in claim 1, characterized in that, Step S6 is as follows: Read the correspondence between fiber optic mileage coordinates and dam body engineering coordinates, represent the correspondence as a segmented mapping composed of multiple control points, and perform linear interpolation on the mileage segments between control points to obtain the dam body engineering coordinates corresponding to any fiber optic mileage coordinate. When the location of the seepage channel is obtained, the fiber optic mileage coordinates corresponding to the location of the seepage channel are input into the segmented mapping to obtain the dam body engineering coordinate positioning result, and the dam body engineering coordinate positioning result is output as the positioning coordinates of the seepage channel. When the no-channel determination result is obtained, the start and end points of the fiber optic mileage coordinates are input into the segmented mapping to obtain the dam body engineering coordinate range, and the dam body engineering coordinate range is associated with the no-channel determination result and output.

Images