Fusion positioning model training and fusion positioning method for power line carrier line

Abstract

The application discloses a fusion positioning model training and a fusion positioning method for a power line carrier line. The training method comprises the following steps: acquiring multi-source space information of a sample target through a plurality of source sensors arranged around a to-be-measured power line carrier line; obtaining multi-source space features after time synchronization, space registration and feature extraction of the multi-source space information; constructing a mask matrix by using the distance between feature points of the multi-source space features and a power transmission line catenary of the to-be-measured power line carrier line; wherein the mask matrix is used to constrain a preset effective area of the to-be-measured power line carrier line; performing mask cross attention processing on the multi-source space features by using the mask matrix to obtain fusion space features; and performing model training by using the fusion space features to obtain a fusion positioning model. The application can improve the positioning result accuracy.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a fusion positioning model training method for power line carrier lines and a fusion positioning method thereof. Background Technology

[0002] With the construction of new power systems and the development of power line carrier (power line sensing integration) technology, transmission lines not only undertake power transmission tasks but have also become ubiquitous communication and sensing network nodes. However, power line carrier infrastructure deployed around transmission lines is exposed to a complex outdoor environment for extended periods, making its physical and electromagnetic space highly vulnerable to intrusion and damage from non-cooperative targets. Examples include unauthorized drones causing power grid discharges or data theft, and large construction machinery illegally operating and touching cables. Therefore, locating non-cooperative targets intruding into the area where power line carrier lines are located has become a key research focus.

[0003] Existing solutions for locating non-cooperative targets in areas intruding power line carrier lines mostly employ attention mechanisms to fuse multi-source data and perform global computation in three-dimensional free space. This lacks constraints on the physical topology of the lines, not only consuming a large amount of computing power for calculations in remote invalid areas, but also introducing interference from remote invalid areas, leading to inaccurate location results. Summary of the Invention

[0004] This application provides a fusion positioning model training and fusion positioning method for power line carrier lines, in order to solve the problem that the existing fusion methods based on attention mechanisms all perform global calculations in three-dimensional free space, which lacks constraints on the physical topology of the transmission line. This not only consumes a lot of computing power for feature matching of the far-end invalid area, but also introduces interference from the far-end invalid area, resulting in inaccurate positioning results.

[0005] To achieve the above objectives, embodiments of this application provide a method for training a fusion positioning model for power line carrier lines, comprising: Multi-source spatial information of the sample target is obtained by deploying multi-source sensors around the power line carrier line under test. After performing time synchronization, spatial registration, and feature extraction on the multi-source spatial information, multi-source spatial features are obtained; A mask matrix is ​​constructed using the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test; wherein, the mask matrix is ​​used to constrain the preset effective area of ​​the power line carrier line under test. The multi-source spatial features are processed by mask cross-attention using the mask matrix to obtain fused spatial features; The fused spatial features are used to train the model, resulting in a fused localization model.

[0006] As an improvement to the above scheme, when the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line to be tested is less than or equal to a preset safety distance, the element of that feature point in the mask matrix is ​​recorded as 0. When the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is greater than the preset safety distance, the element of that feature point in the mask matrix is ​​recorded as negative infinity. The step of using the mask matrix to perform masked cross-attention processing on the multi-source spatial features to obtain fused spatial features includes: Based on the multi-source spatial characteristics, set up the query matrix, key matrix, and value matrix; Using the query matrix and the key matrix, the attention score matrix is ​​obtained; The attention score matrix and the mask matrix are added together and then normalized to obtain the attention fusion weight matrix; The fusion space features are obtained by multiplying the attention fusion weight matrix with the value matrix.

[0007] As an improvement to the above scheme, before acquiring the multi-source spatial information of the sample target through multi-source sensors arranged around the power line carrier line under test, the method further includes: Using the catenary of the power line carrier line under test as a reference, a joint calibration model is established among the multi-source sensors; wherein, the joint calibration model is used for spatial registration; A unified time reference is established among the multi-source sensors; wherein the time reference is used for time synchronization.

[0008] As an improvement to the above scheme, the multi-source sensor includes: at least one radar and at least three antenna arrays; The at least one radar is arranged below the power line carrier line under test; The at least three antenna arrays are respectively arranged to the left, right and below the power line carrier line under test.

[0009] To achieve the above objectives, embodiments of this application also provide a fusion positioning method for power line carrier lines, including: Real-time multi-source spatial information of the target to be located is obtained by deploying multi-source sensors around the power line carrier line to be tested. The real-time multi-source spatial information is synchronized in time and registered in space, and then input into the fusion positioning model as described above for inference to obtain the positioning result of the target to be located.

[0010] As an improvement to the above scheme, the step of synchronizing and spatially registering the real-time multi-source spatial information and then inputting it into the fusion positioning model described above for inference to obtain the positioning result of the target to be located includes: The real-time multi-source spatial information is synchronized in time and spatially registered before being input into the fusion localization model for inference. The intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in combination with the target information to obtain the localization result. The target information includes at least one of the following: The position of the multi-source sensor; Channel interference characteristics of the multi-source sensor; The confidence level of the fusion positioning model.

[0011] As an improvement to the above scheme, the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in conjunction with target information, including: Based on the positions of the multi-source sensors, the propagation distances from the initial positioning results generated during the inference process to each antenna array in the multi-source sensors are obtained. The gain coefficient of each antenna array in the multi-source sensor is calculated based on the propagation distance of each antenna array. The gain coefficients of each antenna array are normalized to obtain the normalized gain coefficients of each antenna array. The normalized gain coefficients of each antenna array are used as weights for their respective antenna arrays, and the intermediate feature matrices of each antenna array generated during the inference process are weighted accordingly.

[0012] As an improvement to the above scheme, the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in conjunction with target information, including: When the channel interference characteristics of at least one of the multi-source sensors meet the preset interference conditions, the intermediate feature matrix of that sensor generated during the inference process is weighted. The channel interference characteristics of the multi-source sensor include at least one of the following: Signal-to-noise ratio of the antenna array; The amplitude of spurious signals in the Doppler spectrum of the radar; The interference conditions include at least one of the following: The signal-to-noise ratio is less than a preset signal-to-noise ratio threshold; The magnitude of the increase or change is greater than a preset change threshold.

[0013] As an improvement to the above scheme, the confidence of the fusion localization model is represented by the information entropy of the attention score matrix constrained by the mask matrix; Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: When the information entropy is greater than a preset information entropy threshold, the attention fusion weight matrix generated during the reasoning process is adjusted.

[0014] As an improvement to the above scheme, after adjusting the attention fusion weight matrix generated during the inference process when the information entropy is greater than a preset information entropy threshold, the method further includes: Based on the historical positioning results, a historical positioning trajectory is formed; The historical positioning trajectory is smoothed and predicted using Kalman filtering to obtain the predicted positioning trajectory. The current positioning result is corrected based on the predicted positioning trajectory.

[0015] Compared with existing technologies, this application provides a fusion positioning model training and fusion positioning method for power line carrier lines. It acquires multi-source spatial information of the sample target by deploying multi-source sensors around the power line carrier line under test. After time synchronization, spatial registration, and feature extraction of the multi-source spatial information, multi-source spatial features are obtained. A mask matrix is ​​constructed using the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test. The mask matrix is ​​used to constrain a preset effective area of ​​the power line carrier line under test. The multi-source spatial features are then subjected to mask cross-attention processing using the mask matrix to obtain fused spatial features. The fused spatial features are then used to train the model, resulting in a fusion positioning model. Therefore, this application utilizes a mask matrix to constrain the preset effective area of ​​the power line carrier line under test during the attention processing, focusing computing power on high-risk intrusion sources in the near field of the line, avoiding redundant calculations and interference in the far-end invalid areas, significantly reducing computing resource consumption while improving the accuracy of the positioning results. Attached Figure Description

[0016] Figure 1 This is a flowchart of a fusion positioning model training method for power line carrier lines provided in an embodiment of this application; Figure 2 This is a flowchart of a fusion positioning method for power line carrier lines provided in an embodiment of this application; Figure 3 This is a structural block diagram of a power line carrier line fusion positioning model training device provided in an embodiment of this application; Figure 4 This is a structural block diagram of a fusion positioning device for power line carrier lines provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] In the description of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0019] In this application description, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0020] In this application description, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The term "based on" means "at least partially based on." The term "according to" means "at least partially according to." The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments." The term "and / or" means at least one of the connected objects, such as A and / or B, indicating three cases: including only A, only B, and both A and B. Unless otherwise stated, the term "multiple" means two or more.

[0021] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0022] See Figure 1 , Figure 1 This is a flowchart illustrating a fusion positioning model training method for power line carrier lines provided in an embodiment of this application. The fusion positioning model training method for power line carrier lines includes: S11. Obtain multi-source spatial information of the sample target by using multi-source sensors arranged around the power line carrier line under test; It is worth noting that the power line carrier line under test is a power line carrier line in an area where it is necessary to monitor whether there is any intrusion by non-cooperative targets. By deploying multi-source sensors around the power line carrier line under test, it is possible to monitor and locate whether non-cooperative targets (such as drone swarms or construction machinery) are approaching the power line carrier line under test.

[0023] When an object intrudes into the area or approaches the power line carrier line to be tested, the object is the sample target or the target to be located. The multi-source sensor collects the multi-source spatial information of the sample target or the real-time multi-source spatial information of the target to be located.

[0024] Furthermore, visual cameras are susceptible to the effects of lighting and severe weather, resulting in inaccurate information. Therefore, multi-source sensors can employ antenna arrays and radar to achieve all-weather, all-time, all-electromagnetic domain perception capabilities, effectively improving the ability to locate target intrusions.

[0025] In one alternative embodiment, the multi-source sensor includes: at least one radar and at least three antenna arrays; The at least one radar is arranged below the power line carrier line under test; The at least three antenna arrays are respectively arranged to the left, right and below the power line carrier line under test.

[0026] For example, a 77GHz FMCW millimeter-wave radar and three 4×4 uniform antenna arrays are configured. The radar is positioned below the power line carrier line under test to collect radar reflected echo signals of the target to be located. The three antenna arrays are respectively deployed to the left, right and below the power line carrier line under test to collect radio frequency radiation signals of the target to be located from three different observation angles.

[0027] S12. After performing time synchronization, spatial registration and feature extraction on the multi-source spatial information, multi-source spatial features are obtained; It is worth noting that due to the different types of multi-source sensors, there are sampling clock deviations, different observation perspectives, and different coordinate systems. This application's embodiment achieves time synchronization and coordinate system unification of the multi-source spatial information through time synchronization and spatial registration, and then performs feature extraction to provide data support for subsequent model training.

[0028] In one specific embodiment, radar point cloud geometric features are extracted from the radar reflected echo signal after time synchronization and spatial registration, spatial spectrum features are extracted from the radio frequency radiation signals of each antenna array after time synchronization and spatial registration, and these spatial spectrum features are spliced ​​together in the channel dimension to form the total spatial spectrum features.

[0029] S13. Construct a mask matrix by utilizing the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test; wherein, the mask matrix is ​​used to constrain the preset effective area of ​​the power line carrier line under test. It is worth noting that, in order to avoid calculations in the invalid region at the far end, this embodiment of the application uses the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test to construct a mask matrix, which is used to constrain the preset valid region of the power line carrier line under test.

[0030] S14. Use the mask matrix to perform masked cross-attention processing on the multi-source spatial features to obtain fused spatial features; The embodiments of this application utilize a mask matrix to perform masked cross-attention processing on multi-source spatial features, which can effectively avoid the calculation of invalid regions at the far end.

[0031] S15. Use the fused spatial features to train the model and obtain the fused localization model.

[0032] In the attention processing, this application uses a mask matrix to constrain the preset effective area of ​​the power line carrier line under test, focusing computing power on high-risk intrusion sources in the near field of the line, avoiding redundant calculations and interference in the invalid area of ​​the line, and improving the accuracy of the positioning results while significantly reducing computing power consumption.

[0033] In one specific embodiment, the fused spatial features are associated with the actual localization results labels of the sample targets to construct a training sample library. The deep localization neural network is trained using this sample library until convergence. The loss function during the training process is described below. The definition is as follows: in, Indicates the network's response to the first... The location prediction results for each sample. Indicates the first Ground Truth labels for each sample's location results. This indicates the total number of training samples (or the training batch size) involved in the calculation. This is the weight decay coefficient. These are network parameters.

[0034] In an optional embodiment, when the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is less than or equal to a preset safety distance, the element of that feature point in the mask matrix is ​​recorded as 0. When the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is greater than the preset safety distance, the element of that feature point in the mask matrix is ​​recorded as negative infinity. The step of using the mask matrix to perform masked cross-attention processing on the multi-source spatial features to obtain fused spatial features includes: Based on the multi-source spatial characteristics, set up the query matrix, key matrix, and value matrix; Using the query matrix and the key matrix, the attention score matrix is ​​obtained; The attention score matrix and the mask matrix are added together and then normalized to obtain the attention fusion weight matrix; The fusion space features are obtained by multiplying the attention fusion weight matrix with the value matrix.

[0035] For example, mask matrix The element expression is: ; In the formula, Represents the mask matrix of the first Line number Column elements, The first representing the multi-source spatial features The distance between each feature point and the catenary of the power line carrier line under test. To establish a preset safe distance; when When the distance is less than or equal to the preset safety distance, it indicates that the feature point belongs to the preset effective area of ​​the power line carrier line under test. Therefore, the element of the mask matrix for this feature point is set to 0, allowing the feature point to participate in subsequent calculations. If the distance exceeds the preset safety distance, it indicates that the feature point does not belong to the preset effective area of ​​the power line carrier line under test. Therefore, the element of the mask matrix for this feature point is set to... This prevents the feature point from participating in subsequent calculations.

[0036] Perform mask cross-attention processing according to the following formula: In the formula, For querying the matrix, The key matrix, For value matrices, For transpose, For the mask matrix, The feature dimension of the key vectors within the key matrix. For attention score matrix, For attention fusion weight matrix, This indicates the characteristics of the integrated space.

[0037] In one specific embodiment, the geometric features of the radar point cloud are transformed by a first linear transformation matrix. Mapped to query matrix This is used to provide distance-depth constraints; the total spatial spectral features are respectively transformed by the second linear transformation matrix. The second linear transformation matrix Mapped to a key matrix Sum matrix This is used to provide multi-view angle constraints. At this time, row index This can be understood as the query matrix's first... Feature points, column index This can be understood as the first... 1 feature point.

[0038] In an optional embodiment, before acquiring multi-source spatial information of the sample target using multi-source sensors arranged around the power line carrier line under test, the method further includes: Using the catenary of the power line carrier line under test as a reference, a joint calibration model is established among the multi-source sensors; wherein, the joint calibration model is used for spatial registration; A unified time reference is established among the multi-source sensors; wherein the time reference is used for time synchronization.

[0039] It is worth noting that this application uses the catenary of the power line carrier transmission line under test as a reference to provide a unified spatial reference for multi-source sensors, thereby establishing a joint calibration model among the multi-source sensors. This joint calibration model can unify the spatial information of multiple sources into the same coordinate system. For example, this joint calibration model is a transformation relationship between the radar coordinate system and the antenna array coordinate system. Using this transformation relationship, the radio frequency radiation signal collected by the antenna array is transformed into the radar coordinate system.

[0040] Establishing a unified time reference among multiple sensors enables the simultaneous acquisition of multi-source spatial information. For example, an FPGA is configured as the hardware triggering core to establish a unified time reference between the antenna array and each radar acquisition and transmission channel. The synchronous trigger pulse width is set to 10 μs to 50 μs to ensure that the acquisition time error of each sensor is less than 1 ms. This time reference is then used to trigger each acquisition and transmission channel in parallel, capturing the radio frequency radiation signal and radar reflected echo signal of the sample target.

[0041] See Figure 2 , Figure 2 This is a flowchart of a fusion positioning method for power line carrier lines provided in an embodiment of this application. The fusion positioning method for power line carrier lines includes: S21. Obtain real-time multi-source spatial information of the target to be located by using multi-source sensors arranged around the power line carrier line to be tested. It is worth noting that the acquisition of real-time multi-source spatial information of the target to be located in this embodiment of the application can refer to the acquisition of multi-source spatial information of the sample target mentioned above, and will not be repeated here.

[0042] S22. The real-time multi-source spatial information is synchronized in time and registered in space, and then input into the fusion positioning model as described above for inference to obtain the positioning result of the target to be located.

[0043] It is worth noting that the time synchronization and spatial registration of real-time multi-source spatial information in this application embodiment can refer to the above-described time synchronization and spatial registration of multi-source spatial information, and will not be repeated here.

[0044] The embodiments of this application achieve the positioning of the target by synchronizing and spatially registering the real-time multi-source spatial information of the target to be located in time and then inputting it into the fusion positioning model.

[0045] In an optional embodiment, the step of synchronizing and spatially registering the real-time multi-source spatial information and then inputting it into the fusion positioning model described above for inference to obtain the positioning result of the target to be located includes: The real-time multi-source spatial information is synchronized in time and spatially registered before being input into the fusion localization model for inference. The intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in combination with the target information to obtain the localization result. The target information includes at least one of the following: The position of the multi-source sensor; Channel interference characteristics of the multi-source sensor; The confidence level of the fusion positioning model.

[0046] It is worth noting that, in order to further improve the robustness of perception, the embodiments of this application dynamically adjust the weights through an adaptive closed-loop mechanism in the inference stage of the model, thereby improving the localization results.

[0047] In one specific embodiment, during the inference process, the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted by using the location of the multi-source sensors, the channel interference characteristics of the multi-source sensors, and the confidence of the fusion positioning model, thereby realizing a triple adaptive closed-loop mechanism adjustment.

[0048] In one optional embodiment, adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process in conjunction with target information includes: Based on the positions of the multi-source sensors, the propagation distances from the initial positioning results generated during the inference process to each antenna array in the multi-source sensors are obtained. The gain coefficient of each antenna array in the multi-source sensor is calculated based on the propagation distance of each antenna array. The gain coefficients of each antenna array are normalized to obtain the normalized gain coefficients of each antenna array. The normalized gain coefficients of each antenna array are used as weights for their respective antenna arrays, and the intermediate feature matrices of each antenna array generated during the inference process are weighted accordingly.

[0049] Specifically, the first is calculated according to the following formula. Gain coefficient of each antenna array : In the formula, This represents the initial localization results generated during the reasoning process up to the [number]th [number]. The propagation distance of an antenna array.

[0050] Using the normalized gain coefficient By weighting the intermediate feature matrices of each antenna array (specifically the key matrix or value matrix) during mask cross-attention calculation, the high signal-to-noise ratio near-field signals can be fully utilized, thereby improving positioning accuracy.

[0051] It is worth noting that the propagation loss of electromagnetic waves in free space is proportional to the square of the distance. The closer the nodes are, the higher the signal-to-noise ratio and the closer the MUSIC direction-finding accuracy is to the Cramer-Rhodes lower bound. Therefore, the attention weight of closer nodes is proportionally amplified, thus making full use of the high signal-to-noise ratio near-field signals.

[0052] In one optional embodiment, adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process in conjunction with target information includes: When the channel interference characteristics of at least one of the multi-source sensors meet the preset interference conditions, the intermediate feature matrix of that sensor generated during the inference process is weighted. The channel interference characteristics of the multi-source sensor include at least one of the following: Signal-to-noise ratio of the antenna array; The amplitude of spurious signals in the Doppler spectrum of the radar; The interference conditions include at least one of the following: The signal-to-noise ratio is less than a preset signal-to-noise ratio threshold; The magnitude of the increase or change is greater than a preset change threshold.

[0053] It is worth noting that at the antenna end, the signal-to-noise ratio (SNR) of the antenna array is monitored in real time. When the SNR is lower than the preset SNR threshold (typically 10dB) caused by the switching of a large load on the power grid or the strong electromagnetic pulse caused by arc discharge, the weight of the key vector or value vector corresponding to the antenna array is multiplied by the first attenuation coefficient. At the radar end, the amplitude of spurious signals in the Doppler spectrum is monitored in real time. When the amplitude increases or changes more than a preset threshold (e.g., more than 3 times the normal level) due to strong winds, rain, snow, or severe multipath reflections from power transmission cables, the weight of the corresponding query vector of the radar is reduced by multiplying it by a second attenuation coefficient (e.g., <0.3).

[0054] In an optional embodiment, the confidence of the fusion localization model is characterized by the information entropy of the attention score matrix constrained by the mask matrix; Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: When the information entropy is greater than a preset information entropy threshold, the attention fusion weight matrix generated during the reasoning process is adjusted.

[0055] It is worth noting that when the model effectively focuses on the near-field target, the attention distribution is highly concentrated and the information entropy is low; when the model encounters complex electromagnetic interference and cannot effectively focus, the attention distribution tends to be uniform and the information entropy surges. In this embodiment, the attention fusion weight matrix generated during the inference process is adjusted by the information entropy being greater than a preset information entropy threshold, thereby improving the positioning accuracy.

[0056] Specifically, the information entropy is calculated according to the following formula: In the formula, Attention score matrix after mask constraint The first in Line 1 The elements of the column.

[0057] Among them, the cross-attention score matrix after mask constraint for The attention fusion weight matrix is ​​as follows: Adjustment is achieved by multiplying it by a third attenuation coefficient (specifically 0.1~0.3).

[0058] In an optional embodiment, after adjusting the attention fusion weight matrix generated during the inference process when the information entropy is greater than a preset information entropy threshold, the method further includes: Based on the historical positioning results, a historical positioning trajectory is formed; The historical positioning trajectory is smoothed and predicted using Kalman filtering to obtain the predicted positioning trajectory. The current positioning result is corrected based on the predicted positioning trajectory.

[0059] It is worth noting that after adjusting the attention fusion weight matrix generated during the inference process, Kalman filtering can be enabled to smooth and predict historical positioning trajectories. The predicted positioning trajectory can then be used to correct the current positioning result, further improving positioning accuracy.

[0060] See Figure 3 , Figure 3 This is a structural block diagram of a fusion positioning model training device 10 for power line carrier lines provided in an embodiment of this application. The fusion positioning model training device 10 for power line carrier lines includes: The first acquisition module 11 is used to acquire multi-source spatial information of the sample target through multi-source sensors arranged around the power line carrier line under test; Extraction module 12 is used to perform time synchronization, spatial registration and feature extraction on the multi-source spatial information to obtain multi-source spatial features; The construction module 13 is used to construct a mask matrix by utilizing the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test; wherein, the mask matrix is ​​used to constrain the preset effective area of ​​the power line carrier line under test. The fusion module 14 is used to perform masked cross-attention processing on the multi-source spatial features using the mask matrix to obtain fused spatial features; Training module 15 is used to train the model using the fused spatial features to obtain a fused localization model.

[0061] Optionally, when the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is less than or equal to a preset safety distance, the element of that feature point in the mask matrix is ​​recorded as 0. When the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is greater than the preset safety distance, the element of that feature point in the mask matrix is ​​recorded as negative infinity. The step of using the mask matrix to perform masked cross-attention processing on the multi-source spatial features to obtain fused spatial features includes: Based on the multi-source spatial characteristics, set up the query matrix, key matrix, and value matrix; Using the query matrix and the key matrix, the attention score matrix is ​​obtained; The attention score matrix and the mask matrix are added together and then normalized to obtain the attention fusion weight matrix; The fusion space features are obtained by multiplying the attention fusion weight matrix with the value matrix.

[0062] Optionally, the device further includes: The calibration module is used to establish a joint calibration model among the multi-source sensors, using the catenary of the power line carrier line under test as a reference; wherein, the joint calibration model is used for spatial registration. A synchronization module is used to establish a unified time reference among the multi-source sensors; wherein the time reference is used for time synchronization.

[0063] Optionally, the multi-source sensor includes: at least one radar and at least three antenna arrays; The at least one radar is arranged below the power line carrier line under test; The at least three antenna arrays are respectively arranged to the left, right and below the power line carrier line under test.

[0064] It is worth noting that the working process of each module in the power line carrier line fusion positioning model training device 10 described in this application embodiment can refer to the working process of the power line carrier line fusion positioning model training method described in the above embodiment and achieve the same beneficial effect, and will not be repeated here.

[0065] See Figure 3 , Figure 3 This is a structural block diagram of a fusion positioning device 20 for power line carrier lines provided in an embodiment of this application. The fusion positioning device 20 for power line carrier lines includes: The second acquisition module 21 is used to acquire real-time multi-source spatial information of the target to be located by using multi-source sensors arranged around the power line carrier line to be tested. The positioning module 22 is used to input the real-time multi-source spatial information after time synchronization and spatial registration into the fusion positioning model as described above for inference, so as to obtain the positioning result of the target to be located.

[0066] Optionally, the step of synchronizing and spatially registering the real-time multi-source spatial information and then inputting it into the fusion positioning model described above for inference to obtain the positioning result of the target to be located includes: The real-time multi-source spatial information is synchronized in time and spatially registered before being input into the fusion localization model for inference. The intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in combination with the target information to obtain the localization result. The target information includes at least one of the following: The position of the multi-source sensor; Channel interference characteristics of the multi-source sensor; The confidence level of the fusion positioning model.

[0067] Optionally, the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process can be adjusted in conjunction with target information, including: Based on the positions of the multi-source sensors, the propagation distances from the initial positioning results generated during the inference process to each antenna array in the multi-source sensors are obtained. The gain coefficient of each antenna array in the multi-source sensor is calculated based on the propagation distance of each antenna array. The gain coefficients of each antenna array are normalized to obtain the normalized gain coefficients of each antenna array. The normalized gain coefficients of each antenna array are used as weights for their respective antenna arrays, and the intermediate feature matrices of each antenna array generated during the inference process are weighted accordingly.

[0068] Optionally, the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process can be adjusted in conjunction with target information, including: When the channel interference characteristics of at least one of the multi-source sensors meet the preset interference conditions, the intermediate feature matrix of that sensor generated during the inference process is weighted. The channel interference characteristics of the multi-source sensor include at least one of the following: Signal-to-noise ratio of the antenna array; The amplitude of spurious signals in the Doppler spectrum of the radar; The interference conditions include at least one of the following: The signal-to-noise ratio is less than a preset signal-to-noise ratio threshold; The magnitude of the increase or change is greater than a preset change threshold.

[0069] Optionally, the confidence level of the fusion localization model is characterized by the information entropy of the attention score matrix constrained by the mask matrix; Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: When the information entropy is greater than a preset information entropy threshold, the attention fusion weight matrix generated during the reasoning process is adjusted.

[0070] Optionally, after adjusting the attention fusion weight matrix generated during the inference process when the information entropy is greater than a preset information entropy threshold, the method further includes: Based on the historical positioning results, a historical positioning trajectory is formed; The historical positioning trajectory is smoothed and predicted using Kalman filtering to obtain the predicted positioning trajectory. The current positioning result is corrected based on the predicted positioning trajectory.

[0071] It is worth noting that the working process of each module in the fusion positioning device 10 for power line carrier lines described in this application embodiment can refer to the working process of the fusion positioning method for power line carrier lines described in the above embodiment and achieve the same beneficial effects, and will not be repeated here.

[0072] Furthermore, this application also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the power line carrier line fusion positioning model training method as described in any of the above embodiments, or to execute the power line carrier line fusion positioning method as described in any of the above embodiments.

[0073] Furthermore, this application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the power line carrier line fusion positioning model training method as described in any of the above embodiments, or executes the power line carrier line fusion positioning method as described in any of the above embodiments.

[0074] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A training method for a fusion positioning model for power line carrier lines, characterized in that, include: Multi-source spatial information of the sample target is obtained by deploying multi-source sensors around the power line carrier line under test. After performing time synchronization, spatial registration, and feature extraction on the multi-source spatial information, multi-source spatial features are obtained; A mask matrix is ​​constructed using the distance between the feature points of the multi-source spatial features and the catenary of the power line carrier line under test; wherein, the mask matrix is ​​used to constrain the preset effective area of ​​the power line carrier line under test. The multi-source spatial features are processed by mask cross-attention using the mask matrix to obtain fused spatial features; The fused spatial features are used to train the model, resulting in a fused localization model.

2. The method for training a fusion positioning model for power line carrier lines as described in claim 1, characterized in that, When the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is less than or equal to a preset safety distance, the element of that feature point in the mask matrix is ​​recorded as 0. When the distance between any feature point of the multi-source spatial features and the catenary of the power line carrier line under test is greater than the preset safety distance, the element of that feature point in the mask matrix is ​​recorded as negative infinity. The step of using the mask matrix to perform masked cross-attention processing on the multi-source spatial features to obtain fused spatial features includes: Based on the multi-source spatial characteristics, set up the query matrix, key matrix, and value matrix; Using the query matrix and the key matrix, the attention score matrix is ​​obtained; The attention score matrix and the mask matrix are added together and then normalized to obtain the attention fusion weight matrix; The fusion space features are obtained by multiplying the attention fusion weight matrix with the value matrix.

3. The method for training a fusion positioning model for power line carrier lines as described in claim 1, characterized in that, Before acquiring multi-source spatial information of the sample target using multi-source sensors arranged around the power line carrier line under test, the method further includes: Using the catenary of the power line carrier line under test as a reference, a joint calibration model is established among the multi-source sensors; wherein, the joint calibration model is used for spatial registration; A unified time reference is established among the multi-source sensors; wherein the time reference is used for time synchronization.

4. The method for training a fusion positioning model for power line carrier lines as described in claim 1, characterized in that, The multi-source sensor includes: at least one radar and at least three antenna arrays; The at least one radar is arranged below the power line carrier line under test; The at least three antenna arrays are respectively arranged to the left, right and below the power line carrier line under test.

5. A fusion positioning method for power line carrier lines, characterized in that, include: Real-time multi-source spatial information of the target to be located is obtained by deploying multi-source sensors around the power line carrier line to be tested. The real-time multi-source spatial information is synchronized in time and registered in space, and then input into the fusion positioning model as described in any one of claims 1 to 4 for inference to obtain the positioning result of the target to be located.

6. The fusion positioning method for power line carrier lines as described in claim 5, characterized in that, The step of synchronizing and spatially registering the real-time multi-source spatial information and then inputting it into the fusion positioning model as described in any one of claims 1 to 4 to obtain the positioning result of the target to be located includes: The real-time multi-source spatial information is synchronized in time and spatially registered before being input into the fusion localization model for inference. The intermediate feature matrix and / or attention fusion weight matrix generated during the inference process are adjusted in combination with the target information to obtain the localization result. The target information includes at least one of the following: The position of the multi-source sensor; Channel interference characteristics of the multi-source sensor; The confidence level of the fusion positioning model.

7. The fusion positioning method for power line carrier lines as described in claim 6, characterized in that, Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: Based on the positions of the multi-source sensors, the propagation distances from the initial positioning results generated during the inference process to each antenna array in the multi-source sensors are obtained. The gain coefficient of each antenna array in the multi-source sensor is calculated based on the propagation distance of each antenna array. The gain coefficients of each antenna array are normalized to obtain the normalized gain coefficients of each antenna array. The normalized gain coefficients of each antenna array are used as weights for their respective antenna arrays, and the intermediate feature matrices of each antenna array generated during the inference process are weighted accordingly.

8. The fusion positioning method for power line carrier lines as described in claim 6, characterized in that, Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: When the channel interference characteristics of at least one of the multi-source sensors meet the preset interference conditions, the intermediate feature matrix of that sensor generated during the inference process is weighted. The channel interference characteristics of the multi-source sensor include at least one of the following: Signal-to-noise ratio of the antenna array; The amplitude of spurious signals in the Doppler spectrum of the radar; The interference conditions include at least one of the following: The signal-to-noise ratio is less than a preset signal-to-noise ratio threshold; The magnitude of the increase or change is greater than a preset change threshold.

9. The fusion positioning method for power line carrier lines as described in claim 6, characterized in that, The confidence level of the fusion localization model is represented by the information entropy of the attention score matrix constrained by the mask matrix; Adjusting the intermediate feature matrix and / or attention fusion weight matrix generated during the inference process based on target information, including: When the information entropy is greater than a preset information entropy threshold, the attention fusion weight matrix generated during the reasoning process is adjusted.

10. The fusion positioning method for power line carrier lines as described in claim 9, characterized in that, After adjusting the attention fusion weight matrix generated during the inference process when the information entropy is greater than a preset information entropy threshold, the method further includes: Based on the historical positioning results, a historical positioning trajectory is formed; The historical positioning trajectory is smoothed and predicted using Kalman filtering to obtain the predicted positioning trajectory. The current positioning result is corrected based on the predicted positioning trajectory.

Images