Power transmission line risk parameter determination method and device and electronic equipment

Abstract

The application discloses a power transmission line risk parameter determination method and device and electronic equipment. The method comprises the following steps: obtaining target multi-modal data of a target power transmission line; determining meteorological spatio-temporal characteristics corresponding to meteorological data, defect risk characteristics corresponding to defect data, and equipment state characteristics corresponding to equipment data; determining target multi-modal characteristics corresponding to the target power transmission line according to the meteorological spatio-temporal characteristics, the defect risk characteristics, and the equipment state characteristics; determining a risk propagation matrix corresponding to the target power transmission line according to the meteorological spatio-temporal characteristics and the equipment state characteristics; and determining a target risk parameter corresponding to the target power transmission line according to the target multi-modal characteristics and the risk propagation matrix. The application solves the technical problem of inaccurate determination of power transmission line risk parameters in the related art.

Description

Technical Field

[0001] This invention relates to the field of power systems, and more specifically, to a method, apparatus, and electronic device for determining risk parameters of transmission lines. Background Technology

[0002] As a critical infrastructure for power transmission, the reliability of power transmission lines is affected by various factors, such as severe weather, equipment aging, and external damage. Therefore, it is necessary to assess the risk parameters of power transmission lines to identify potential risks in advance, reduce the occurrence of power outages, and ensure the stability of power supply. However, in related technologies, there are technical problems with the inaccurate determination of risk parameters for power transmission lines.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a method, apparatus, and electronic device for determining risk parameters of transmission lines, in order to at least solve the technical problem of inaccurate determination of risk parameters of transmission lines in related technologies.

[0005] According to one aspect of the present invention, a method for determining risk parameters of a transmission line is provided, comprising: acquiring target multimodal data of a target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, the defect data being used to characterize defects of the target transmission line; determining meteorological spatiotemporal features corresponding to the meteorological data, defect risk features corresponding to the defect data, and equipment status features corresponding to the equipment data; determining target multimodal features corresponding to the target transmission line based on the meteorological spatiotemporal features, the defect risk features, and the equipment status features; determining a risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal features and the equipment status features, wherein the risk propagation matrix represents the risk propagation capability between any two towers among a plurality of towers on the target transmission line; and determining target risk parameters corresponding to the target transmission line based on the target multimodal features and the risk propagation matrix.

[0006] Optionally, determining the meteorological spatiotemporal features corresponding to the meteorological data includes: when the meteorological data includes meteorological time-series data corresponding to multiple towers on the target transmission line, determining meteorological frequency domain data corresponding to each of the multiple towers based on the meteorological time-series data corresponding to each of the multiple towers; retrieving a meteorological feature extraction model, wherein the meteorological feature extraction model includes a first convolutional kernel and a second convolutional kernel, the first convolutional kernel being determined based on multiple frequency band weights and frequency band masks corresponding to the multiple frequency band weights, the first convolutional kernel being used to extract temporal features, and the second convolutional kernel being used to extract spatial features; obtaining meteorological time features corresponding to each of the multiple towers based on the first convolutional kernel and the meteorological frequency domain data corresponding to each of the multiple towers; determining a tower association graph, wherein the tower association graph includes nodes corresponding to each of the multiple towers, edges corresponding to the association relationship between any two towers among the multiple towers, and edge weights corresponding to the association strength of the corresponding association relationship; determining the meteorological spatiotemporal features corresponding to the meteorological data based on the second convolutional kernel, the meteorological time features corresponding to each of the multiple towers, and the tower association graph.

[0007] Optionally, determining the defect risk features corresponding to the defect data includes: when the defect data includes a defect image, determining an initial defect feature map corresponding to the defect image; determining a first mask region and a second mask region corresponding to the initial defect feature map, wherein the first mask region is a feature region in the initial defect feature map whose feature change degree is greater than or equal to a feature change threshold, and the second mask region is a feature region in the initial defect feature map whose feature change degree is less than a feature change threshold; adjusting the defect feature map based on the first mask region and the second mask region to obtain a target defect feature map; and determining the defect risk features corresponding to the defect image based on the target defect feature map.

[0008] Optionally, determining the equipment status characteristics corresponding to the equipment data includes: when the equipment data includes fault timing data, determining defect events, maintenance events, and fault events corresponding to the fault timing data; determining a first event association relationship between the defect event and the maintenance event; determining a second event association relationship between the maintenance event and the fault event; determining a target event association relationship between the defect event and the fault event based on the first event association relationship and the second event association relationship; and determining the equipment status characteristics corresponding to the equipment data based on the target event association relationship.

[0009] Optionally, determining the target risk parameter corresponding to the target transmission line based on the target multimodal features and the risk propagation matrix includes: retrieving a risk association set, wherein the risk association set includes multiple risk association relationships corresponding to a reference transmission line, and the multiple risk association relationships are respectively the association relationships between reference multimodal features and reference risk parameters corresponding to the reference transmission line; filtering target association relationships from the risk association set based on the target multimodal features, wherein the target association relationship is the risk association relationship corresponding to a reference multimodal feature whose similarity index with the target multimodal features is greater than a similarity threshold among the multiple risk association relationships; and determining the target risk parameter corresponding to the target transmission line based on the risk propagation matrix and the target association relationships.

[0010] Optionally, determining the risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal characteristics and equipment status characteristics includes: determining the target geometric parameters corresponding to the target transmission line, wherein the target geometric parameters include the deformation index corresponding to multiple towers and the tower spacing between any two towers among the multiple towers, and the corresponding deformation index represents the degree of deformation of the corresponding equipment; retrieving the target prediction model, wherein the target prediction model is obtained by training an initial prediction model based on an auxiliary enhancement model and sample data, the auxiliary enhancement model is used to augment the sample images, and the sample data includes the sample images; inputting the deformation index corresponding to the multiple towers, the tower spacing between any two towers among the multiple towers, the meteorological spatiotemporal characteristics, and the equipment status characteristics into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line.

[0011] Optionally, before retrieving the target prediction model, the method includes: inputting the sample three-dimensional coordinate data and sample multimodal features corresponding to the sample transmission line into the structure recognition module of the auxiliary enhancement model to obtain the sample geometric structure parameters corresponding to the sample transmission line; inputting the sample image corresponding to the sample transmission line into the light removal feature recognition module of the auxiliary enhancement model to obtain the anti-light interference features corresponding to the sample transmission line, wherein the light removal feature recognition module is used to extract defect features that are not affected by light interference; inputting the sample geometric structure parameters and the anti-light interference features into the data augmentation module of the auxiliary enhancement model to augment the sample image to obtain multiple augmented images, wherein the multiple augmented images correspond to different visual parameters, and the visual parameters include geometric pose and illumination parameters.

[0012] Optionally, the method of determining the defect risk features corresponding to the defect data includes: when the defect data includes defect text, determining a semantic vector corresponding to the defect text; determining a reference risk label; extracting a risk keyword vector from the semantic vector based on the reference risk label; and determining the defect risk features corresponding to the defect text based on the risk keyword vector.

[0013] According to one aspect of the present invention, a device for determining risk parameters of a transmission line is provided, comprising: an acquisition module, configured to acquire target multimodal data of a target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is data containing defect information of the target transmission line; a first determination module, configured to determine meteorological spatiotemporal features corresponding to the meteorological data, defect risk features corresponding to the defect data, and equipment status features corresponding to the equipment data; a second determination module, configured to determine target multimodal features corresponding to the target transmission line based on the meteorological spatiotemporal features, the defect risk features, and the equipment status features; a third determination module, configured to determine a risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal features and the equipment status features; and a fourth determination module, configured to determine target risk parameters corresponding to the target transmission line based on the target multimodal features and the risk propagation matrix.

[0014] According to one aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the transmission line risk parameter determination method described in any of the preceding claims.

[0015] In this embodiment of the invention, target multimodal data of a target transmission line is acquired, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is used to characterize the defects of the target transmission line; the spatiotemporal meteorological features corresponding to the meteorological data, the defect risk features corresponding to the defect data, and the equipment status features corresponding to the equipment data are determined; based on the spatiotemporal meteorological features, the defect risk features, and the equipment status features, target multimodal features corresponding to the target transmission line are determined; based on the spatiotemporal meteorological features and the equipment status features, a risk propagation matrix corresponding to the target transmission line is determined, wherein the risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line; based on the target multimodal features and the risk propagation matrix, target risk parameters corresponding to the target transmission line are determined. By acquiring target multimodal data of the target transmission line, identifying the corresponding meteorological and spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics, and fusing them to generate target multimodal features, it is possible to capture the correlation characteristics between different data. Furthermore, by combining meteorological and spatiotemporal characteristics and equipment status characteristics, a risk propagation matrix is ​​generated to quantify the propagation law of risk in the target transmission line. In this way, by combining the target multimodal features and the risk propagation matrix, the risk parameters of the target transmission line can be determined more accurately, thus solving the technical problem of inaccurate determination of transmission line risk parameters in related technologies. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0017] Figure 1 This is a flowchart of a method for determining transmission line risk parameters according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the process for determining transmission line risk parameters in an optional embodiment of the present invention;

[0019] Figure 3 This is a structural block diagram of a transmission line risk parameter determination device according to an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention 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 the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a 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.

[0022] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0023] FLOPs: FLOPs (Floating Point Operations per Second) are used to measure the computational workload of a task, expressed in units of floating-point operations.

[0024] Swin Transformer V2: Swin Transformer V2 is a deep learning model based on the Transformer architecture.

[0025] Sigmoid: Sigmoid is a common activation function whose output ranges between (0, 1).

[0026] BERT-4: BERT-4 is a pre-trained language model used for natural language processing tasks.

[0027] Prompt Tuning: Prompt Tuning is a method of guiding a pre-trained language model to generate or understand specific text content by designing specific prompts.

[0028] Causal Transformer: Causal Transformer is an improved Transformer architecture.

[0029] RGB: RGB is a color representation of a pixel, using a combination of red, green, and blue light to represent various colors. In image processing, each pixel is typically composed of three components (R, G, B), representing the intensity of the red, green, and blue colors, respectively.

[0030] ST-GCN: ST-GCN is a deep learning model for processing spatiotemporal data.

[0031] Transformer: A deep learning architecture based on self-attention mechanism.

[0032] RAG: RAG is a model that combines retrieval and generation, enhancing the generation task by retrieving relevant information from a knowledge base.

[0033] FAISS Vector Database: A high-performance similarity search library used to quickly find the most similar vector to a query vector in a large dataset.

[0034] Focal Loss is an improved cross-entropy loss function.

[0035] Example 1

[0036] According to an embodiment of the present invention, an embodiment of a method for determining risk parameters of transmission lines is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0037] Figure 1 This is a flowchart of a method for determining transmission line risk parameters according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0038] S102, acquire target multimodal data of the target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is used to characterize the defects of the target transmission line.

[0039] In step S102 of this application, target multimodal data of the target transmission line is obtained.

[0040] This involves the target transmission line, which is a transmission line for which risk parameters need to be determined. This target transmission line is a line in the power system used to transmit electrical energy, including multiple towers, conductors, etc.

[0041] This involves target multimodal data, which comprises various types of data related to the target transmission line. These data originate from different sources and possess different formats and characteristics, specifically including meteorological data, defect data, and equipment data. Through fusion and analysis, this data can provide comprehensive information support for the risk assessment of the transmission line, helping to more accurately determine risk parameters.

[0042] This involves meteorological data, which reflects meteorological information corresponding to the target transmission line, including but not limited to temperature, humidity, wind speed, wind direction, rainfall, and the number of lightning strikes. This data can be time-series data. The meteorological data reflects the meteorological conditions of the environment in which the transmission line is located. For example, extreme weather (such as heavy rain, strong winds, and thunderstorms) may threaten the stability and safety of the transmission line. Analyzing the meteorological data can help predict and assess these risks.

[0043] This involves defect data, which is information characterizing defects in the target transmission line; that is, the defect data contains information about the defects in the target transmission line. Defect data includes defect images (such as images of transmission lines taken by drones during inspections, which include defects such as insulator cracks and conductor wear), defect text (such as descriptions of defects in inspection reports), or other forms of text data.

[0044] This includes equipment data, which reflects information about the equipment used in the target transmission line, such as its operating status, maintenance records, and fault history. For example, data such as the equipment's service life, voltage level, maintenance cycle, and number of faults all fall under equipment data. This data helps in understanding the health and reliability of the equipment and is of significant reference value for assessing the risk parameters of the transmission line.

[0045] This involves defects in the target transmission line, which refer to various problems and faults that occur during the operation of the transmission line, potentially affecting its safe operation. The types of defects are diverse, including tower tilting, conductor wear, insulator cracks, and loose hardware. By detecting and analyzing defect data, these problems can be identified and addressed promptly, reducing the risks to the transmission line.

[0046] The target multimodal data of the target transmission line includes data corresponding to various modes, such as meteorological data, defect data, and equipment data. Through the target multimodal data, the risk status of the target transmission line can be understood from multiple perspectives, thus providing a reliable data analysis basis for subsequent risk assessment.

[0047] S104, determine the spatiotemporal characteristics of meteorological data, the defect risk characteristics of defect data, and the equipment status characteristics of equipment data.

[0048] In step S104 of this application, meteorological spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics are determined.

[0049] This involves meteorological spatiotemporal characteristics, which are determined based on meteorological data and reflect the characteristics of the target transmission line in both time and space. It reflects the changing patterns and trends of meteorological conditions at different times and spatial locations. Specifically, meteorological spatiotemporal characteristics can be understood from both temporal and spatial perspectives:

[0050] Meteorological spatiotemporal characteristics, at the temporal level, reflect the changes in meteorological data over time, such as periodic changes (e.g., seasonal variations, diurnal variations), trend changes (e.g., long-term global warming trends), and abrupt changes (e.g., sudden thunderstorms, heavy rains). For example, by analyzing the time-series information of meteorological data (e.g., meteorological time-series data corresponding to multiple power poles), characteristics such as low-frequency trends (e.g., annual average temperature variation trends), medium-frequency periods (e.g., seasonal temperature variations), and high-frequency abrupt changes (e.g., an increase in the number of sudden lightning strikes) can be extracted.

[0051] At the spatial level, the spatiotemporal characteristics of meteorological data reflect its spatial distribution, such as differences in geographical location and spatial propagation paths. For example, analyzing the spatial distribution of meteorological data can determine the distribution patterns of meteorological conditions along power transmission lines, as well as the propagation paths and impact ranges of meteorological conditions from one location to another, such as the impact of typhoon paths on power transmission towers along the line.

[0052] This involves defect risk characteristics, which are determined based on defect data and reflect the characteristics of defects in the target transmission line (i.e., defect features), as well as the corresponding risk characteristics. These defect risk characteristics reflect the potential threat posed by defects to the safe operation of the target transmission line. Specifically, defect features can be understood from the following aspects: defect type characteristics, defect location characteristics, defect severity characteristics, and defect association characteristics.

[0053] Defect type characteristics: These indicate the type of defect, such as insulator cracks, conductor wear, loose hardware, etc. Different types of defects have different impacts on the safety of transmission lines; therefore, defect type is one of the important factors in assessing defect risk.

[0054] Defect location characteristics: These indicate the specific location of the defect within the transmission line, such as the tower number or conductor segment. Location information helps determine the extent to which a defect impacts the overall safety of the transmission line; for example, a defect located in a critical position may pose a greater threat to the stability of the transmission line.

[0055] Defect severity characteristics: These represent the severity of a defect, such as crack length or degree of wear. The severity of a defect directly affects its risk level. By analyzing defect data, the severity of a defect can be quantified, thereby assessing its risk.

[0056] Defect correlation characteristics: These indicate the relationships between defects, such as whether multiple defects occur simultaneously or whether a causal relationship exists. The correlation between defects can affect the overall risk assessment; for example, the simultaneous occurrence of multiple defects may lead to greater safety hazards.

[0057] Additionally, the risk characteristics corresponding to defects can include the risk level and risk type, which can be represented by risk keyword vectors.

[0058] This involves equipment status characteristics, which are determined based on equipment data and characterize the operating status of equipment on the target transmission line, reflecting the equipment's health and reliability. Specifically, equipment status characteristics can be understood from the following aspects: operating status characteristics, maintenance characteristics, fault history characteristics, and event correlation characteristics.

[0059] Operating status characteristics of equipment: These represent the current operating status of the equipment, such as whether the equipment is operating normally or whether there are any abnormal operating conditions. For example, parameters such as the equipment's service life and voltage level can reflect the basic operating status of the equipment.

[0060] Maintenance characteristics of equipment status reflect the maintenance status of the equipment, such as the maintenance cycle and the time of the most recent maintenance. Maintenance characteristics help to understand the maintenance status of the equipment, identify and deal with potential problems in a timely manner, and thus assess the health status of the equipment.

[0061] Fault history characteristics of equipment status reflect the equipment's historical failure history, such as the number of failures and the types of failures. Fault history characteristics can reflect the equipment's reliability and stability. By analyzing the fault history, potential future problems can be predicted, providing a reference for risk assessment.

[0062] Event correlation characteristics of equipment status reflect the relationships between different events in equipment data, such as the correlation between defect events and maintenance events, or the correlation between maintenance events and failure events. By analyzing the correlations between events, we can better understand the operating status and potential risks of the equipment.

[0063] S106. Based on meteorological spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics, determine the target multimodal characteristics corresponding to the target transmission line.

[0064] In step S106 provided in this application, the target multimodal characteristics corresponding to the target transmission line are determined based on meteorological spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics.

[0065] This involves target multimodal features, which are comprehensive features obtained by fusing meteorological spatiotemporal features, defect risk features, and equipment status features. It is a multi-dimensional feature set that can comprehensively reflect the overall status and risk of the target transmission line under different modal data. Through this fusion, the correlation and interaction between different modal data can be captured, thereby providing more comprehensive and accurate information support for the risk assessment of transmission lines.

[0066] Specifically, the process of constructing target multimodal features can include the following aspects:

[0067] Feature fusion: This involves fusing meteorological spatiotemporal features, defect risk features, and equipment status features to form a unified feature space. For example, this fusion can be a simple concatenation, or it can involve weighted or nonlinear combinations using specific algorithms to better capture the relationships between different features.

[0068] Feature optimization: In the fused feature space, feature selection or dimensionality reduction operations are required to remove redundant information, improve the quality and effectiveness of features, thereby reducing computational complexity and improving the generalization ability of the model.

[0069] Feature representation: The final target multimodal feature can be a vector or tensor representation that contains comprehensive information about the transmission line in terms of weather, defects and equipment status.

[0070] S108. Based on the meteorological spatiotemporal characteristics and equipment status characteristics, determine the risk propagation matrix corresponding to the target transmission line. The risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line.

[0071] In step S108 provided in this application, a risk propagation matrix corresponding to the target transmission line is determined based on meteorological spatiotemporal characteristics and equipment status characteristics.

[0072] This involves a risk propagation matrix, which is used to describe the risk propagation relationship between multiple towers on a target transmission line. The risk propagation matrix is ​​constructed based on meteorological spatiotemporal characteristics and equipment status characteristics, and can quantify the propagation law of risk in transmission lines, providing important dynamic information for risk assessment.

[0073] This involves multiple power poles, which refer to the various supporting structures that constitute the target transmission line. These poles support the conductors of the transmission line, ensuring its stability and safety. Each pole has a specific location, structural parameters (such as height and material), and operating status. In the risk propagation matrix, each pole is considered a node, and the risk propagation path is described through the connections between poles (such as geographical distance and electrical connections).

[0074] This involves risk propagation capability, which is the ability of a risk to spread from one pole to another. It is influenced by various factors, including meteorological conditions (such as wind speed and lightning strikes), the geographical distance between poles, the structural condition of the poles (such as the pole's tilt angle and conductor sag), and the operational status of the equipment (such as the equipment's health condition and failure history). For example, if two poles are close together and the meteorological conditions are severe (such as strong winds), the risk propagation capability may be high, meaning that the risk is more easily transmitted from one pole to the other.

[0075] By comprehensively considering meteorological conditions (such as wind speed and lightning strikes) and equipment conditions (such as tower tilt and conductor sag), a risk propagation matrix is ​​determined, which can accurately quantify the ability of risk to propagate from one tower to another. This provides dynamic information for risk assessment of target transmission lines, helping to predict and prevent the spread of risks in advance.

[0076] S110, based on the target's multimodal characteristics and the risk propagation matrix, determines the target risk parameters corresponding to the target transmission line.

[0077] In step S110 provided in this application, the target risk parameters corresponding to the target transmission line are determined based on the target multimodal characteristics and the risk propagation matrix.

[0078] This involves target risk parameters, which are parameters determined by comprehensively considering the multimodal characteristics of the target and the risk propagation matrix to characterize the overall risk level of the target transmission line. These parameters can quantify the risk level of the transmission line under its current operating state and environmental conditions. Target risk parameters may include the risk level of the target transmission line, spatiotemporal risk heatmaps, etc.

[0079] Since the target multimodal features are obtained based on multimodal data, they can comprehensively reflect the characteristics of the target transmission line. The risk propagation matrix can quantify the propagation of risk. Therefore, by combining the target multimodal features and the risk propagation matrix, the risk situation of the transmission line under the current operating status and environmental conditions can be assessed more comprehensively and accurately, thereby achieving accurate determination of the target risk parameters corresponding to the target transmission line.

[0080] Through the above steps S102-S110, target multimodal data of the target transmission line is obtained. This target multimodal data includes meteorological data, defect data, and equipment data. The defect data is used to characterize the defects of the target transmission line. The spatiotemporal characteristics corresponding to the meteorological data, the defect risk characteristics corresponding to the defect data, and the equipment status characteristics corresponding to the equipment data are determined. Based on the spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics, the target multimodal characteristics corresponding to the target transmission line are determined. Based on the spatiotemporal characteristics and equipment status characteristics, the risk propagation matrix corresponding to the target transmission line is determined. This risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line. Based on the target multimodal characteristics and the risk propagation matrix, the target risk parameters corresponding to the target transmission line are determined. By acquiring target multimodal data of the target transmission line, identifying the corresponding meteorological and spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics, and fusing them to generate target multimodal features, it is possible to capture the correlation characteristics between different data. Furthermore, by combining meteorological and spatiotemporal characteristics and equipment status characteristics, a risk propagation matrix is ​​generated to quantify the propagation law of risk in the target transmission line. In this way, by combining the target multimodal features and the risk propagation matrix, the risk parameters of the target transmission line can be determined more accurately, thus solving the technical problem of inaccurate determination of transmission line risk parameters in related technologies.

[0081] As an optional embodiment, determining the meteorological spatiotemporal features corresponding to meteorological data includes: when the meteorological data includes meteorological time-series data corresponding to multiple towers on the target transmission line, determining meteorological frequency domain data corresponding to each of the multiple towers based on the meteorological time-series data corresponding to each of the multiple towers; retrieving a meteorological feature extraction model, wherein the meteorological feature extraction model includes a first convolutional kernel and a second convolutional kernel, the first convolutional kernel being determined based on multiple frequency band weights and frequency band masks corresponding to the multiple frequency band weights, the first convolutional kernel being used to extract temporal features, and the second convolutional kernel being used to extract spatial features; obtaining meteorological time features corresponding to each of the multiple towers based on the first convolutional kernel and the meteorological frequency domain data corresponding to each of the multiple towers; determining a tower association graph, wherein the tower association graph includes nodes corresponding to each of the multiple towers, edges corresponding to the association relationship between any two towers among the multiple towers, and edge weights corresponding to the association strength of the corresponding association relationship; and determining the meteorological spatiotemporal features corresponding to the meteorological data based on the second convolutional kernel, the meteorological time features corresponding to each of the multiple towers, and the tower association graph.

[0082] This embodiment describes the specific steps for determining the spatiotemporal characteristics of meteorological data.

[0083] This includes meteorological time-series data, which consists of meteorological data for the corresponding towers on the target transmission line. This data is recorded in chronological order, reflecting changes in meteorological conditions over time. For example, it includes the time-series changes in wind speed, temperature, humidity, and other data for each tower location.

[0084] This includes meteorological frequency domain data, which is obtained by performing time-frequency conversion on meteorological time-series data using frequency domain analysis methods such as Fourier transform. It reflects information about meteorological data at different frequency components and can reveal the periodicity and trend characteristics of the meteorological data.

[0085] This involves a meteorological feature extraction model, which is used to extract temporal and spatial features from meteorological frequency domain data. The meteorological feature extraction model includes a first convolution kernel and a second convolution kernel.

[0086] This involves a first convolutional kernel, specifically designed for extracting temporal features. This kernel is determined based on multiple frequency band weights and corresponding frequency band masks. By performing a convolution operation with meteorological frequency domain data, the first convolutional kernel can extract the temporal features of the meteorological data, such as meteorological trend features, meteorological periodic features, and meteorological abrupt change features. This first convolutional kernel can be a frequency dynamic convolution (FDConv).

[0087] This involves a second convolutional kernel, which is specifically designed for extracting spatial features. Based on the first convolutional kernel, the output of the first convolutional kernel is convolved to extract features in the spatial dimension, such as the spatial distribution and propagation path of meteorological conditions between different towers.

[0088] This involves multiple frequency band weights, which are used to weight meteorological frequency domain data in different frequency bands. Each frequency band weight corresponds to a type of frequency band, such as high-frequency, mid-frequency, and low-frequency. The high-frequency, mid-frequency, and low-frequency bands are determined based on the frequency range of the meteorological frequency domain data. For example, a first frequency threshold and a second frequency threshold are set, where the first frequency threshold is greater than the second frequency threshold. Then, in the meteorological frequency domain data, the frequency bands greater than or equal to the first frequency threshold are considered high-frequency, those between the first and second frequency thresholds are considered mid-frequency, and those less than or equal to the second frequency threshold are considered low-frequency.

[0089] This involves frequency band masks, which are used to highlight the corresponding frequency bands. Each frequency band mask corresponds to a specific frequency band. By performing element-wise multiplication with meteorological frequency domain data, data of a specific frequency band can be extracted. For example, for a high-frequency band mask, the data corresponding to the mid-frequency and low-frequency bands are masked, retaining only the high-frequency band.

[0090] This involves time characteristics, which are data features reflected at the time level. For example, meteorological time characteristics are time-related features extracted from meteorological frequency domain data. These features reflect the changing patterns of meteorological data in the time dimension, such as periodicity, trend, and abrupt change.

[0091] This involves spatial characteristics, which are data features reflected at the spatial level. For example, based on meteorological time characteristics, combined with the pole-to-tower correlation diagram, further spatially related features are extracted to obtain meteorological spatiotemporal characteristics, which further reflect the distribution pattern of meteorological data in the spatial dimension, such as the spatial propagation path and correlation of meteorological conditions between different poles.

[0092] This includes a tower association diagram, a graph structure used to describe the relationships between multiple towers in a transmission line. It includes:

[0093] Nodes corresponding to multiple towers: Each tower is considered a node in the diagram, and the node contains information such as the tower's location and structural parameters.

[0094] Edges representing the connections between any two towers in a set of towers: An edge represents the connection between two towers.

[0095] The edge weights corresponding to the strength of the association relationship: The edge weights represent the strength of the association between two towers, which can be determined by meteorological correlation coefficients, geographical distance, etc.

[0096] By performing time-frequency conversion to obtain meteorological frequency domain data, and combining it with the first convolution kernel, the second convolution kernel, and the tower correlation diagram, the spatiotemporal variation patterns of meteorological conditions along the transmission line can be more accurately depicted, providing precise spatiotemporal characteristic information for risk assessment.

[0097] As an optional embodiment, determining the defect risk characteristics corresponding to defect data includes: when the defect data includes a defect image, determining an initial defect feature map corresponding to the defect image; determining a first mask region and a second mask region corresponding to the initial defect feature map, wherein the first mask region is a feature region in the initial defect feature map whose feature change degree is greater than or equal to a feature change threshold, and the second mask region is a feature region in the initial defect feature map whose feature change degree is less than a feature change threshold; adjusting the defect feature map based on the first mask region and the second mask region to obtain a target defect feature map; and determining the defect risk characteristics corresponding to the defect image based on the target defect feature map.

[0098] This embodiment describes the specific steps for determining the defect risk characteristics corresponding to defect data.

[0099] This includes defect images, which are images containing information about defects in power transmission lines. These defect images can be obtained through drone inspections or other methods.

[0100] This involves an initial defect feature map, which is a preliminary feature map extracted from the defect image. This initial feature map includes multi-scale features. For example, multi-scale features (such as the local texture and global structure of insulator cracks) of the defect image can be extracted through image processing and feature extraction algorithms combined with a sliding window mechanism to obtain the initial defect feature map.

[0101] This involves a first mask region, which is the feature area in the initial defect feature map where the degree of feature change is greater than or equal to a feature change threshold. These areas typically contain obvious defect features, such as cracks and wear. By setting a feature change threshold, these important defect areas can be filtered out for further analysis and processing.

[0102] This involves a second mask region, which is a feature region in the initial defect feature map where the degree of feature change is less than the feature change threshold. These regions are relatively smooth and can be regarded as areas without defects.

[0103] This involves the degree of feature change, which is the magnitude of the change in the feature value. It can be measured by calculating the difference or gradient between adjacent pixels in the initial feature map. The degree of feature change reflects the prominence of the defect; a larger feature change usually corresponds to a significant defect, while a smaller feature change may correspond to noise or a weaker defect.

[0104] This involves a feature change threshold, which is used to distinguish between the first mask region and the second mask region. This threshold can be adjusted according to the actual application and dataset to achieve the best feature extraction results.

[0105] This involves a target defect feature map, which is the final defect feature map after adjustment. By combining information from the first and second mask regions, the initial defect feature map is adjusted to obtain a more accurate and robust defect feature map. The target defect feature map can more clearly reflect the location, shape, and intensity of the defect, providing a more reliable foundation for subsequent defect risk feature extraction.

[0106] By determining the initial defect feature map corresponding to the defect image and identifying the first mask region with significant feature changes and the second mask region with smaller changes, defect information can be enhanced and non-defect information weakened. Based on these mask regions, the initial defect feature map can be adjusted to obtain the target defect feature map, which can more clearly present the details of the defect.

[0107] As an optional embodiment, determining the equipment status characteristics corresponding to the equipment data includes: when the equipment data includes fault timing data, determining defect events, maintenance events, and fault events corresponding to the fault timing data; determining a first event correlation between the defect events and maintenance events; determining a second event correlation between the maintenance events and fault events; determining a target event correlation between the defect events and fault events based on the first event correlation and the second event correlation; and determining the equipment status characteristics corresponding to the equipment data based on the target event correlation.

[0108] This embodiment describes the specific steps for determining the device status characteristics corresponding to the device data.

[0109] This includes fault timing data, which is a data sequence that records fault information that occurs in the device at different points in time.

[0110] This includes defect events, which reflect potential malfunctions or anomalies that occur during equipment operation. These events may not immediately lead to complete equipment failure, but may develop into failures over time or under specific conditions. Defect event records typically include information such as the type of defect, the time of discovery, and the location.

[0111] This includes maintenance events, which are maintenance events performed to address equipment defects or malfunctions. Records of these events typically include information such as the time and content of the maintenance. Maintaining maintenance event records helps in understanding the equipment's maintenance history and the effectiveness of maintenance.

[0112] This involves the first-event correlation, which is the relationship between defect events and maintenance events. This correlation reflects whether maintenance was performed after a defect event occurred, whether it was performed on time, and the extent of the maintenance. For example, a defect event may trigger one or more maintenance events, the purpose of which is to repair or mitigate the impact of the defect event. By analyzing the first-event correlation, we can understand the impact of defect events on equipment operation and the effectiveness of maintenance activities.

[0113] This involves the second event correlation, which is the relationship between maintenance events and failure events. This correlation reflects the impact of maintenance activities on equipment failure. For example, a maintenance event might successfully prevent a failure from occurring, or fail to repair a defect, or cause a maintenance delay, ultimately leading to a failure. By analyzing the second event correlation, the effectiveness of maintenance activities and the reliability of the equipment can be evaluated.

[0114] This involves the target event correlation, which refers to the final correlation between defect events and failure events. This correlation integrates the first event correlation and the second event correlation, reflecting how defect events affect the occurrence of failure events through maintenance events. By determining the target event correlation, a more comprehensive understanding of the equipment's operating status and failure modes can be achieved.

[0115] As an optional embodiment, the target risk parameters corresponding to the target transmission line are determined based on the target multimodal features and the risk propagation matrix, including: retrieving a risk association set, wherein the risk association set includes multiple risk associations corresponding to the reference transmission line, and the multiple risk associations are the associations between the reference multimodal features and the reference risk parameters corresponding to the reference transmission line; based on the target multimodal features, selecting target associations from the risk association set, wherein the target associations are the risk associations corresponding to the reference multimodal features whose similarity index with the target multimodal features is greater than a similarity threshold among the multiple risk associations; and determining the target risk parameters corresponding to the target transmission line based on the risk propagation matrix and the target associations.

[0116] This embodiment describes the specific steps for determining the target risk parameters corresponding to the target transmission line based on the target's multimodal characteristics and the risk propagation matrix.

[0117] This involves a risk association set, which is a collection of multiple risk associations. These risk associations describe the correspondence between the multimodal characteristics of a reference transmission line and its risk parameters. Through these associations, reference features similar to the multimodal characteristics of the target transmission line can be found, thereby inferring the risk parameters of the target transmission line. This risk association set can be a set of domain knowledge related to the transmission line.

[0118] This involves reference transmission lines, which are used to establish a risk association set. These lines have known multimodal characteristics and risk parameters and can be used as a reference to assess the risk of the target transmission line.

[0119] This involves referencing multimodal features, which are multimodal data features related to a reference transmission line. These features include meteorological spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics, corresponding to the multimodal features of the target transmission line. The reference multimodal features are used to establish a risk association set, providing a reference for assessing the risk of the target transmission line.

[0120] This involves reference risk parameters, which are known risk parameters related to the reference transmission line.

[0121] This involves target associations, which are selected from the risk association set based on similarity to the target multimodal features. This selection is based on a similarity index; only when the similarity index between the reference multimodal feature and the target multimodal feature is greater than a set similarity threshold will the corresponding association be selected as a target association.

[0122] This involves a similarity index, which measures the degree of similarity between target multimodal features and reference multimodal features. The similarity index can be calculated using various methods, such as Euclidean distance and cosine similarity. A higher similarity index indicates greater similarity between the two features.

[0123] This involves a similarity threshold, which is used to determine whether two features are sufficiently similar. Only when the similarity index is greater than or equal to the similarity threshold is the reference multimodal feature considered sufficiently similar to the target multimodal feature, and the corresponding association is selected as the target association. The similarity threshold can be adjusted according to the actual application and dataset to achieve the best matching results.

[0124] The risk association set contains multiple risk associations corresponding to reference transmission lines. These associations reflect the mapping relationship between reference multimodal features and reference risk parameters. By comparing the multimodal features of the target transmission line with the reference features in the risk association set, and selecting target associations with high similarity, known risk parameters can be used to infer the risk status of the target transmission line. This allows for full utilization of historical data and prior knowledge, avoiding risk assessment from scratch, reducing uncertainty, and making the risk assessment results closer to reality, thereby improving the accuracy and reliability of the assessment.

[0125] As an optional embodiment, based on meteorological spatiotemporal characteristics and equipment status characteristics, a risk propagation matrix corresponding to the target transmission line is determined, including: determining the target geometric parameters corresponding to the target transmission line, wherein the target geometric parameters include the deformation index corresponding to each of the multiple towers, and the tower spacing between any two towers among the multiple towers, and the corresponding deformation index represents the degree of deformation of the corresponding equipment; retrieving the target prediction model, wherein the target prediction model is obtained by training an initial prediction model based on an auxiliary enhancement model and sample data, the auxiliary enhancement model is used to augment the sample images, and the sample data includes sample images; inputting the deformation index corresponding to each of the multiple towers, the tower spacing between any two towers among the multiple towers, the meteorological spatiotemporal characteristics, and the equipment status characteristics into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line.

[0126] This embodiment describes the specific steps for determining the risk propagation matrix corresponding to the target transmission line based on meteorological spatiotemporal characteristics and equipment status characteristics.

[0127] This involves target geometric structure parameters, which are parameters reflecting the geometric structural characteristics of the target transmission line. Specifically, these target geometric structure parameters include:

[0128] Deformation indices for multiple towers: These represent the degree of deformation of each tower on the target transmission line. A higher deformation index indicates greater tower deformation, potentially having a greater impact on the safe operation of the transmission line. Deformation indices can be calculated by measuring the tower's geometric parameters (such as tilt angle and displacement).

[0129] Tower spacing between any two towers in a group of towers: the distance between any two towers, used to describe the distribution of towers.

[0130] This involves a target prediction model, which is used to predict the risk capability of a target transmission line. This model is trained based on an auxiliary enhancement model and sample data. The target prediction model can be a spatiotemporal risk propagation model.

[0131] This involves an auxiliary enhancement model, which is used to augment sample images. This auxiliary enhancement model is implemented through neural implicit field (NIF) risk modeling.

[0132] This involves sample data, which is the dataset used to train the target prediction model. The sample data includes sample images and other relevant feature data (such as deformation index, tower spacing, meteorological spatiotemporal characteristics, equipment status characteristics, etc.).

[0133] By using an auxiliary enhancement model, sample images can be effectively expanded, enhancing the generalization ability of the target prediction model. Furthermore, it can comprehensively consider multi-dimensional information such as the geometric structural parameters of the target transmission line (e.g., the deformation index and tower spacing of the towers), meteorological spatiotemporal characteristics, and equipment status characteristics. By using the target prediction model trained based on the auxiliary enhancement model and sample data, the risk propagation matrix can be accurately determined, thereby better reflecting the propagation law of risk in the transmission line.

[0134] As an optional embodiment, before retrieving the target prediction model, the method includes: inputting the sample's three-dimensional coordinate data and multimodal features corresponding to the sample transmission line into the structure recognition module of the auxiliary enhancement model to obtain the sample's geometric structure parameters corresponding to the sample transmission line; inputting the sample image corresponding to the sample transmission line into the light removal feature recognition module of the auxiliary enhancement model to obtain the anti-light interference features corresponding to the sample transmission line, wherein the light removal feature recognition module is used to extract defect features that are not affected by light interference; and inputting the sample's geometric structure parameters and anti-light interference features into the data augmentation module of the auxiliary enhancement model to augment the sample image to obtain multiple augmented images, wherein the multiple augmented images correspond to different visual parameters, including geometric pose and illumination parameters.

[0135] This embodiment describes the specific steps before retrieving the target prediction model.

[0136] This involves sample 3D coordinate data, which consists of the 3D spatial coordinate information of each tower and equipment in the sample transmission line. This data is used to describe the geometry of the transmission line, including the position, height, and orientation of the towers.

[0137] This involves multimodal features of the samples, which are various types of data features extracted from the sample transmission lines, including meteorological data, defect data, and equipment data. These features reflect the state and performance of the transmission lines under different modes, providing comprehensive information support for risk assessment.

[0138] This includes a structure recognition module, which is used to identify the geometric structural parameters of transmission lines from sample three-dimensional coordinate data and sample multimodal features.

[0139] This involves sample geometric parameters, which are the geometric information of the transmission line extracted by the structure recognition module. These parameters include the deformation index of the towers and the tower spacing.

[0140] This includes a light-removal feature recognition module, which is used to extract defect features from sample images that are not affected by illumination. By extracting defect features that are not affected by illumination, the influence of illumination changes on the image can be removed, making the obtained defect features more stable, thereby helping to improve the accuracy and robustness of defect detection.

[0141] This includes anti-light interference features, which are defect features extracted by the light removal feature recognition module that are unaffected by light interference. These features can more accurately reflect the true condition of the defect and are unaffected by changes in lighting conditions.

[0142] This includes a data augmentation module, which is used to augment sample images, increasing the diversity and quantity of sample data and improving the generalization ability of the target prediction model. Augmented images can be generated by changing visual parameters such as the geometric pose and illumination parameters of the image.

[0143] This involves multiple augmented images, which are generated from sample images by changing visual parameters such as geometric pose and lighting, thus increasing the diversity and quantity of sample data. These augmented images are used to train the target prediction model, improving its generalization ability and robustness.

[0144] This involves visual parameters, which are various parameters that affect the visual effect of an image, including geometric pose and lighting parameters. Variations in these parameters can simulate different shooting conditions and environments, increasing the diversity of sample data.

[0145] This involves geometric pose, which reflects the geometric features of an object, such as position, orientation, and angle. By changing the geometric pose, image variants at different angles and positions can be generated, increasing the diversity of the sample data.

[0146] This involves lighting parameters, which are various parameters that affect the lighting effect of an image, such as light intensity, light direction, and light color. By changing the lighting parameters, image variants under different lighting conditions can be generated, increasing the diversity of the sample data.

[0147] Through these steps and modules, the augmentation model can effectively extract and expand sample data, improve the performance and generalization ability of the target prediction model, and thus more accurately determine the risk propagation matrix.

[0148] As an optional embodiment, determining the defect risk features corresponding to the defect data includes: when the defect data includes defect text, determining the semantic vector corresponding to the defect text; determining a reference risk label; extracting a risk keyword vector from the semantic vector based on the reference risk label; and determining the defect risk features corresponding to the defect text based on the risk keyword vector.

[0149] This embodiment describes the specific steps for determining the defect risk characteristics corresponding to defect data.

[0150] This includes defect text, which is descriptive textual information about defects in transmission lines, typically recorded in inspection reports. This text can describe the type, location, and severity of the defect, for example, "An insulator has a crack, located on insulator number 12 on tower, with a crack length of approximately 5 centimeters."

[0151] This involves semantic vectors, which are vectors obtained by semantically processing defective text. These vectors capture the semantic information of the text. For example, a pre-trained language model can encode defective text into a high-dimensional semantic vector, where each dimension represents a feature of the text in a certain semantic space.

[0152] This involves reference risk labels, which are at least one predefined risk-related label or keyword that is directly related to the defect risk of transmission lines, in order to guide the language processing model to analyze information related to the labels.

[0153] This involves risk keyword vectors, which are vectors extracted from semantic vectors that are related to the reference risk tags. These keyword vectors can more accurately reflect the risk information in the defective text.

[0154] By transforming complex textual expressions into risk keyword vectors relevant to risk assessment through the above steps, irrelevant textual interference can be eliminated, thereby improving the accuracy of risk analysis while reducing the complexity of subsequent calculations.

[0155] Based on the above embodiments and optional embodiments, an optional implementation method is provided, which is described in detail below.

[0156] In related technologies, transmission lines, as critical infrastructure for power transmission, are subject to various factors affecting their operational reliability, such as severe weather, equipment aging, and external damage. Therefore, it is necessary to assess the risk parameters of transmission lines to identify potential risks in advance, reduce the occurrence of power outages, and ensure the stability of power supply. However, in related technologies, there is a technical problem of inaccurate determination of transmission line risk parameters.

[0157] Specifically, in the field of transmission line operation and maintenance, the accuracy and timeliness of risk assessment are the core challenges to ensuring power grid security, and the following technical bottlenecks exist in related technologies:

[0158] (1) Insufficient multimodal data fusion. Meteorological data (time series), defect data (image / text), and ledger data (structured) lack unified feature space modeling, and traditional splicing and fusion cannot capture nonlinear correlations (such as the synergistic risk of "ice thickness + insulator crack").

[0159] (2) The spatiotemporal feature modeling is crude. Existing models do not adequately characterize the spatial propagation of meteorological data (such as the impact of typhoon paths on towers along the line) and temporal dependence (such as equipment aging trends), resulting in long-term risk prediction bias;

[0160] (3) Lack of knowledge-driven ability. There is a lack of dynamic retrieval and reasoning ability for relevant knowledge of transmission lines and historical cases, making it difficult to explain the causes of risks (such as "whether the high lightning tripping rate is related to the excessive grounding resistance of the tower").

[0161] (4) Conflict between real-time performance and accuracy. The computational load of deep learning models increases dramatically in complex scenarios (e.g., FLOPs > 2000G), making it difficult to meet the real-time requirements of online early warning (e.g., second-level response is required in extreme weather).

[0162] There is currently no effective solution to the above problems.

[0163] In view of this, the optional embodiments of the present invention provide a method for determining the risk parameters of transmission lines, which can also be called a real-time risk assessment method for transmission lines, and can effectively solve the above-mentioned technical problems existing in related technologies.

[0164] To address these issues, a unified framework integrating multimodal dynamic convolution, neural implicit fields, and RAG techniques is proposed, with the following objectives:

[0165] (1) High-precision risk feature extraction. By dynamically convolutionally decomposing the frequency domain features of multimodal data (such as low-frequency trends and high-frequency abrupt changes in meteorological data), the ability to perceive small defects (such as microcracks in conductors) and spatiotemporal associated risks (such as subsidence of tower groups caused by regional rainstorms) is improved;

[0166] (2) Robust spatiotemporal risk modeling. By combining neural implicit field and spatiotemporal graph network, the geometric structure of the equipment (such as the 3D model of the tower) and the spatiotemporal propagation path (such as the movement trajectory of thunderstorm) are explicitly modeled to enhance the invariance of illumination and the robustness of occlusion;

[0167] (3) Knowledge enhances reasoning ability. By using RAG technology to search industry knowledge bases and historical cases, the causes of risks can be explained (e.g., "The current risk level has been increased because the discharge probability of a certain type of insulator increases by 30% when the humidity is >85%").

[0168] (4) Real-time and efficient evaluation. The design of a shared feature backbone and lightweight modules reduces the computational load by more than 40%. At the same time, edge device deployment is achieved through model compression technology (such as knowledge distillation) to meet the real-time early warning requirements.

[0169] Figure 2 This is a schematic diagram of the process for determining transmission line risk parameters in an optional embodiment of the present invention, as shown below. Figure 2 As shown, this paper mainly constructs a four-module collaborative transmission line risk assessment framework, including a multimodal feature extraction module, a spatiotemporal dynamic modeling module, a RAG knowledge enhancement module, and a joint optimization module. Through cross-module frequency domain interaction, spatiotemporal graph propagation, and knowledge injection mechanisms, the risk parameters of transmission lines are accurately determined. A detailed description follows.

[0170] S1. Obtain target multimodal data of the target transmission line, including meteorological data, defect data, and equipment data. The defect data is used to characterize the defects of the target transmission line.

[0171] S2. Determine the spatiotemporal characteristics corresponding to meteorological data, the defect risk characteristics corresponding to defect data, and the equipment status characteristics corresponding to equipment data. This can be achieved through a multimodal feature extraction module.

[0172] Specifically as follows:

[0173] (1) For determining the spatiotemporal characteristics of meteorology, including:

[0174] Given that the meteorological data includes meteorological time-series data corresponding to multiple towers on the target transmission line, the following steps are taken: First, a meteorological feature extraction model is retrieved. This model includes a first convolutional kernel and a second convolutional kernel. The first convolutional kernel is determined based on multiple frequency band weights and corresponding frequency band masks. The first convolutional kernel is used to extract temporal features, and the second convolutional kernel is used to extract spatial features. Based on the first convolutional kernel and the meteorological frequency-series data corresponding to the multiple towers, meteorological temporal features corresponding to each tower are obtained. A tower association graph is then determined. This graph includes nodes corresponding to each tower, edges representing the association relationships between any two towers, and edge weights representing the association strength. Finally, based on the second convolutional kernel, the meteorological temporal features corresponding to the multiple towers, and the tower association graph, the meteorological spatiotemporal features corresponding to the meteorological data are determined.

[0175] The first convolutional kernel can be a spatial convolutional kernel, determined through frequency dynamic convolution (FDConv). The specific steps are as follows:

[0176] Introducing Fourier Disjoint Weights (FDW), meteorological time-series data is decomposed into low (trend) frequency bands, medium (periodic) frequency bands, and high (abrupt) frequency bands, as shown in the formula:

[0177]

[0178] in:

[0179] This is the first convolution kernel, also known as the spatial convolution kernel;

[0180] B=3 represents the number of frequency bands;

[0181] F is the Fourier transform;

[0182] For frequency band masks (e.g., low-frequency mask is [1,1,0,0]);

[0183] This refers to the frequency band weight, also known as the independent weight of each frequency band.

[0184] Through inverse Fourier transform ( It generates spatial convolution kernels to achieve decoupled modeling of long-term trends (such as the annual average icing cycle) and short-term abrupt changes (such as sudden thunderstorms) in meteorological data.

[0185] The second convolution kernel can be a spatiotemporal graph convolution kernel, denoted as: (K is the size of the second convolution kernel, T is the time step, and C is the number of feature channels) to capture the spatiotemporal propagation characteristics of meteorological data.

[0186] The tower association diagram can be obtained in the following ways:

[0187] The towers of the transmission line (the same as the target transmission line mentioned above) are considered as nodes in the diagram. ), edge weight In other words, the strength of the corresponding correlation is dynamically calculated based on the correlation between tower spacing and meteorological conditions.

[0188]

[0189] in:

[0190] Let be the geographical distance between towers i and j;

[0191] This represents the correlation coefficient of meteorological data.

[0192] (2) For determining defect risk characteristics, including:

[0193] When the defect data includes a defect image, an initial defect feature map corresponding to the defect image is determined. A first mask region and a second mask region corresponding to the initial defect feature map are also determined. The first mask region is a feature region in the initial defect feature map whose feature change is greater than or equal to a feature change threshold, and the second mask region is a feature region in the initial defect feature map whose feature change is less than the feature change threshold. Based on the first and second mask regions, the defect feature map is adjusted to obtain the target defect feature map. Specifically, this is achieved through the following method:

[0194] An improved Swin Transformer V2 is used to extract multi-scale features of the defect image (such as the local texture and global structure of insulator cracks) through a sliding window mechanism, and the output feature vector is the same as the initial defect feature map mentioned above. ,in, Let H represent the three-dimensional tensor space, where H represents the height of the tensor, corresponding to the vertical resolution of the image, and W represents the width of the tensor, corresponding to the horizontal resolution of the image. This represents the number of channels in a tensor, corresponding to the feature dimension of the image.

[0195] Frequency band modulation (FBM) is introduced to dynamically adjust the feature weights of high-frequency (edge) and low-frequency (body) frequencies to adjust the initial defect feature map, as shown in the following formula:

[0196]

[0197] in:

[0198] For target defect feature map;

[0199] The feature weights corresponding to the high-frequency mask, i.e., the high-frequency (edge) feature weights, are generated by the convolutional layer and the Sigmoid function.

[0200] The feature weights corresponding to the low-frequency mask, i.e., the low-frequency (main) feature weights, are generated by the convolutional layer and the Sigmoid function.

[0201] This is a high-frequency mask used to obtain the first mask region;

[0202] This is a low-frequency mask used to obtain the second mask region.

[0203] When the defect data includes defect text (such as inspection report text), determine the semantic vector corresponding to the defect text; determine the reference risk label; and extract the risk keyword vector from the semantic vector based on the reference risk label. Specifically, this is achieved through the following methods:

[0204] Semantic encoding is performed using BERT-4, and risk keyword vectors (e.g., mapping "severe corrosion" to a risk level vector) are generated through Prompt Tuning, using the following formula:

[0205]

[0206] in:

[0207] This is a semantic embedding vector, that is, a risk keyword vector. , Represents a D-dimensional vector space;

[0208] t represents the input text;

[0209] For reference to the risk label, the value is .

[0210] Finally, based on the target defect feature map and the risk keyword vector, the defect risk features corresponding to the defect text are determined.

[0211] (3) For determining the characteristics of equipment status, including:

[0212] When equipment data (such as ledger data) includes fault time-series data, the following steps are taken: First, identify the defect events, maintenance events, and fault events corresponding to the fault time-series data. Second, determine the first event correlation between the defect events and maintenance events. Third, determine the second event correlation between the maintenance events and fault events. Fourth, based on the first and second event correlations, determine the target event correlation between the defect events and fault events. Fifth, based on the target event correlation, determine the equipment status characteristics corresponding to the equipment data, where the equipment data can be structured data. Specifically, this is achieved through the following methods:

[0213] Structured feature encoding is performed on equipment data, including equipment parameters (such as years of operation). Voltage level Normalize:

[0214]

[0215] in:

[0216] For equipment parameters;

[0217] These are the normalized device parameters;

[0218] This represents the average value of the corresponding equipment parameters;

[0219] This represents the variance of the corresponding equipment parameters.

[0220] Furthermore, temporal causal modeling was performed, and historical fault time-series data was analyzed using the Causal Transformer and a causal mask matrix. ,in, It is the number of time steps, which restricts the attention calculation of non-causal relationships and identifies the correlation between target events, that is, the causal chain of "defect type → maintenance delay → failure".

[0221] Finally, based on the correlation of target events, the device status characteristics corresponding to the device data are determined.

[0222] S3. Based on the spatiotemporal characteristics of meteorology, the characteristics of defect risks, and the characteristics of equipment status, determine the target multimodal characteristics corresponding to the target transmission line.

[0223] The multimodal feature extraction module can determine the target multimodal features corresponding to the target transmission line based on meteorological spatiotemporal characteristics, defect risk characteristics, and equipment status characteristics. Specifically, meteorological data, defect data, and equipment data are mapped to a unified feature space, and cross-modal correlation analysis is performed to ultimately obtain the target multimodal features.

[0224] S4. Based on the meteorological spatiotemporal characteristics and equipment status characteristics, determine the risk propagation matrix corresponding to the target transmission line. The risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line.

[0225] Specifically, it includes:

[0226] The three-dimensional coordinate data of the sample transmission line and its multimodal features are input into the structure recognition module of the auxiliary enhancement model to obtain the geometric structure parameters of the sample transmission line. The sample image of the sample transmission line is input into the light removal feature recognition module of the auxiliary enhancement model to obtain the anti-light interference features of the sample transmission line. The light removal feature recognition module is used to extract defect features unaffected by illumination. The sample geometric structure parameters and anti-light interference features are input into the data augmentation module of the auxiliary enhancement model to augment the sample image, resulting in multiple augmented images. Each augmented image corresponds to different visual parameters, including geometric pose and illumination parameters. Using these multiple augmented images and sample data, the initial prediction model is trained to obtain the target prediction model.

[0227] Next, the target geometric parameters corresponding to the target transmission line are determined. These parameters include the deformation index of each of the multiple towers and the tower spacing between any two towers. The deformation index represents the degree of deformation of the corresponding equipment. The target prediction model is then retrieved. This model is trained on an initial prediction model using an auxiliary enhancement model and sample data. The auxiliary enhancement model is used to augment the sample images. The sample data includes the sample images. The deformation index of each of the multiple towers, the tower spacing between any two towers, the meteorological spatiotemporal characteristics, and the equipment status characteristics are input into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line.

[0228] The spatiotemporal dynamic modeling module can be used to model the spatiotemporal evolution of transmission line risks, so as to achieve short-term early warning and long-term trend prediction.

[0229] The auxiliary enhancement model can utilize neural implicit field (NIF) to strengthen the constraints of three-dimensional structures and perform neural implicit field (NIF) risk modeling.

[0230] The three-dimensional coordinate data of the sample transmission line and the multimodal features of the sample are input into the structure recognition module in the auxiliary enhancement model to obtain the sample geometric structure parameters corresponding to the sample transmission line. This can be done in the following way:

[0231] (1) Geometric field construction:

[0232] Input the three-dimensional coordinates of the sample transmission line towers, conductors, etc. Multimodal features of samples The signed distance field (SDF) is predicted using three layers of frequency dynamic convolution (FDconv), as shown in the following formula:

[0233]

[0234] in:

[0235] This represents the distance from a point in space to the surface of the device, also known as the SDF value.

[0236] This indicates a 3-layer dynamic convolution operation.

[0237] (2) Deformation rate calculation:

[0238] SDF (Self-Definition Quantification) can quantify equipment deformation (such as tower tilt angle and conductor sag). The formula is:

[0239]

[0240] in:

[0241] DA is the deformation rate;

[0242] This is the SDF value of the device in its current state;

[0243] This is the SDF baseline value under the health condition of the device.

[0244] Among them, the geometrical parameters of the sample include the geometrical field and the deformation rate.

[0245] Similarly, the target geometric parameters corresponding to the target transmission line can also be determined by determining the sample geometric parameters, which will not be elaborated further.

[0246] The sample image corresponding to the sample transmission line is input into the light removal feature recognition module in the auxiliary enhancement model to obtain the anti-light interference features corresponding to the sample transmission line. The anti-light interference features can be represented by the appearance field. The steps are as follows:

[0247] RGB pixel values ​​of the input sample image Normal vector 2D features output by a two-dimensional (2D) feature extraction model Illumination-invariant features are generated through fully connected layers, as shown in the following formula:

[0248]

[0249] in:

[0250] Operations are performed to generate illumination-invariant features of fully connected layers;

[0251] Use the Sigmoid activation function;

[0252] The normalized RGB value (same as the anti-light interference feature mentioned above).

[0253] By separating the effects of lighting (such as shadow elimination under backlight), the robustness of defects is improved. Experiments show that the feature stability is improved by 35% in backlight scenes.

[0254] The sample's geometric structure parameters and anti-light interference features are input into the data augmentation module of the auxiliary enhancement model to augment the sample image, resulting in multiple augmented images. These can be rendered using a geometry-appearance joint rendering method, with the following steps:

[0255] The SDF predicted by the geometric field is compared with that generated by the appearance field. The process of blending and rendering device views under different lighting / poses is described by the following formula:

[0256]

[0257] in:

[0258] This represents a composite image (same as the augmented image described above).

[0259] This is a geometry-appearance joint rendering operation.

[0260] The above methods are used to augment data (generate samples of extreme scenarios such as rainstorms and icing) and expand the diversity of the training set of sample images.

[0261] The target prediction model is implemented using a spatiotemporal risk propagation model. This model inputs the deformation index of multiple towers, the tower spacing between any two towers, meteorological spatiotemporal characteristics, and equipment status characteristics into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line. This can be achieved through the following steps:

[0262] A spatiotemporal encoder based on ST-GCN and Transformer is constructed, with meteorological feature sequences as input. (Same as the above-mentioned meteorological spatiotemporal characteristics) and equipment status sequence (Similar to the equipment status characteristics mentioned above, such as the status characteristics of each tower), where, The number of time steps in the meteorological feature sequence. for Time step sequence number This represents the number of time steps in the device state sequence. for Output the risk propagation matrix based on the time step sequence number. ,in, Let N represent the three-dimensional tensor space, where N is the number of towers and T is the number of time steps.

[0263] Based on the features of the 3D structural matrix generated by NIF (similar to the target geometric parameters mentioned above, such as tower spacing and conductor tension distribution), the risk propagation weights are dynamically adjusted, as shown in the following formula:

[0264]

[0265] in:

[0266] For each element in the risk propagation matrix S, specifically representing the risk propagation weight from tower i to tower j at time step t;

[0267] The meteorological characteristics of tower i at time step t. Meteorological characteristics of pole j The correlation coefficient between them.

[0268] Let i be the distance between towers i and j;

[0269] These are the SDF values ​​predicted by NIF for tower i and tower j, respectively.

[0270] This is the maximum allowable deformation threshold for the device.

[0271] S5. Based on the target's multimodal characteristics and the risk propagation matrix, determine the target risk parameters corresponding to the target transmission line. The target risk parameters include risk assessment report, risk level, risk type, risk explanation, spatiotemporal risk heat map, etc.

[0272] Specifically, this includes: retrieving a risk association set, wherein the risk association set includes multiple risk associations corresponding to a reference transmission line, and the multiple risk associations are the associations between reference multimodal features and reference risk parameters corresponding to the reference transmission line; based on the target multimodal features, selecting target associations from the risk association set, wherein the target associations are the risk associations corresponding to reference multimodal features whose similarity index with the target multimodal features is greater than a similarity threshold among the multiple risk associations; and determining the target risk parameters corresponding to the target transmission line based on the risk propagation matrix and the target associations. This can be implemented using the RAG knowledge enhancement module, with the following steps:

[0273] Construct a knowledge graph (KG, the same as the risk association set mentioned above) corresponding to the target transmission line, which includes entities and relationships such as equipment type, defect mode, and meteorological risk (e.g., "icing → insulator flashover").

[0274] Using the FAISS vector database to index knowledge embeddings, retrieve the knowledge fragments most similar to the target multimodal features. (Related to the above objectives).

[0275]

[0276] in:

[0277] For target multimodal features;

[0278] This indicates the number of the most similar knowledge fragments retrieved;

[0279] The knowledge embedding vector, concatenated with the target multimodal features, is then input into the decision layer. for A dimensional vector space.

[0280] The aim is to enhance the interpretability and generalization ability of risk assessment by retrieving knowledge and historical cases corresponding to transmission lines.

[0281] Furthermore, through a joint optimization module, the multimodal feature extraction, spatiotemporal modeling, and knowledge enhancement processes corresponding to the aforementioned multimodal feature extraction module, spatiotemporal dynamic modeling module, and RAG knowledge enhancement module are collaboratively optimized to balance accuracy and efficiency. The collaborative optimization process through the joint optimization module includes: sharing the feature backbone and multi-task loss functions.

[0282] (1) Shared feature backbone:

[0283] The detection and risk prediction branches share the first 12 layers of FDConv parameters, and the output feature map sizes are respectively (Risk Classification) and (Spatiotemporal prediction) reduces computational redundancy by 40%;

[0284] Among them, the detection branch corresponds to the multimodal feature extraction module, and the risk prediction branch corresponds to the spatiotemporal dynamic modeling module.

[0285] (2) Multi-task loss function:

[0286] The design incorporates joint loss mechanisms including risk classification, spatiotemporal prediction, and knowledge alignment. :

[0287]

[0288] in:

[0289] , , As weight;

[0290] For risk classification loss, the corresponding multimodal feature extraction module is used;

[0291] For spatiotemporal prediction loss, corresponding to the spatiotemporal dynamic modeling module;

[0292] This represents the knowledge alignment loss, corresponding to the RAG knowledge enhancement module.

[0293] The classification loss uses Focal Loss to handle the imbalance between positive and negative samples, and the formula is as follows:

[0294]

[0295] in:

[0296] This represents the probability of the c-th class predicted by the model.

[0297] Total number of categories;

[0298] Category weights;

[0299] For focusing parameters.

[0300] Spatiotemporal risk loss is calculated based on the deviation between the predicted risk value and the actual value constrained by the mean squared error (MSE), as shown in the following formula:

[0301]

[0302] in:

[0303] The size of the spatial dimension, i.e., the number of towers in the transmission line;

[0304] This refers to the size of the time dimension, i.e., the number of time steps.

[0305] In time step At that time, the first Predicted risk value for each tower;

[0306] In time step At that time, the first The true risk value of each pole / tower.

[0307] Knowledge alignment loss ensures multimodal features through contrastive learning. With retrieval knowledge Semantic consistency, the formula is as follows:

[0308]

[0309] in:

[0310] Multimodal features used for training With retrieved knowledge The similarity between them;

[0311] sim is the cosine similarity function.

[0312] The following description is based on the steps outlined above, with specific examples.

[0313] B1. Data preprocessing.

[0314] (1) Image enhancement.

[0315] Adaptive median filtering (window size) is applied to drone inspection images. Noise is removed, and homomorphic filtering (cutoff frequency 0.3) is used to eliminate uneven illumination; the depth map is processed using bilateral filtering (standard deviation). Smooth and fill in invalid pixels;

[0316] (2) Sensor calibration.

[0317] Estimating the camera intrinsic parameter matrix based on camera calibration method The camera-robotic arm extrinsic parameter matrix is ​​obtained through hand-eye calibration.

[0318] B2. Multimodal feature extraction.

[0319] (1) Meteorological data.

[0320] Input the hourly meteorological data (wind speed, humidity, and lightning strike frequency) for the past 7 days, decompose it into three frequency band features using FDConv, and generate a spatiotemporal feature vector using ST-GCN. , for A 3D space tensor.

[0321] (2) Defect data.

[0322] Visual features were extracted from drone images using Swin Transformer V2. ,in, For defect image features, These are features of defective text.

[0323] (3) Ledger data.

[0324] Structured parameters (service life, maintenance cycle) and time-series fault records are used to generate causal features through Causal Transformer. ,in, for Tensors of dimension.

[0325] B3. Spatiotemporal dynamic modeling.

[0326] (1) Neural implicit field rendering.

[0327] Based on the 3D model of the tower and meteorological data, synthetic images under different lighting conditions are rendered using NIF to generate training samples. ,in, For synthesized images, This provides the pose information for the synthesized image.

[0328] (2) Risk propagation prediction.

[0329] ST-Transformer outputs a risk propagation matrix for the next 24 hours to identify high-risk pole clusters.

[0330] B4. Enhanced RAG knowledge.

[0331] (1) Real-time retrieval.

[0332] Input the current multimodal features By searching the knowledge graph through FAISS, relevant historical cases can be obtained, among which... As a meteorological feature, As a defect feature, This is a characteristic of the ledger.

[0333] (2) Explanation of generation.

[0334] Generate risk explanation text (e.g., "The current risk level is high because the wind speed exceeds the design value by 20% and there are 3 serious cracks in the insulator. Refer to the planned maintenance plan").

[0335] B5. Joint Reasoning and Output.

[0336] Multimodal features are output as risk classification results (levels 1-5) and spatiotemporal risk heatmaps via a shared backbone. Combined with RAG interpretation, a final report is generated, achieving a second-level response through edge computing nodes.

[0337] The above optional implementation methods can achieve at least the following beneficial effects:

[0338] (1) Frequency domain feature decoupling. Multimodal data is decomposed into different frequency bands by FDConv, and frequency-level correlation analysis of meteorological trends and equipment defects is realized for the first time (such as the slow-varying correlation between low-frequency meteorological trends and equipment aging).

[0339] (2) Spatiotemporal risk propagation modeling. ST-Transformer combines the geographical topology of transmission lines to quantify the dynamic propagation path of risks in tower groups (such as the risk cascading effect on typhoon paths);

[0340] (3) Knowledge-data dual-drive. RAG technology dynamically binds relevant knowledge of transmission lines, historical cases and real-time data to realize a closed loop of "data prediction + knowledge interpretation" and solve the black box problem of deep learning;

[0341] (4) Lightweight design. The shared backbone and model compression technology halve the computational load, allowing edge devices to run directly and meet the needs of distributed monitoring of transmission lines.

[0342] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0343] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0344] Example 2

[0345] According to embodiments of the present invention, an apparatus for implementing the above-described method for determining transmission line risk parameters is also provided. Figure 3 This is a structural block diagram of a transmission line risk parameter determination device according to an embodiment of the present invention, such as... Figure 3 As shown, the device includes: an acquisition module 302, a first determination module 304, a second determination module 306, a third determination module 308, and a fourth determination module 310. The device will be described in detail below.

[0346] The acquisition module 302 is used to acquire target multimodal data of the target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is data containing defect information of the target transmission line; the first determination module 304 is connected to the acquisition module 302, and is used to determine the meteorological spatiotemporal characteristics corresponding to the meteorological data, the defect risk characteristics corresponding to the defect data, and the equipment status characteristics corresponding to the equipment data; the second determination module 306 is connected to the first determination module 304, and is used to determine the target multimodal characteristics corresponding to the target transmission line based on the meteorological spatiotemporal characteristics, the defect risk characteristics, and the equipment status characteristics; the third determination module 308 is connected to the second determination module 306, and is used to determine the risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal characteristics and the equipment status characteristics; the fourth determination module 310 is connected to the third determination module 308, and is used to determine the target risk parameters corresponding to the target transmission line based on the target multimodal characteristics and the risk propagation matrix.

[0347] It should be noted that the above-mentioned acquisition module 302, first determination module 304, second determination module 306, third determination module 308 and fourth determination module 310 correspond to steps S102 to S110 in the method for determining transmission line risk parameters. The multiple modules and the corresponding steps are the same in terms of implementation examples and application scenarios, but are not limited to the content disclosed in the above embodiment 1.

[0348] Example 3

[0349] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor; and a memory for storing processor-executable instructions, wherein the processor is configured to execute instructions to implement the transmission line risk parameter determination method of any of the above embodiments.

[0350] Example 4

[0351] According to another aspect of the present invention, a computer-readable storage medium is also provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the transmission line risk parameter determination method described above.

[0352] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0353] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0354] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0355] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0356] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0357] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0358] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining risk parameters of transmission lines, characterized in that, include: Acquire target multimodal data of the target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is used to characterize the defects of the target transmission line; Determine the meteorological spatiotemporal characteristics corresponding to the meteorological data, the defect risk characteristics corresponding to the defect data, and the equipment status characteristics corresponding to the equipment data; Based on the meteorological spatiotemporal characteristics, the defect risk characteristics, and the equipment status characteristics, the target multimodal characteristics corresponding to the target transmission line are determined; Based on the meteorological spatiotemporal characteristics and the equipment status characteristics, a risk propagation matrix corresponding to the target transmission line is determined, wherein the risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line; Based on the target multimodal characteristics and the risk propagation matrix, the target risk parameters corresponding to the target transmission line are determined; The step of determining the target risk parameters corresponding to the target transmission line based on the target multimodal features and the risk propagation matrix includes: retrieving a risk association set, wherein the risk association set includes multiple risk association relationships corresponding to the reference transmission line, and the multiple risk association relationships are respectively the association relationships between the reference multimodal features and the reference risk parameters corresponding to the reference transmission line; filtering target association relationships from the risk association set based on the target multimodal features, wherein the target association relationship is the risk association relationship corresponding to the reference multimodal features whose similarity index with the target multimodal features is greater than a similarity threshold among the multiple risk association relationships; and determining the target risk parameters corresponding to the target transmission line based on the risk propagation matrix and the target association relationships, wherein the target risk parameters include a risk assessment report, risk level, risk type, risk explanation, and spatiotemporal risk heatmap. The step of determining the risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal characteristics and the equipment status characteristics includes: determining the target geometric parameters corresponding to the target transmission line, wherein the target geometric parameters include the deformation index corresponding to multiple towers and the tower spacing between any two towers among the multiple towers, and the corresponding deformation index represents the degree of deformation of the corresponding equipment; retrieving the target prediction model, wherein the target prediction model is obtained by training an initial prediction model based on an auxiliary enhancement model and sample data, wherein the auxiliary enhancement model is used to augment the sample images, and the sample data includes the sample images; inputting the deformation index corresponding to the multiple towers, the tower spacing between any two towers among the multiple towers, the meteorological spatiotemporal characteristics, and the equipment status characteristics into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line, wherein the elements in the risk propagation matrix represent the risk propagation weight from one tower to another at a time step.

2. The method according to claim 1, characterized in that, Determining the spatiotemporal characteristics of the meteorological data includes: When the meteorological data includes meteorological time-series data corresponding to multiple towers on the target transmission line, meteorological frequency domain data corresponding to each of the multiple towers is determined based on the meteorological time-series data corresponding to each of the multiple towers. The meteorological feature extraction model is retrieved, wherein the meteorological feature extraction model includes a first convolution kernel and a second convolution kernel. The first convolution kernel is determined based on multiple frequency band weights and frequency band masks corresponding to the multiple frequency band weights respectively. The first convolution kernel is used to extract temporal features, and the second convolution kernel is used to extract spatial features. Based on the first convolution kernel and the meteorological frequency domain data corresponding to the plurality of towers, meteorological time features corresponding to the plurality of towers are obtained respectively; Determine the pole-tower association graph, wherein the pole-tower association graph includes the nodes corresponding to the plurality of poles, the edges corresponding to the association relationship between any two poles among the plurality of poles, and the edge weights corresponding to the association strength of the corresponding association relationship; Based on the second convolution kernel, the meteorological time characteristics corresponding to the multiple towers, and the tower association diagram, the meteorological spatiotemporal characteristics corresponding to the meteorological data are determined.

3. The method according to claim 1, characterized in that, Determining the defect risk characteristics corresponding to the defect data includes: When the defect data includes a defect image, an initial defect feature map corresponding to the defect image is determined; A first mask region and a second mask region corresponding to the initial defect feature map are determined, wherein the first mask region is a feature region in the initial defect feature map whose feature change degree is greater than or equal to a feature change threshold, and the second mask region is a feature region in the initial defect feature map whose feature change degree is less than a feature change threshold. Based on the first mask region and the second mask region, the defect feature map is adjusted to obtain the target defect feature map; Based on the target defect feature map, the defect risk features corresponding to the defect image are determined.

4. The method according to claim 1, characterized in that, Determining the device status characteristics corresponding to the device data includes: When the equipment data includes fault timing data, the defect event, maintenance event, and fault event corresponding to the fault timing data are determined. Determine the first event correlation between the defect event and the maintenance event; Determine a second event correlation between the maintenance event and the failure event; Based on the first event association relationship and the second event association relationship, determine the target event association relationship between the defect event and the fault event; Based on the target event correlation, determine the device status characteristics corresponding to the device data.

5. The method according to claim 1, characterized in that, Before retrieving the target prediction model, the following steps are included: The sample three-dimensional coordinate data and sample multimodal features corresponding to the sample transmission line are input into the structure recognition module in the auxiliary enhancement model to obtain the sample geometric structure parameters corresponding to the sample transmission line. The sample image corresponding to the sample transmission line is input into the light removal feature recognition module in the auxiliary enhancement model to obtain the anti-light interference feature corresponding to the sample transmission line. The light removal feature recognition module is used to extract defect features that are not affected by light interference. The sample geometric structure parameters and the anti-light interference features are input into the data augmentation module in the auxiliary enhancement model to augment the sample image and obtain multiple augmented images. The multiple augmented images correspond to different visual parameters, including geometric pose and illumination parameters.

6. The method according to any one of claims 1 to 5, characterized in that, Determining the defect risk characteristics corresponding to the defect data includes: If the defect data includes defect text, determine the semantic vector corresponding to the defect text; Determine the reference risk label; Based on the reference risk tags, risk keyword vectors are extracted from the semantic vectors; Based on the risk keyword vector, the defect risk features corresponding to the defect text are determined.

7. A device for determining risk parameters of transmission lines, characterized in that, include: An acquisition module is used to acquire target multimodal data of a target transmission line, wherein the target multimodal data includes meteorological data, defect data, and equipment data, and the defect data is used to characterize the defects of the target transmission line; The first determining module is used to determine the meteorological spatiotemporal characteristics corresponding to the meteorological data, the defect risk characteristics corresponding to the defect data, and the equipment status characteristics corresponding to the equipment data. The second determining module is used to determine the target multimodal features corresponding to the target transmission line based on the meteorological spatiotemporal features, the defect risk features, and the equipment status features. The third determining module is used to determine the risk propagation matrix corresponding to the target transmission line based on the meteorological spatiotemporal characteristics and the equipment status characteristics, wherein the risk propagation matrix represents the risk propagation capability between any two towers among multiple towers on the target transmission line. The fourth determining module is used to determine the target risk parameters corresponding to the target transmission line based on the target multimodal characteristics and the risk propagation matrix. The fourth determining module is further configured to retrieve a risk association set, wherein the risk association set includes multiple risk associations corresponding to a reference transmission line, and the multiple risk associations are respectively the associations between reference multimodal features and reference risk parameters corresponding to the reference transmission line; based on the target multimodal features, target associations are selected from the risk association set, wherein the target associations are the risk associations corresponding to reference multimodal features whose similarity index with the target multimodal features is greater than a similarity threshold among the multiple risk associations; based on the risk propagation matrix and the target associations, target risk parameters corresponding to the target transmission line are determined, wherein the target risk parameters include a risk assessment report, risk level, risk type, risk explanation, and spatiotemporal risk heatmap; The third determining module is further configured to determine the target geometric parameters corresponding to the target transmission line, wherein the target geometric parameters include the deformation index corresponding to each of the multiple towers, and the tower spacing between any two towers among the multiple towers, and the corresponding deformation index represents the degree of deformation of the corresponding equipment; retrieve the target prediction model, wherein the target prediction model is obtained by training an initial prediction model based on an auxiliary enhancement model and sample data, wherein the auxiliary enhancement model is used to augment the sample images, and the sample data includes the sample images; input the deformation index corresponding to each of the multiple towers, the tower spacing between any two towers among the multiple towers, the meteorological spatiotemporal characteristics, and the equipment state characteristics into the target prediction model to obtain the risk propagation matrix corresponding to the target transmission line, wherein the elements in the risk propagation matrix represent the risk propagation weight from one tower to another at a time step.

8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the method for determining transmission line risk parameters as described in any one of claims 1 to 6.

Images