Electrical equipment corrosion detection method based on double-spectrum collaborative analysis and space-time prediction

Abstract

The invention is suitable for the technical field of electrical equipment corrosion detection, and provides an electrical equipment corrosion detection method based on double-spectrum collaborative analysis and space-time prediction, and the method comprises the following steps: carrying out fusion analysis on a visible light image and an infrared image of acquired electrical equipment, and outputting position information and morphological category of a corrosion region; a binocular vision system and an IMU sensor are combined to calculate three-dimensional space coordinates, main direction angles and environmental parameters of the corrosion area, and pose data are generated; inputting the pose data into a preset corrosion extension path prediction model for processing, and outputting physical characteristic parameters; fusing the dual-frequency RTK GPS positioning data, the IMU attitude data and the physical characteristic parameters to construct a space-time association data set, and inputting the space-time association data set into a preset space-time diagram neural network model to analyze a space-time evolution relationship of diagram structure modeling corrosion nodes; and outputting microscopic and macroscopic prediction results. The identification accuracy of corrosion characteristics and the corrosion detection efficiency can be improved, and the labor cost and the operation risk are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment corrosion detection, and in particular to a power equipment corrosion detection method based on dual-spectrum collaborative analysis and spatiotemporal prediction. Background Art

[0002] Power equipment is exposed to complex outdoor environments for long periods of time. Affected by environmental factors such as humidity, salt spray, acid rain, and temperature differences, it is prone to corrosion, which can lead to a decrease in mechanical strength and deterioration of electrical performance. In severe cases, this can cause safety incidents such as equipment breakage and insulation failure. Currently, corrosion detection for power equipment relies mainly on manual inspections or single-sensor detection. Single-spectrum detection is susceptible to environmental interference. For example, visible light imaging is poor at night or under strong light conditions, while infrared imaging suffers from reduced signal-to-noise ratios when the temperature fluctuates drastically, resulting in missed detection or misjudgment of corrosion features. Furthermore, it is difficult to accurately calibrate the three-dimensional spatial position of the corrosion area, making it impossible to quantify the geometric characteristics of corrosion, such as direction and area expansion trends. The lack of high-precision geographic information association makes it difficult to link corrosion data with the equipment operation and maintenance database.

[0003] In view of this, a corrosion detection method for power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction is needed. Summary of the Invention

[0004] The present application provides a method for detecting corrosion of power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction, which is used to solve the problem that existing models for corrosion detection cannot quantify geometric features and lack geographic information association.

[0005] This application provides a method for detecting corrosion of power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction, including:

[0006] Using a drone equipped with a dual-spectrum camera to synchronously collect visible light images and infrared images of power equipment, the visible light images and infrared images are fused and analyzed to output the location information and morphological classification of the corrosion area;

[0007] Based on the location information and morphological category of the corrosion area, the three-dimensional spatial coordinates, main direction angles and environmental parameters of the corrosion area are calculated by combining the binocular vision system and the IMU sensor to generate pose data;

[0008] The posture data is input into a preset corrosion extension path prediction model for processing, and physical characteristic parameters including corrosion propagation direction probability, corrosion rate and environmental coupling coefficient are output;

[0009] The dual-frequency RTK GPS positioning data, IMU attitude data and the physical characteristic parameters are integrated to construct a spatiotemporal correlation dataset, and the spatiotemporal correlation dataset is input into a preset spatiotemporal graph neural network model to analyze the spatiotemporal evolution relationship of the corrosion nodes in the graph structure modeling;

[0010] The output includes the microscopic prediction results of the level probability distribution, expansion direction angle and corrosion area within the preset time of each corrosion node; as well as the macroscopic prediction results of the corrosion risk heat map that integrates the physical diffusion path and data-driven rules.

[0011] Furthermore, the UAV equipped with a dual-spectrum camera synchronously collects visible light images and infrared images of the power equipment, performs fusion analysis on the visible light images and infrared images, and outputs the location information and morphological category of the corrosion area, including:

[0012] The texture features of the corrosion area are extracted based on the visible light image, and the thermal gradient anomalies and component infrared spectrum anomalies of the corrosion area are extracted based on the infrared image.

[0013] Dynamically adjusting the weight coefficients of the visible light image and the infrared image according to the ambient temperature and humidity parameters, and determining the location information of the corrosion area according to the weight coefficients;

[0014] Based on the texture features, thermal gradient anomalies, component infrared spectrum anomalies and location information of the corrosion area, preset corrosion morphology categories are matched, and the morphology categories include initial corrosion, local corrosion, regional corrosion and general corrosion.

[0015] Furthermore, based on the position information and morphological category of the corrosion area, the three-dimensional spatial coordinates, main direction angles and environmental parameters of the corrosion area are calculated in combination with the binocular vision system and the IMU sensor to generate the pose data, including:

[0016] A fast directional binary feature detection algorithm is used to extract corner and edge features of the corrosion area, and a fast approximate nearest neighbor matcher is used to generate matching point pairs.

[0017] The depth information of the corrosion area is calculated based on the disparity value of the matching point pair, and the three-dimensional space coordinates are determined by combining the camera intrinsic parameter matrix and the baseline distance;

[0018] The principal component analysis method is used to calculate the main direction angle, and the extended Kalman filter algorithm is used to fuse the posture data of the IMU sensor with the main direction angle to output the corrected main direction angle;

[0019] The temperature, humidity, and salt spray concentration parameters collected by environmental sensors are associated to generate a pose dataset containing three-dimensional coordinates, main direction angles, and environmental parameters.

[0020] Furthermore, the calculation of the main direction angle using the principal component analysis method includes:

[0021] Convert the pixel point set of the eroded area into a two-dimensional coordinate matrix and calculate its covariance matrix;

[0022] Perform eigenvalue decomposition on the covariance matrix, extract the eigenvector corresponding to the maximum eigenvalue as the main direction vector, and calculate the main direction angle according to the coordinate components of the main direction vector.

[0023] Furthermore, the posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, including:

[0024] Corrosion propagation direction probability P d The calculation formula is:

[0025]

[0026] Where: α, β and γ are material property weight coefficients, satisfying α + β + γ = 1, θ and θ e are the main direction angles of the corrosion area and equipment structure stress, S salt is the current salt spray concentration, H is the salt spray concentration threshold to which the material can withstand, humidity is the ambient humidity, is the critical humidity threshold.

[0027] Furthermore, the posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, and also includes:

[0028] Corrosion rate V r The calculation formula is:

[0029]

[0030] Where: k0 is the corrosion reaction frequency factor, E a is the corrosion activation energy, R and T are the gas constant and ambient temperature respectively, λ is the shape influence coefficient, A corrode and A total is the current corrosion area and the total equipment surface area.

[0031] Furthermore, the posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, and also includes:

[0032] Environmental coupling coefficient C e The calculation formula is:

[0033]

[0034] Where: w1, w2 and w3 are weight factors of salt spray concentration, humidity and temperature gradient respectively, which are calibrated by material corrosion experiments. salt is the normalized salt spray concentration, which is calculated as the ratio of the current measured salt spray concentration to the material tolerance threshold. humidity The normalized humidity is calculated as the ratio of the current humidity measurement value to the critical humidity threshold, ΔT gradient is the temperature gradient between the corrosion area and the surrounding environment.

[0035] Furthermore, the preset spatiotemporal graph neural network model includes:

[0036] The preset spatiotemporal graph neural network model includes a spatial modeling module, a temporal modeling module, and a spatiotemporal attention mechanism dedicated to power equipment, wherein:

[0037] The spatial modeling module constructs an adjacency matrix of the graph structure based on the spatial proximity of the corrosion nodes, the physical diffusion probability and the environmental coupling coefficient;

[0038] The temporal modeling module adopts a gated recurrent unit network, taking the temporal changes of the corrosion rate and environmental parameters as inputs to model the dynamic evolution of the corrosion area;

[0039] The power equipment-specific spatiotemporal attention mechanism is used to dynamically assign the importance weights of nodes and time series, focusing on high corrosion risk areas or periods of environmental mutation.

[0040] Furthermore, the dual-frequency RTK GPS positioning data, IMU attitude data and the physical characteristic parameters are integrated to construct a spatiotemporal correlation dataset, and the spatiotemporal correlation dataset is input into a preset spatiotemporal graph neural network model to analyze the spatiotemporal evolution relationship of the corrosion nodes in the graph structure modeling, including:

[0041] Standardizing the spatiotemporal correlation dataset, aligning timestamps and removing outliers;

[0042] Based on the three-dimensional coordinates, physical characteristic parameters and environmental parameters of the corrosion nodes, a spatiotemporal graph containing node characteristics and edge weights is generated;

[0043] Input the spatiotemporal graph into the spatiotemporal graph neural network model, extract the spatial correlation of nodes through the spatial modeling module, learn the corrosion evolution law through the time series modeling module, and optimize the prediction weight through the power equipment-specific spatiotemporal attention mechanism;

[0044] Decode the hidden state of the spatiotemporal graph neural network model and output micro-prediction results and macro-prediction results.

[0045] Furthermore, the optimization of prediction weights by using a spatiotemporal attention mechanism dedicated to power equipment includes:

[0046]

[0047] Where: Q, K and V are the linear mapping matrices of node features, time series features and environment features respectively, d k is the feature dimension scaling factor, λ is the power equipment scenario adjustment coefficient, ranging from 0.5 to 1.5, θ i is the main direction angle of node i, cos(θ i -θ e ) is the matching degree between the corrosion direction and the equipment stress direction, is the environmental coupling coefficient of node i.

[0048] It can be seen from the above technical solutions that this application has the following advantages:

[0049] This application significantly improves the accuracy of positioning corrosion areas of power equipment through dual-spectrum collaborative analysis, visible light and infrared images, binocular vision, RTK-GPS and IMU. The physical and data dual-driven prediction model can dynamically evolve predictions and generate risk classifications; combining environmental parameters and material properties, the spatiotemporal graph neural network is used to model the spatiotemporal expansion laws of corrosion, and output microscopic corrosion status and macroscopic risk heat maps, effectively improving detection efficiency and scientific decision-making. It not only improves the overall performance of the system, but also provides closed-loop support for the health management of equipment throughout its life cycle, effectively preventing structural failures and reducing operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The figure is a flow chart of an embodiment of a method for detecting corrosion of electric equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction in the present invention. DETAILED DESCRIPTION

[0051] The terms "first," "second," "third," "fourth," etc. (if any) in the specification and claims of the present application and in the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential sequence. It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to," and any variations thereof, are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or apparatus.

[0052] Example 1

[0053] The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, without specific limitation. The following will introduce the power equipment corrosion detection method based on dual-spectrum collaborative analysis and spatiotemporal prediction in this application from the perspective of system implementation. Figure 1 , the method provided in the embodiment of the present application includes the following steps:

[0054] S11. Using a drone equipped with a dual-spectral camera, the drone simultaneously captures visible light and infrared images of the power equipment, performs a fusion analysis of the visible light and infrared images, and outputs the location and morphological classification of the corrosion areas.

[0055] In this embodiment, the output location information and morphological categories of the corrosion area include the following:

[0056] 1. Extract the texture features of the corrosion area based on the visible light image, and extract the thermal gradient anomaly and component infrared spectrum anomaly of the corrosion area based on the infrared image;

[0057] 2. Dynamically adjust the weight coefficients of visible light images and infrared images based on ambient temperature and humidity parameters, and determine the location information of the corrosion area based on the weight coefficients;

[0058] 3. Based on texture features, thermal gradient anomalies, component infrared spectrum anomalies and location information of the corrosion area, match the preset corrosion morphology categories, which include initial corrosion, localized corrosion, regional corrosion and general corrosion.

[0059] Specifically, after simultaneously capturing visible light and infrared images of power equipment using a drone equipped with a dual-spectral camera, the team first extracts texture features of the corrosion area from the visible light image. These features include edge sharpness, illumination uniformity, and other indicators. Simultaneously, the team calculates thermal gradient anomalies and component infrared spectral anomalies from the infrared image to reflect sudden changes in local temperature and corrosion products. The fusion weights of the two spectra are dynamically adjusted based on real-time ambient temperature and humidity parameters, for example, increasing the weight of infrared data in high-temperature and high-humidity environments. A weighted averaging strategy is then used to fuse the detection frame coordinates and class probabilities of the visible and infrared images, outputting high-precision location information for the corrosion area.

[0060] The texture features, thermal gradient anomalies, component infrared spectrum anomalies and geometric parameters of the corrosion area are further combined, where the geometric parameters include area, perimeter, and main direction angle, to match the preset corrosion morphology classification standards. Here, the corrosion morphology includes at least four forms: initial corrosion, local corrosion, regional corrosion, and comprehensive corrosion, to achieve automatic classification of the corrosion state.

[0061] The above steps effectively solve the problem of missed detection of a single spectrum under complex working conditions such as strong light and haze through dual-spectrum complementarity and dynamic weight adjustment, and improve the accuracy of morphological classification based on multi-dimensional feature fusion.

[0062] S12. Based on the location information and morphological category of the corrosion area, the binocular vision system and the IMU sensor are combined to calculate the three-dimensional spatial coordinates, main direction angles, and environmental parameters of the corrosion area to generate pose data;

[0063] In this embodiment, a binocular vision system simulates the principle of human binocular parallax to achieve three-dimensional measurement. It consists of two synchronized cameras. The IMU (Inertial Measurement Unit) consists of a three-axis accelerometer and a gyroscope. It is used to measure the drone's three-axis angular velocity and linear acceleration in real time, and obtain the aircraft's pitch and roll angles through quaternion calculations. Based on the location information and morphological classification of the corrosion area, the specific steps for calculating the three-dimensional spatial coordinates, main direction angles, and environmental parameters by combining the binocular vision system and IMU sensors are as follows:

[0064] 1. A fast directional binary feature detection algorithm is used to extract corner and edge features of the corrosion area, and a fast approximate nearest neighbor matcher is used to generate matching point pairs.

[0065] The fast oriented binary feature detection algorithm (ORB) is used to extract the corner and edge features of the corrosion area from the left and right camera images of the binocular vision system. The fast approximate nearest neighbor matcher (FLANN) is used to generate high-confidence matching point pairs to solve the feature matching problem caused by the reflection or single texture of the metal surface.

[0066] 2. Calculate the depth information of the corrosion area based on the disparity value of the matching point pair, and determine the 3D space coordinates by combining the camera intrinsic parameter matrix and the baseline distance;

[0067] The initial depth value is dynamically compensated based on the temperature offset coefficient and the baseline calibration error. The calculation formula for the three-dimensional space coordinates is:

[0068]

[0069] Where: Z is the depth value, f is the focal length of the camera, B is the baseline distance of the binocular camera, d is the parallax value, ΔT is the temperature drift coefficient, δB is the baseline calibration error, and k1 and k2 are experimental calibration parameters.

[0070] 3. Calculate the principal direction angle using principal component analysis, fuse the IMU sensor's attitude data with the principal direction angle using an extended Kalman filter algorithm, and output the corrected principal direction angle.

[0071] Specifically, the principal component analysis method is used to calculate the main direction angle, including:

[0072] (1) Convert the pixel point set of the eroded area into a two-dimensional coordinate matrix and calculate its covariance matrix;

[0073] (2) Perform eigenvalue decomposition on the covariance matrix, extract the eigenvector corresponding to the maximum eigenvalue as the main direction vector, and calculate the main direction angle based on the coordinate components of the main direction vector.

[0074] The pixel point set of the eroded area is converted into a two-dimensional coordinate matrix, its covariance matrix is calculated and eigenvalue decomposition is performed, and the eigenvector corresponding to the maximum eigenvalue is extracted as the main direction vector. According to the coordinate components of the main direction vector (v x ,v y ), through the formula Calculate the principal direction angle θ. The extended Kalman filter algorithm fuses the pitch, roll, and yaw angle data from the IMU to dynamically compensate for angular deviations caused by drone flight jitter, outputting a corrected principal direction angle θ with an accuracy of ±0.5°.

[0075] 4. Associate the temperature, humidity, and salt spray concentration parameters collected by environmental sensors to generate a pose dataset containing three-dimensional coordinates, main direction angles, and environmental parameters.

[0076] The temperature, humidity and salt spray concentration parameters collected in real time by the associated environmental sensors are used to generate a pose dataset containing three-dimensional coordinates, main direction angles and environmental parameters, providing multi-dimensional input for subsequent corrosion extension path prediction.

[0077] Taking the corrosion inspection of a transmission tower as an example, the coordinates of the corrosion point calculated by binocular vision are (12.3, 5.6, 2.1), the main direction angle θ = 45°, and the angle is 44.8° after IMU correction. At the same time, the salt spray concentration is bound to 0.6 mg / m3 to form a complete pose data record.

[0078] S13. Input the posture data into a preset corrosion extension path prediction model for processing, and output physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate, and environmental coupling coefficient;

[0079] In this embodiment, the posture data obtained above is input into a pre-set corrosion extension path prediction model to calculate the corrosion propagation direction probability, corrosion rate and environmental coupling coefficient respectively. The calculation formulas are as follows:

[0080] Corrosion propagation direction probability P d The calculation formula is:

[0081]

[0082] Where: α, β and γ are material property weight coefficients, satisfying α + β + γ = 1, θ and θ eare the main direction angles of the corrosion area and equipment structure stress, S salt is the current salt spray concentration, H is the salt spray concentration threshold to which the material can withstand, humidity is the ambient humidity, is the critical humidity threshold.

[0083] Corrosion rate V r The calculation formula is:

[0084]

[0085] Where: k0 is the corrosion reaction frequency factor, E a is the corrosion activation energy, R and T are the gas constant and ambient temperature respectively, λ is the shape influence coefficient, A corrode and A total is the current corrosion area and the total equipment surface area.

[0086] Environmental coupling coefficient C e The calculation formula is:

[0087]

[0088] Where: w1, w2 and w3 are weight factors of salt spray concentration, humidity and temperature gradient respectively, which are calibrated by material corrosion experiments. salt is the normalized salt spray concentration, which is calculated as the ratio of the current measured salt spray concentration to the material tolerance threshold. humidity The normalized humidity is calculated as the ratio of the current humidity measurement value to the critical humidity threshold, ΔT gradient is the temperature gradient between the corrosion area and the surrounding environment.

[0089] S14. Integrate dual-frequency RTK GPS positioning data, IMU attitude data, and physical feature parameters to construct a spatiotemporal correlation dataset. This dataset is then fed into a pre-defined spatiotemporal graph neural network model to analyze the spatiotemporal evolution of the corrosion nodes in the graph structure modeling.

[0090] In this embodiment, the preset spatiotemporal neural network model (ST-GNN model) includes a spatial modeling module, a temporal modeling module, and a spatiotemporal attention mechanism dedicated to power equipment, wherein:

[0091] 1. The spatial modeling module constructs an adjacency matrix of the graph structure based on the spatial proximity of corrosion nodes, physical diffusion probability, and environmental coupling coefficient;

[0092] Based on the spatial proximity (Euclidean distance) of the corrosion nodes, the physical diffusion probability and environmental coupling coefficient Dynamically calculate the edge weights between nodes. The edge weight formula is:

[0093]

[0094] Where α, β, and γ are weight coefficients, calibrated experimentally. For example, in salt spray corrosion scenarios, β (the diffusion probability weight) may be significantly higher than the other coefficients. A graph convolution operation (GCN) is used to extract the spatial correlation of corrosion nodes, capturing the mutual influence of adjacent corrosion areas and the diffusion patterns driven by the environment.

[0095] 2. The time series modeling module uses a gated recurrent unit network to model the dynamic evolution of the corrosion area using the corrosion rate and the time series changes of environmental parameters as input;

[0096] The Gated Recurrent Unit (GRU) network is based on the corrosion rate V r The model uses the temporal changes of environmental parameters (temperature, humidity, and salt spray concentration) as input to model the dynamic evolution of the corrosion area over time. Through the forget gate and update gate mechanism of the GRU, it learns long-term dependencies in corrosion evolution, such as the lagged effect of a sudden increase in humidity on the corrosion rate.

[0097] 3. A spatiotemporal attention mechanism dedicated to power equipment is used to dynamically assign the importance weights of nodes and time series, focusing on areas with high corrosion risk or periods of environmental mutation.

[0098] Attention weight allocation is to dynamically adjust the importance of nodes and time sequences, focusing on the following high corrosion risk scenarios: Corrosion rate V r ≥0.2mm / month, indicating accelerated corrosion, environmental coupling coefficient C e A value of ≥0.8 reflects high salt fog or humidity environments; the node is located in stress concentration areas of the equipment structure, such as welds and bolted joints. By focusing on key nodes and time periods, the model's prediction accuracy for high-risk corrosion areas can be improved.

[0099] In this embodiment, based on the physical characteristic parameters determined in the above steps, dual-frequency RTK GPS positioning data, IMU attitude data and the physical characteristic parameters are integrated to construct a spatiotemporal correlation dataset, and the spatiotemporal evolution relationship of the corrosion nodes in the graph structure modeling is analyzed by inputting the preset spatiotemporal graph neural network model. The steps include:

[0100] 1. Standardize the spatiotemporal correlation dataset, align the timestamps, and remove outliers;

[0101] Standardization normalizes GPS positioning data, IMU attitude data, and physical characteristic parameters to eliminate dimensional differences. Timestamp alignment uses the inspection period (e.g., weekly) as a benchmark to unify the time series of multi-source data. Outlier removal filters out abnormal data based on sudden changes in corrosion area (e.g., daily increases exceeding 10%) or environmental parameter outliers (e.g., salt spray concentration exceeding a threshold).

[0102] 2. Generate a spatiotemporal graph containing node features and edge weights based on the three-dimensional coordinates, physical characteristic parameters, and environmental parameters of the corrosion nodes;

[0103] Each corrosion node contains geometric features, physical features and environmental features. The geometric features include three-dimensional coordinates, area, and main direction angle; the physical features include corrosion rate and diffusion direction probability; and the environmental features include temperature, humidity, and salt spray concentration.

[0104] 3. Input the spatiotemporal graph into the spatiotemporal graph neural network model. The spatial modeling module extracts the spatial correlation of nodes, the time series modeling module learns the corrosion evolution law, and the power equipment-specific spatiotemporal attention mechanism optimizes the prediction weight.

[0105] The spatial modeling module extracts spatial correlations between nodes through graph convolution operations on the adjacency matrix. For example, if adjacent nodes share a high environmental coupling coefficient, their edge weights are significantly increased. The temporal modeling module inputs the temporal sequence of corrosion rate and environmental parameters into the GRU network to predict future changes in corrosion area.

[0106] Using a customized attention formula:

[0107]

[0108] Where: Q, K and V are the linear mapping matrices of node features, time series features and environment features respectively, d k is the feature dimension scaling factor, λ is the power equipment scenario adjustment coefficient, ranging from 0.5 to 1.5, θ i is the main direction angle of node i, cos(θ i -θ e ) is the matching degree between the corrosion direction and the equipment stress direction, is the environmental coupling coefficient of node i.

[0109] Weight rule: If the node satisfies V r ≥0.2mm / month and cos(θ i -θ e )≥0.8, the weight is increased by 30%; if the node is located in the stress area and The weight increased by 50%.

[0110] 4. Decode the hidden state of the spatiotemporal graph neural network model and output micro-prediction results and macro-prediction results.

[0111] Micro-level prediction: Decodes the model's hidden state and outputs the probability distribution of each corrosion node's level (e.g., the probability of initial corrosion -> full corrosion), the expansion direction angle (0-360°), and the corrosion area forecast for the next 1-3 months. Macro-level prediction: Generates a corrosion risk heat map that integrates physical diffusion paths with data-driven rules, annotating high-risk areas and their maintenance priorities, such as "red areas require repair within 30 days."

[0112] Taking the detection of a high-voltage transmission tower as an example, the spatiotemporal neural network model identifies the corrosion rate V at the node coordinates A(12.3,5.6,2.1). r =0.25mm / month, environmental coupling coefficient C e = 0.9, and is located at a weld. The spatiotemporal attention mechanism increases its weight to 0.9. The model predicts that its area will increase by 60% over the next three months and labels it a "Level 1 risk," guiding operations personnel to prioritize its handling.

[0113] S15. Output includes the micro-prediction results of the level probability distribution, expansion direction angle, and corrosion area within the preset time for each corrosion node; as well as the macro-prediction results of the corrosion risk heat map that integrates the physical diffusion path and data-driven rules.

[0114] In this embodiment, the preset time refers to a certain period of time in the future. During the prediction phase of the spatiotemporal graph neural network model, the system outputs microscopic and macroscopic prediction results based on the spatiotemporal correlation dataset. The microscopic prediction results generate a level probability distribution, an expansion direction angle, and the corrosion area within the preset time for each corrosion node by decoding the hidden state of the model. The level probability distribution is based on the historical evolution of the corrosion morphology category and outputs the probability distribution of each corrosion level. For example, the probability of a node entering the full corrosion stage in the next three months is 65%. The expansion direction angle combines the main direction angle with the stress direction of the equipment structure to predict the expansion direction of corrosion within a range of ±10° and annotate it with an arrow in the three-dimensional model. The corrosion area within the preset time is calculated based on the time series prediction value of the corrosion rate, and the corrosion area growth in the future preset time (such as 1 month, 3 months) is calculated.

[0115] The macro prediction results generate a corrosion risk heat map and maintenance priority map by aggregating the spatiotemporal evolution data of all nodes. The corrosion risk heat map integrates the physical diffusion path prediction model (such as the high-probability path along the stress direction) with the data-driven law of the ST-GNN model (such as the correlation of neighboring nodes), and uses a color gradient (green → red) to mark the risk level of different areas, where the red area represents the corrosion rate V r ≥0.2mm / month and environmental coupling coefficient C eHigh-risk areas with a corrosion rate of ≥0.8. The maintenance priority map integrates micro-predictions of corrosion expansion direction, area growth rate, and the importance of equipment structures (such as load-bearing components) to classify maintenance levels by urgency, such as level 1 requiring treatment within 7 days and level 2 requiring treatment within 30 days. This level of maintenance is then overlaid onto the equipment's 3D model or geographic information map. Micro-predictions provide precise analysis of corrosion point status to guide local repair strategies. Macro-predictions, through heat maps and priority maps, provide managers with a basis for global operations and maintenance planning.

[0116] The above embodiment uses multispectral data fusion to improve the recognition accuracy of corrosion features, ensures the reliability of spatial positioning through high-precision RTKGPS and IMU sensors, and realizes dynamic prediction of corrosion development through spatiotemporal graph neural network. It not only greatly improves detection efficiency and reduces labor costs and operational risks, but also provides a scientific decision-making basis for preventive maintenance of power equipment, which has important practical value for ensuring the safe and stable operation of the power system.

[0117] It is understandable that those skilled in the art can, under the guidance of the above embodiments, combine various implementation methods in the above embodiments to obtain technical solutions of multiple implementation methods.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting corrosion of power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction, characterized in that: include: Using a drone equipped with a dual-spectrum camera to synchronously collect visible light images and infrared images of power equipment, the visible light images and infrared images are fused and analyzed to output the location information and morphological classification of the corrosion area; Based on the location information and morphological category of the corrosion area, the three-dimensional spatial coordinates, main direction angles and environmental parameters of the corrosion area are calculated by combining the binocular vision system and the IMU sensor to generate pose data; The posture data is input into a preset corrosion extension path prediction model for processing, and physical characteristic parameters including corrosion propagation direction probability, corrosion rate and environmental coupling coefficient are output; The dual-frequency RTK GPS positioning data, IMU attitude data and the physical characteristic parameters are integrated to construct a spatiotemporal correlation dataset, and the spatiotemporal correlation dataset is input into a preset spatiotemporal graph neural network model to analyze the spatiotemporal evolution relationship of the corrosion nodes in the graph structure modeling; The output includes the probability distribution of the level of each corrosion node, the expansion direction angle, and the microscopic prediction results of the corrosion area within the preset time; And the macro prediction results of the corrosion risk heat map that integrates the physical diffusion path and data-driven rules.

2. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 1 is characterized in that: The method includes synchronously collecting visible light images and infrared images of the power equipment by using a drone equipped with a dual-spectrum camera, fusing and analyzing the visible light images and infrared images, and outputting location information and morphological categories of the corrosion areas, including: The texture features of the corrosion area are extracted based on the visible light image, and the thermal gradient anomalies and component infrared spectrum anomalies of the corrosion area are extracted based on the infrared image. Dynamically adjusting the weight coefficients of the visible light image and the infrared image according to the ambient temperature and humidity parameters, and determining the location information of the corrosion area according to the weight coefficients; Based on the texture features, thermal gradient anomalies, component infrared spectrum anomalies and location information of the corrosion area, preset corrosion morphology categories are matched, and the morphology categories include initial corrosion, local corrosion, regional corrosion and general corrosion.

3. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 1 is characterized in that: The method of calculating the three-dimensional spatial coordinates, main direction angles and environmental parameters of the corrosion area based on the position information and morphological category of the corrosion area in combination with the binocular vision system and the IMU sensor to generate the pose data includes: A fast directional binary feature detection algorithm is used to extract corner and edge features of the corrosion area, and a fast approximate nearest neighbor matcher is used to generate matching point pairs. The depth information of the corrosion area is calculated based on the disparity value of the matching point pair, and the three-dimensional space coordinates are determined by combining the camera intrinsic parameter matrix and the baseline distance; The principal component analysis method is used to calculate the main direction angle, and the extended Kalman filter algorithm is used to fuse the posture data of the IMU sensor with the main direction angle to output the corrected main direction angle; The temperature, humidity, and salt spray concentration parameters collected by environmental sensors are associated to generate a pose dataset containing three-dimensional coordinates, main direction angles, and environmental parameters.

4. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 3 is characterized in that: The method of calculating the main direction angle by using the principal component analysis method includes: Convert the pixel point set of the eroded area into a two-dimensional coordinate matrix and calculate its covariance matrix; Perform eigenvalue decomposition on the covariance matrix, extract the eigenvector corresponding to the maximum eigenvalue as the main direction vector, and calculate the main direction angle according to the coordinate components of the main direction vector.

5. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 1, characterized in that: The posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, including: Corrosion propagation direction probability P d The calculation formula is: Where: α, β and γ are material property weight coefficients, satisfying α + β + γ = 1, θ and θ e are the main direction angles of the corrosion area and equipment structure stress, S salt is the current salt spray concentration, H is the salt spray concentration threshold to which the material can withstand, humidity is the ambient humidity, is the critical humidity threshold.

6. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 5 is characterized in that: The posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, and also includes: Corrosion rate V r The calculation formula is: Where: k0 is the corrosion reaction frequency factor, E a is the corrosion activation energy, R and T are the gas constant and ambient temperature respectively, λ is the shape influence coefficient, A corrode and A total is the current corrosion area and the total equipment surface area.

7. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 6, characterized in that: The posture data is input into a preset corrosion extension path prediction model for processing, and the output includes physical characteristic parameters including the probability of corrosion propagation direction, corrosion rate and environmental coupling coefficient, and also includes: Environmental coupling coefficient C e The calculation formula is: Where: w1, w2 and w3 are weight factors of salt spray concentration, humidity and temperature gradient respectively, which are calibrated by material corrosion experiments. salt is the normalized salt spray concentration, which is calculated as the ratio of the current measured salt spray concentration to the material tolerance threshold. humidity The normalized humidity is calculated as the ratio of the current humidity measurement value to the critical humidity threshold, ΔT gradient is the temperature gradient between the corrosion area and the surrounding environment.

8. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to any one of claims 1 to 7, characterized in that: The preset spatiotemporal graph neural network model includes: The preset spatiotemporal graph neural network model includes a spatial modeling module, a temporal modeling module, and a spatiotemporal attention mechanism dedicated to power equipment, wherein: The spatial modeling module constructs an adjacency matrix of the graph structure based on the spatial proximity of the corrosion nodes, the physical diffusion probability and the environmental coupling coefficient; The temporal modeling module adopts a gated recurrent unit network, taking the temporal changes of the corrosion rate and environmental parameters as inputs to model the dynamic evolution of the corrosion area; The power equipment-specific spatiotemporal attention mechanism is used to dynamically assign the importance weights of nodes and time series, focusing on high corrosion risk areas or periods of environmental mutation.

9. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 8, characterized in that: The method of fusing dual-frequency RTK GPS positioning data, IMU attitude data, and the physical characteristic parameters to construct a spatiotemporal correlation dataset, and inputting the spatiotemporal correlation dataset into a preset spatiotemporal graph neural network model to analyze the spatiotemporal evolution relationship of the corrosion nodes in the graph structure modeling includes: Standardizing the spatiotemporal correlation dataset, aligning timestamps and removing outliers; Based on the three-dimensional coordinates, physical characteristic parameters and environmental parameters of the corrosion nodes, a spatiotemporal graph containing node characteristics and edge weights is generated; Input the spatiotemporal graph into the spatiotemporal graph neural network model, extract the spatial correlation of nodes through the spatial modeling module, learn the corrosion evolution law through the time series modeling module, and optimize the prediction weight through the power equipment-specific spatiotemporal attention mechanism; Decode the hidden state of the spatiotemporal graph neural network model and output micro-prediction results and macro-prediction results.

10. The method for detecting corrosion of electric power equipment based on dual-spectrum collaborative analysis and spatiotemporal prediction according to claim 9, characterized in that: The optimization of prediction weights through a spatiotemporal attention mechanism dedicated to power equipment includes: Where: Q, K and V are the linear mapping matrices of node features, time series features and environment features respectively, d k is the feature dimension scaling factor, λ is the power equipment scenario adjustment coefficient, ranging from 0.5 to 1.5, θ i is the main direction angle of node i, cos(θ i ―θ e ) is the matching degree between the corrosion direction and the equipment stress direction, is the environmental coupling coefficient of node i.

Images