Method for predicting corrosion depth of power transmission tower member based on corrosion current monitoring

Abstract

This invention discloses a method for predicting the corrosion depth of transmission tower components based on corrosion current monitoring, belonging to the field of transmission tower operation and maintenance monitoring technology. The method includes: acquiring the initial corrosion current signal and meteorological environmental data of the transmission tower components to be monitored, extracting a macroscopic evolution benchmark, and constructing an environmental feature matrix; constructing and training a deep learning network that integrates the cumulative transferred charge conservation constraint to obtain a trained metal corrosion current evolution trend prediction model; predicting the corrosion current of the transmission tower components to be monitored and performing time-domain discrete integration to obtain the predicted charge accumulation; and converting the predicted charge accumulation into an initial predicted corrosion depth based on a target mapping fitting curve and then correcting it to obtain the target predicted corrosion depth. This invention alleviates the technical problems of overfitting to high-frequency physical noise and the inability to quantitatively equivalence due to physical differences between the sensing mechanism and the actual material degradation in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of transmission tower operation and maintenance monitoring technology, and in particular to a method for predicting the corrosion depth of transmission tower members based on corrosion current monitoring. Background Technology

[0002] As a core infrastructure of the power system, the structural safety of transmission towers directly affects the stable operation of the power grid. Tower components are exposed to the outdoor atmosphere year-round, suffering severe corrosion from industrial, marine, and humid environments. In particular, the load-bearing critical components of the towers, in addition to enduring environmental corrosion, are subjected to enormous mechanical loads over long periods.

[0003] Currently, industry assessments of the corrosion life of power transmission towers primarily rely on periodic manual inspections, drone image recognition, or the application of empirical formulas for macroscopic atmospheric corrosion. Meanwhile, in materials science research, some cutting-edge explorations are beginning to deploy environmental sensors (collecting parameters such as temperature, humidity, sulfur dioxide, and chloride ion concentrations) and inputting them into pre-trained deep learning models to predict the corrosion current of metallic materials. The core principle is to utilize the nonlinear feature extraction capabilities of neural networks to uncover the statistical patterns between environmental factors and corrosion evolution.

[0004] However, directly applying this purely data-driven material-level prediction technology to the complex and ever-changing real-world service environment of power transmission towers presents the following serious engineering and theoretical shortcomings: Existing machine learning-based atmospheric corrosion prediction algorithms typically train directly on raw sensor current data, which includes rainwater short-circuit spikes, thermal noise from the circuit itself, and electromagnetic interference. Because the amplitude of random fluctuations in the nanoampere (nA) level raw signal is often much larger than the signal reflecting the actual corrosion evolution, data-driven models are prone to overfitting to high-frequency noise when there are too many feature dimensions or limited data, leading to a sharp drop in generalization ability. This approach also obscures the true dynamic benchmarks that determine the lifespan of metal facilities, resulting in a severe lack of physical interpretability in the model.

[0005] Currently, commonly used atmospheric corrosion monitors (ACMs) are essentially macroscopic galvanocouplers composed of dissimilar metals (such as Zn / Ag). Their output, a weak short-circuit current, reflects the degree of galvanocoupler corrosion activation at a specific electrode spacing. However, the actual corrosion on the galvanized steel surface of transmission towers is predominantly large-area, uniform micro-cell corrosion. This fundamental physical difference means that the current signal directly output by the algorithm cannot accurately and quantitatively represent the actual weight loss rate of the galvanized component.

[0006] Existing corrosion life assessment methods mostly rely on predictions based on purely material-level natural exposure data or empirical formula fitting from accelerated corrosion tests conducted indoors. These models are physically limited to ideal, stress-free test conditions, completely severing the synergistic effect between environmental erosion and structural stress, thus neglecting the stress-accelerated corrosion damage effect of mechanical-electrochemical coupling under real service conditions. Transmission towers, as typical tall, flexible frame structures, bear enormous alternating wind load stress on their windward main materials and critical nodes under extreme weather conditions. The combined effect of corrosion and wind load can easily lead to local instability or even collapse of the tower's load-bearing components. This macro- and micro-stress-strain can cause the passivation film on the galvanized layer to rupture, making the actual corrosion rate of the stressed members several times higher than the theoretically measured value under stress-free conditions. This blind spot can easily lead to a serious underestimation of damage by power grid maintenance personnel, creating significant safety hazards. Summary of the Invention

[0007] To address the aforementioned technical problems in existing technologies, this invention provides a method for predicting the corrosion depth of transmission tower components based on corrosion current monitoring. The technical solution is as follows: On the one hand, a method for predicting the corrosion depth of transmission tower components based on corrosion current monitoring is provided. The method includes: acquiring initial corrosion current signals and meteorological environmental data on the metal electrode surface of the transmission tower component to be monitored, and recording the corresponding timestamp information; extracting a macroscopic evolution benchmark by performing moving average filtering on the initial corrosion current signal; constructing an environmental feature matrix by performing feature extraction and time period encoding on the meteorological environmental data and corresponding timestamp information; constructing and training a deep learning network that integrates cumulative transferred charge conservation constraints using the environmental feature matrix as input and the macroscopic evolution benchmark as a supervision label, to obtain a trained metal corrosion current evolution trend prediction model; and using the meteorological forecast sequence for the transmission tower component to be monitored for a future preset period as input, based on the trained... A metal corrosion current evolution trend prediction model is used to predict the corrosion current of the transmission tower member to be monitored, and the predicted corrosion current is obtained. The predicted corrosion current is then discretely integrated in the time domain to obtain the predicted charge accumulation. Based on the target mapping fitting curve between charge and corrosion depth, the predicted charge accumulation is converted into the initial corrosion depth under stress-free conditions. The target mapping fitting curve is a mapping fitting curve between charge and corrosion depth constructed based on the weight loss data of the natural hanging plates deployed in the same position on the transmission tower member to be monitored by an atmospheric corrosion monitor. Based on the three-dimensional finite element model of the transmission tower member to be monitored, the dynamic stress of the transmission tower member to be monitored is extracted, and the initial corrosion depth is corrected based on the correction fitting curve between dynamic stress and corrosion acceleration factor to obtain the target corrosion depth.

[0008] Optionally, the initial corrosion current signal and meteorological environmental data of the metal electrode surface of the transmission tower member to be monitored are acquired, and the corresponding timestamp information is recorded. This includes: mounting an atmospheric corrosion monitor and a meteorological sensor at the target monitoring location of the transmission tower member to be monitored; acquiring the initial corrosion current signal of the metal electrode surface of the transmission tower member to be monitored at a preset sampling frequency based on the atmospheric corrosion monitor; and synchronously collecting the corresponding meteorological environmental data based on the meteorological sensor and recording the corresponding timestamp information.

[0009] Optionally, the macroscopic evolution benchmark includes: ; In the formula, This refers to the initial corrosion current signal. Let W be the macroscopic evolution benchmark, W represent the time window length, t represent the sampling time, and i represent the sampling point index.

[0010] Optionally, feature extraction and time-period encoding are performed on the meteorological environmental data and corresponding timestamp information to construct an environmental feature matrix, including: coupling and multiplying the current local ambient temperature of the transmission tower member to be monitored with the damping time within a preset monitoring period to construct a damp heat index feature; constructing a cumulative damping damage memory feature based on the cumulative duration of the relative humidity of the transmission tower member to be monitored exceeding a preset humidity threshold within a historical monitoring period; constructing a transient environmental factor based on the ambient temperature, relative humidity, and dew point difference derived from the ambient temperature and relative humidity of the transmission tower member to be monitored; periodically encoding the time scale of the meteorological environmental data based on the timestamp information to construct a time-period encoding; the time-period encoding includes time sine encoding and time cosine encoding; constructing a damping time feature based on the damping duration of the transmission tower member to be monitored within the preset monitoring period; and constructing an environmental feature matrix based on the damp heat index feature, the cumulative damping damage memory feature, the transient environmental factor, the time-period encoding, and the damping time feature.

[0011] Optionally, the deep learning network integrating the cumulative transferred charge conservation constraint includes: a multi-scale local feature embedding layer, a long-range physical evolution coding layer, and a fully connected linear mapping layer; wherein, the multi-scale local feature embedding layer includes a one-dimensional convolutional neural network; the long-range physical evolution coding layer includes a Transformer encoder incorporating a time decay mask matrix; and the loss function of the deep learning network integrating the cumulative transferred charge conservation constraint includes a joint objective function integrating SmoothL1 loss, physical area conservation loss, and total variation smoothing penalty.

[0012] Optionally, the method further includes: acquiring weight loss data of natural hanging plates deployed in the same location as the atmospheric corrosion monitor on the transmission tower member to be monitored; and based on the weight loss data, constructing a target mapping fitting curve for the conversion of the cumulative transferred charge of the atmospheric corrosion monitor to the initial corrosion depth under stress-free conditions.

[0013] Optionally, the dynamic stress of the transmission tower member to be monitored is extracted based on the three-dimensional finite element model of the member, and the initial corrosion depth is corrected based on the modified fitting curve between the dynamic stress and the corrosion acceleration factor to obtain the target corrosion depth. This includes: constructing a three-dimensional finite element model based on the topology and dimensional parameters of the transmission tower member; converting historical wind speed and direction time history data of the area where the transmission tower member is located into dynamic wind pressure load according to the structural wind load calculation specifications; inputting the dynamic wind pressure load into the three-dimensional finite element model to extract the dynamic stress of the transmission tower member; obtaining a first corrosion rate under different stress conditions and a second corrosion rate under no stress conditions based on accelerated experiments of metals under the same environment under different stress conditions; calculating the corrosion acceleration factor under different stress conditions based on the first corrosion rate and the second corrosion rate; performing nonlinear curve fitting on the corrosion acceleration factor under different stress conditions to obtain a modified fitting curve between the dynamic stress and the corrosion acceleration factor; and correcting the initial corrosion depth based on the modified fitting curve and the dynamic stress of the transmission tower member to obtain the target corrosion depth.

[0014] On the other hand, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.

[0015] On the other hand, a computer-readable storage medium is also provided, wherein program code is stored in the computer-readable storage medium, and the program code can be called by a processor to execute the method provided in the embodiments of the present invention.

[0016] This invention provides a method for predicting the corrosion depth of transmission tower members based on corrosion current monitoring. It constructs a target transformation mechanism based on integral area equivalence to solve the problem of overfitting due to high-frequency noise; combines real corrosion data of hanging plates to eliminate the physical difference between the sensed current and the actual uniform corrosion; and establishes a stress acceleration correction model for transmission tower members to solve the safety hazard of traditional static models seriously underestimating the damage of key load-bearing components of the tower under extreme weather conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for predicting the corrosion depth of transmission tower members based on corrosion current monitoring, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a deep learning network that integrates the cumulative transfer charge conservation constraint provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the initial corrosion power circuit signal with the macroscopic evolution benchmark provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of full-cycle cumulative deep evolution consistency verification provided in an embodiment of the present invention; Figure 5 This is a schematic diagram comparing the method provided in the embodiments of the present invention with a traditional method. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Figure 1 This is a flowchart illustrating a method for predicting the corrosion depth of transmission tower members based on corrosion current monitoring, according to an embodiment of the present invention. Figure 1 As shown, the method specifically includes the following steps: Step S102: Obtain the initial corrosion current signal and meteorological environment data of the metal electrode surface of the transmission tower member to be monitored, and record the corresponding timestamp information.

[0023] Step S104: The initial corrosion current signal is extracted by moving average filtering to obtain the macroscopic evolution benchmark. The meteorological environment data and the corresponding timestamp information are used to extract features and encode time periodicity to construct an environmental feature matrix.

[0024] Step S106: Using the environmental feature matrix as input and the macroscopic evolution benchmark as supervision label, a deep learning network integrating the cumulative transferred charge conservation constraint is constructed and trained to obtain a predicted model of metal corrosion current evolution trend after training.

[0025] Step S108: Using the meteorological forecast sequence of the future preset period of the transmission tower member to be monitored as input, the corrosion current of the transmission tower member to be monitored is predicted based on the metal corrosion current evolution trend prediction model after training, and the predicted corrosion current is obtained.

[0026] Step S110: Perform time-domain discrete integration on the predicted corrosion current to obtain the predicted charge accumulation.

[0027] Step S112: Based on the target mapping fitting curve between charge and corrosion depth, the predicted charge accumulation is converted into the initial corrosion depth under stress-free conditions; the target mapping fitting curve is the mapping fitting curve between charge and corrosion depth constructed based on the weight loss data of the natural hanging plates deployed in the same position on the transmission tower member to be monitored by the atmospheric corrosion monitor.

[0028] Step S114: Extract the dynamic stress of the transmission tower member to be monitored based on the three-dimensional finite element model of the member, and correct the initial corrosion depth based on the corrected fitting curve between the dynamic stress and the corrosion acceleration factor to obtain the target corrosion depth.

[0029] Specifically, step S102 includes the following steps: Step S1021: Install an atmospheric corrosion monitor and a meteorological sensor on the target monitoring part of the transmission tower component to be monitored; Step S1022: Based on the atmospheric corrosion monitor, acquire the initial corrosion current signal on the surface of the metal electrode of the transmission tower member to be monitored at a preset sampling frequency; for example, the preset sampling frequency is once every 10 minutes. Step S1023: Based on the meteorological sensor, synchronously collect the corresponding meteorological environment data and record the corresponding timestamp information.

[0030] Specifically, a time window length of W is set (e.g., 24 hours, i.e., 144 sampling points) for the initial corrosion current signal. A smoothed macroscopic evolution baseline is obtained by performing a moving average filter. :

[0031] In the formula, t represents the sampling time, and i represents the sampling point index. This step effectively filters out non-stationary components caused by instantaneous weather changes and circuit thermal noise.

[0032] Proof of the area equivalence of the moving average operator: To filter out transient environmental noise, this invention modifies the initial corrosion current signal. Macroscopic evolution benchmark is extracted using a moving average filter. .

[0033] Let the time window length be ,but:

[0034] Proof: Within the evaluation period N (N is much larger than M), for the smoothed sequence Perform discrete integration over the entire period:

[0035] By changing the order of summation, it can be seen that for any original data point in the middle of the sequence (i.e., W≤k≤NW), The number of times it is included in the summation term during the movement of the sliding window is exactly 10. .

[0036] Therefore, in the summation formula, the weight coefficient of this point is: .

[0037] Therefore, we can deduce that:

[0038] in, This represents the boundary residual term affected by the beginning and end of the sequence. (When monitoring duration...) Much larger than the filter window hour, The boundary residuals account for a negligible proportion in the calculation of the total charge.

[0039] Conclusion: The moving average method filters out high-frequency random fluctuations while strictly preserving the total charge transferred during the electrochemical reaction, achieving a balance between noise reduction and physical fidelity.

[0040] Based on the mathematical proof above, the integral of the smooth evolution benchmark strictly preserves the total transferred charge. However, due to the scale difference between macroscopic electrocouple sensing and tower micro-cell corrosion, this virtual charge needs to be further output to the subsequent multiphysics coupling correction module for secondary physical calibration through actual plate weight loss.

[0041] Specifically, step S104, which involves feature extraction and time-period encoding of meteorological environmental data and corresponding timestamp information to construct an environmental feature matrix, includes: Step S1041: The current local ambient temperature of the transmission tower member to be monitored is coupled and multiplied with the moisture time within a preset monitoring period to construct a damp heat index feature. For example, the preset monitoring period is the most recent 24 hours.

[0042] In the atmospheric corrosion mechanism of metals, the duration of the surface liquid film determines the persistence of the micro-battery reaction, while the ambient temperature directly determines the kinetic rate and activation energy of the electrochemical reaction. This invention innovatively introduces the damp heat index feature to characterize the nonlinear synergistic destructive force of heat and humidity, enabling deep learning models to keenly capture and effectively amplify corrosion acceleration abrupt signals under high temperature and high humidity conditions.

[0043] Step S1042: Based on the cumulative duration of the relative humidity of the transmission tower members under monitoring exceeding the preset humidity threshold during the historical monitoring period, construct the cumulative moisture damage memory feature.

[0044] Electrochemical corrosion of metals occurs only when a liquid film is present on the surface. This invention constructs a cumulative moisture damage memory feature by calculating the cumulative duration for which the relative humidity of the environment exceeds a preset threshold (e.g., 80%) over a historical period. This feature simulates the historical damage accumulation memory that the sensor lacks under specific conditions, enabling the model to perceive the current electrochemical fatigue level of the tower.

[0045] Step S1043: Based on the ambient temperature, relative humidity, and dew point difference derived from the ambient temperature and relative humidity of the transmission tower members to be monitored, a transient environmental factor is constructed. This feature is used to characterize the kinetic intensity of corrosion at the current moment.

[0046] Step S1044: Based on the timestamp information, the time scale of the meteorological and environmental data is periodically encoded to construct a time periodic code. The time periodic code includes time sine coding and time cosine coding. Specifically, the first step is: timestamp parsing and continuous floating-point conversion: Timestamp information is obtained from the raw multi-source monitoring data, and the hour and minute values ​​corresponding to each sampling point are extracted. To eliminate the step error caused by discrete time, the minute values ​​are converted into the decimal part of the hour, constructing a continuous floating-point time variable. The calculation formula is as follows:

[0047] Step 2: Two-dimensional physical cycle mapping based on sine and cosine functions: Taking 24 hours as a single complete diurnal physical cycle of meteorological and environmental changes, the above one-dimensional continuous time variable is mapped using sine and cosine functions. Mapped to two-dimensional periodic coding features (including time sine coding) Cosine coding of time (Two feature dimensions). The specific mathematical mapping formula is:

[0048] In traditional linear time coding, time series experience abrupt numerical drops (from near 24 to 0) when crossing the midnight boundary (i.e., from 23:59 to 00:00 the next day). These non-physical numerical abrupt changes easily lead to false gradient fluctuations and misjudgments in the network during long-term extrapolation. The sine and cosine dual-channel periodic coding method introduced in this invention maps one-dimensional linear time sequentially onto the unit circle of a two-dimensional coordinate system, eliminating the numerical discontinuity at the zero-point time boundary. This allows the model to perfectly and smoothly capture the nonlinear evolution rhythm of metal micro-battery corrosion driven by the diurnal cycle.

[0049] Step S1045: Based on the duration of moisture absorption of the transmission tower components to be monitored within a preset monitoring time period, construct moisture absorption time characteristics. For example, statistically analyze the duration of moisture absorption in the 24 hours prior to the current moment to characterize the abundance of surface liquid film caused by recent weather conditions, helping the model accurately capture the current activation state of the micro-batteries.

[0050] Step S1046: Construct an environmental feature matrix based on the characteristics of the damp heat index, the cumulative damp damage memory characteristics, transient environmental factors, time period coding, and damp time characteristics.

[0051] Through the above steps, this invention constructs an 8-dimensional feature matrix that includes damp heat index features, cumulative moisture damage memory features, ambient temperature, relative humidity, dew point difference features, time sine coding, time cosine coding, and moisture time features, ensuring that the system still has complete evolutionary deduction capabilities even when only a few environmental features are present.

[0052] Figure 2 This is a schematic diagram of a deep learning network that incorporates cumulative transferred charge conservation constraints according to an embodiment of the present invention. Figure 2 As shown, the deep learning network integrating the cumulative transferred charge conservation constraint includes: a multi-scale local feature embedding layer, a long-range physical evolution coding layer, and a fully connected linear mapping layer; among which, (1) The multi-scale local feature embedding layer includes a one-dimensional convolutional neural network.

[0053] Specifically, the input of the deep learning network receives a constructed 8-dimensional environmental feature matrix. ,in For batch size, For the historical sliding window length, For feature dimensions.

[0054] 1D-CNN Architecture Design: The multi-scale local feature embedding layer employs a one-dimensional convolutional neural network (1D-CNN), performing a sliding scan along the time axis with a preset stride and convolutional kernel. The convolutional operation automatically extracts local spatial features from meteorological sequences, such as capturing key transient signals that trigger condensation on metal surfaces, like sudden temperature drops or humidity spikes. Through convolutional mapping, the original low-dimensional features are projected into a high-dimensional latent space, achieving a preliminary characterization of complex physical triggering mechanisms.

[0055] (2) The long-range physical evolution coding layer includes a Transformer encoder that introduces a time decay mask matrix.

[0056] Specifically, the extracted local features are processed by position encoding and then fed into the Transformer encoder.

[0057] Multi-head self-attention mechanism: The module achieves cross-time domain information aggregation by calculating the correlation weight matrix of each time point within the sequence.

[0058] Mechanistic significance: Addressing the state-dependent nature of corrosion processes, the self-attention mechanism can identify the contribution weight of historical key meteorological events to the current cumulative depth increment. This mechanism simulates the physical memory of metal facilities, effectively capturing the long-range dynamic causal relationship between environmental factors and corrosion damage accumulation.

[0059] Residual connections and layer normalization: Residual connections are introduced after the attention layer, supplemented by layer normalization operations, to ensure that gradient propagation is smooth when the model processes cumulative deep data with long-term monotonically increasing characteristics, effectively preventing gradient vanishing or diverging during deep network training.

[0060] This invention introduces a time decay mask matrix inside the Transformer encoder. Since the memory of early meteorological conditions caused by electrochemical corrosion decays over time, the time decay mask matrix dynamically penalizes the attention weights of too far back in historical time steps through a negative exponential decay function. This avoids prediction distortion caused by the attention mechanism over-focusing on distant irrelevant meteorological changes and improves the local smoothness of time series features.

[0061] (3) Fully connected linear mapping layer: Unlike the autoregressive mode generated by point-by-point iteration in traditional recurrent neural networks, this invention adopts a direct multi-step prediction architecture.

[0062] Feature aggregation: The encoder's output tensor extracts the hidden state features of the last time step, which incorporates the macroscopic evolution of the entire input sequence.

[0063] Linear mapping decoding: Through a fully connected linear mapping layer and the ReLU nonlinear activation function, the macroscopic corrosion current baseline sequence (i.e. the predicted smooth current evolution trend) for a future preset prediction window (such as 24 hours) is directly obtained.

[0064] This invention avoids the snowball effect of errors generated during iterative prediction by using direct mapping design, significantly improves the robustness of long-period extrapolation prediction, and ensures that the output curve strictly conforms to the physical evolution trajectory of metal corrosion damage.

[0065] (4) The loss function of the deep learning network that integrates the cumulative transfer charge conservation constraint includes a joint objective function that integrates SmoothL1 loss, physical area conservation loss and total variation smoothing penalty.

[0066] Specifically, the joint objective function includes:

[0067] In the formula, L SmoothL1 The loss is SmoothL1, where y represents the true macroscopic corrosion evolution baseline sequence. This represents the predicted corrosion current sequence output by the deep learning network, where t and t-1 represent consecutive time steps within the prediction time window. The area conservation penalty weight is used to control the weight given to physical laws in the total error. A larger weight forces the model to adhere more strictly to the conservation of total charge integral. This represents the smoothing penalty weight, used to control the smoothness of the prediction curve. It allows for the existence of real-physics spikes but severely suppresses meaningless sawtooth oscillations.

[0068] Furthermore, during the network backpropagation parameter update phase, gradient-trimmed explosion-proof valves are configured.

[0069] Specifically, the physical area conservation loss acts as a macroscopic integral constraint of Faraday's law, forcing the model to conserve the "cumulative total corrosion charge" within the prediction window; the total variation smoothing penalty allows for the existence of real physical spikes, but severely punishes high-frequency sawtooth oscillations that have no physical meaning. This mechanism completely solves the problem of exponential error divergence under long-period offline extrapolation.

[0070] Preferably, in step S108, the input to the trained metal corrosion current evolution trend prediction model is a weather forecast sequence for the future preset period of the transmission tower member to be monitored, which is composed of the following two parts: (1) Short-term forecast area (e.g., within the next seven days): Data source: Dynamically retrieves real-time numerical weather forecast data issued by national or regional meteorological bureaus.

[0071] Function: To capture upcoming real extreme weather, the model outputs transient corrosion current based on this, which is used for short-term high-risk early warning of the power grid.

[0072] (2) Long-range statistical extension area (e.g., from the 8th to the 30th day in the future): Data source: retrieve the standard meteorological sequence for the corresponding month from the historical typical meteorological year database of the meteorological zone where the tower is located.

[0073] Function: Provides highly statistically significant meteorological data. By statistically analyzing meteorological data, it obtains extended values ​​of meteorological data for a future period as model input, ensuring that the data conforms to local climate characteristics.

[0074] Specifically, in step S110, based on the environmental and state characteristics over a historical period, the total cumulative charge transfer amount for the next complete sample collection cycle (e.g., the next month) is obtained by integrating the data through multiple step-by-step sliding splicing within a short window. This ensures that the total macroscopic charge integral is completely consistent with the actual physical weightlessness data measured by the sample collection within the same cycle on the same time scale.

[0075] Specifically, the method provided in this embodiment of the invention further includes: Obtain weight loss data of natural hanging plates deployed at the same location as the atmospheric corrosion monitor on the transmission tower pole to be monitored; specifically, deploy several standard galvanized steel hanging plates at the same location of the monitoring point, periodically collect them and perform acid washing and weighing to obtain the actual mass loss (weight loss) data under outdoor natural exposure.

[0076] Then, based on the weightlessness data, a target mapping fitting curve is constructed to transform the cumulative transferred charge of the atmospheric corrosion monitor into the initial corrosion depth under stress-free conditions.

[0077] Specifically, data acquisition is aligned with time: an alignment period is established, assuming the physical sampling period of the outdoor standard mounting patch is... (For example, T=30 days). Extract X-axis data: Based on the real historical meteorological data from the ACM instrument for these 30 days, perform time-domain integration on the current for these 30 days to obtain the predicted cumulative transferred charge for this month. Extract Y-axis data (physical true value): Remove the naturally exposed plates from the same location for these 30 days, perform standard pickling, rust removal, and weighing, calculate the actual mass loss, and convert it into the physical corrosion depth under stress-free conditions.

[0078] Establishing the fitting curve: As the monitoring time progresses (e.g., after several months or quarters), multiple monthly or quarterly data pairs can be accumulated. Based on these scattered data, a target mapping fitting curve between charge and corrosion depth is constructed using polynomial fitting algorithms such as the least squares method.

[0079] Engineering significance: Once this curve is fitted, you only need to convert the current calculated by the prediction model into the total charge each month, and then substitute the total charge into this equation to calculate how many millimeters the tower corroded under natural conditions in that month.

[0080] Specifically, the initial corrosion depth calculation method in step S112 is as follows: by substituting the predicted charge accumulation into the target mapping fitting curve, the initial corrosion depth under stress-free conditions corresponding to the predicted charge accumulation can be obtained.

[0081] Specifically, step S114 further includes the following steps: Step S1141: Based on the topology and dimensional parameters of the transmission tower members to be monitored, construct a three-dimensional finite element model; Step S1142: Convert the historical wind speed and direction time history data of the area where the transmission tower member to be monitored is located into dynamic wind pressure load according to the structural wind load calculation specification. Step S1143: Input the dynamic wind pressure load into the three-dimensional finite element model and extract the dynamic stress of the transmission tower members to be monitored.

[0082] Step S1144: Based on accelerated experiments of metals under different stress conditions in the same environment, obtain the first corrosion rate under different stress conditions and the second corrosion rate under no stress conditions. Step S1145: Based on the first corrosion rate and the second corrosion rate, calculate the corrosion acceleration factor under different stress conditions. Step S1146: Nonlinear curve fitting is performed on the corrosion acceleration factor under different stress conditions to obtain the corrected fitting curve between dynamic stress and corrosion acceleration factor. Step S1147: Based on the corrected fitting curve and the dynamic stress of the transmission tower member to be monitored, the initial corrosion depth is corrected to obtain the target corrosion depth.

[0083] Preferably, in this embodiment of the invention, a three-dimensional finite-state model of the transmission tower member to be monitored is constructed using Ansys. Specifically: (1) Three-dimensional geometric modeling and material property assignment: Import the design drawings and topological dimensions of the transmission tower components to be monitored, and build a high-fidelity 3D geometric model in ANSYS simulation software. A three-tower, four-line structure is ultimately adopted. Based on the actual materials of each component of the tower (such as Q235, Q345 angle steel or steel pipe), assign corresponding physical property parameters to the model, including but not limited to the material's elastic modulus, Poisson's ratio, and material density.

[0084] (2) Mesh generation and boundary condition constraints: A three-dimensional finite element model was constructed by discretizing the three-dimensional geometric model using finite element elements suitable for truss structures. At the base nodes of the tower model, a fixed constraint with full degrees of freedom was applied to simulate the anchorage state of the tower to the real foundation.

[0085] (3) Wind load time history conversion: Historical typical wind speed and direction time history data of the terrain area where the transmission tower components to be monitored are located are simulated in MATLAB. According to the structural wind load calculation specifications, the wind speed time history is converted into dynamic wind pressure loads acting on the windward nodes of each node of the tower, and then imported into Ansys.

[0086] (4) Transient solution and target nodal stress extraction: Initiate transient structural analysis in the ANSYS solver. After the solution converges, accurately locate the "target load-bearing member" and nodes prone to stress concentration in the model that correspond to the actual assessment prediction. Extract the time series of local dynamic principal stresses or equivalent stresses of the target member under the set wind load time history, and calculate its representative effective stress value (such as peak stress or cyclic stress amplitude), which will serve as the mechanical input parameters for subsequent calls to correct the fitted curve.

[0087] In reality, the transmission tower components under monitoring bear enormous wind loads and their own weight, and are subjected to tensile or alternating stress. This can tear the passivation film on the metal surface, leading to a sharp acceleration of corrosion. The corrected fitting curve provided in this embodiment of the invention is specifically designed to correct the corrosion depth under stress-free conditions in light of the above-mentioned circumstances. Specifically, the process of constructing the corrected fitting curve between dynamic stress and the corrosion acceleration factor is as follows: (1) Data acquisition: Conduct accelerated metal stress test under the same environment: In the same environment, use a strain tensile testing machine to simulate the corrosion process of galvanized steel material used in the transmission tower members to be monitored under different tensile stress levels.

[0088] Extract X-axis data: Set different dynamic or static stress levels For example, 10%, 30%, 50%, 70% of the yield strength, etc. Extract Y-axis data: Measure the corrosion rate at each stress level and divide it by the corrosion rate under no-stress conditions. The resulting ratio is the corrosion acceleration factor. .

[0089] (2) Establishment of the fitting curve: based on the scatter pairs obtained from the experiment Nonlinear curve fitting (e.g., exponential or quadratic polynomial fitting) is performed to obtain a corrected fitting curve between dynamic stress and corrosion acceleration factor.

[0090] Engineering significance: In practical applications, by simply using the ANSYS three-dimensional finite element model of the transmission tower member to be monitored, the local stress of a certain member under the current wind speed can be found, and by substituting it into this curve, it can be immediately known how much faster its corrosion rate is than usual.

[0091] Specifically, the initial corrosion depth is corrected based on the modified fitting curve and the dynamic stress of the transmission tower member to be monitored, to obtain the target corrosion depth. This includes: substituting the dynamic stress of the transmission tower member to be monitored into the modified fitting curve to obtain the corresponding corrosion acceleration factor; then multiplying the corrosion acceleration factor with the initial corrosion depth to obtain the target corrosion depth. The calculation formula is as follows:

[0092] In the formula, D actual Indicates the target corrosion depth. D free Indicates the initial corrosion depth. K stress Indicates the corrosion accelerating factor. This represents dynamic stress.

[0093] This invention realistically reproduces the stress-accelerated corrosion effect caused by stress on rods under field working conditions.

[0094] As described above, the embodiments of the present invention provide a method for predicting the corrosion depth of transmission tower members based on corrosion current monitoring, which has the following technical advantages compared with the prior art: (1) Break through the limitations of original data-driven approach and realize the filtering of high-frequency physical noise and the reconstruction of long-cycle evolution state.

[0095] Figure 3 This is a schematic diagram comparing the initial corrosion power circuit signal with the macroscopic evolution benchmark according to an embodiment of the present invention. Figure 4 This is a schematic diagram of full-cycle cumulative deep evolution consistency verification provided by an embodiment of the present invention. For example... Figure 3 and Figure 4 As shown, directly predicting transient corrosion currents containing a large amount of high-frequency physical noise can easily lead to model non-convergence or divergence. This invention first performs a sliding filter on the original transient current to remove environmental thermal noise, and mathematically proves that the time-series charge integral area before and after the smoothing filter is essentially equivalent. Based on this, this invention innovatively introduces a PINN area conservation constraint based on Faraday's law into the network. This mechanism forces the model to strictly adhere to the cumulative transferred charge conservation under long-period derivation while discarding high-frequency transient noise, completely overcoming the pain point of exponentially amplifying errors over time in traditional derivations, and endowing the black-box algorithm with extremely strong physical interpretability and long-term stability.

[0096] (2) Significantly improved the prediction accuracy and stability of long-period nonlinear corrosion evolution.

[0097] To address the challenges of long-range memory and environmental hysteresis in the atmospheric corrosion process of metals, this solution designs a dedicated CorrTransformer hybrid neural network architecture. Specifically, the model utilizes 1D-CNN to extract local micro-meteorological transient features, and employs a Transformer encoder with a time decay mask and a multi-head self-attention mechanism to establish long-range physical aging memory. Simultaneously, a direct multi-step mapping head is used instead of the traditional point-by-point autoregressive iterative prediction, effectively avoiding the snowball effect of errors in long-range extrapolation. Figure 5 This is a schematic diagram comparing the method provided by the embodiments of the present invention with a conventional method. Figure 5 As shown, compared with traditional machine learning algorithms such as random forests, this invention has outstanding advantages in prediction accuracy and robustness when dealing with complex micro-meteorological coupled inputs.

[0098] (3) Establish a multi-physics field correction closed loop that integrates "evolutionary trend deduction - in-situ mapping of hanging plates - structural stress correction".

[0099] This invention innovatively introduces in-situ mapping of outdoor naturally exposed mounting plates, eliminating corrosion data errors between macroscopic electrocouple sensing and microscopic uniform corrosion. Furthermore, by combining finite element analysis of transmission towers with laboratory datasets, dynamic principal stresses are extracted and a corrosion acceleration factor is assigned. This mechanism accurately reproduces the accelerated damage effect of passivation film rupture caused by "mechanical-electrochemical" coupling under natural conditions, solving the safety hazard of traditional static models severely underestimating damage to the load-bearing components of the tower, and achieving true engineering-level lifespan early warning.

[0100] The present invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.

[0101] The present invention also provides a computer-readable storage medium storing program code, which can be called by a processor to execute the method provided in the embodiments of the present invention.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0105] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, 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.

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a 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 this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting corrosion depth of transmission tower members based on corrosion current monitoring, characterized in that, The method includes: Acquire the initial corrosion current signal and meteorological environmental data of the metal electrode surface of the transmission tower member to be monitored, and record the corresponding timestamp information; The initial corrosion current signal is filtered by moving average to obtain a macroscopic evolution benchmark. The meteorological environment data and the corresponding timestamp information are used to extract features and encode time periodicity to construct an environmental feature matrix. Using the environmental feature matrix as input and the macroscopic evolution benchmark as supervision label, a deep learning network integrating the cumulative transferred charge conservation constraint is constructed and trained to obtain a metal corrosion current evolution trend prediction model after training. Using the meteorological forecast sequence of the future preset period of the transmission tower member to be monitored as input, the corrosion current of the transmission tower member to be monitored is predicted based on the metal corrosion current evolution trend prediction model after training, and the predicted corrosion current is obtained. The predicted corrosion current is discretely integrated in the time domain to obtain the predicted charge accumulation. Based on the target mapping fitting curve between charge and corrosion depth, the predicted charge accumulation is converted into the initial corrosion depth under stress-free conditions; the target mapping fitting curve is a mapping fitting curve between charge and corrosion depth constructed based on the weight loss data of the natural hanging plates deployed in the same position on the transmission tower members to be monitored by the atmospheric corrosion monitor. The dynamic stress of the transmission tower member to be monitored is extracted based on the three-dimensional finite element model of the member, and the initial corrosion depth is corrected based on the modified fitting curve between the dynamic stress and the corrosion acceleration factor to obtain the target corrosion depth.

2. The method according to claim 1, characterized in that, Acquire the initial corrosion current signal and meteorological environmental data of the metal electrode surface of the transmission tower components to be monitored, and record the corresponding timestamp information, including: Atmospheric corrosion monitors and meteorological sensors are mounted on the target monitoring locations of the transmission tower components to be monitored. Based on the atmospheric corrosion monitor, the initial corrosion current signal of the metal electrode surface of the transmission tower member to be monitored is obtained at a preset sampling frequency. Based on the meteorological sensor, corresponding meteorological environmental data are collected synchronously, and corresponding timestamp information is recorded.

3. The method according to claim 1, characterized in that, The macroscopic evolution benchmark includes: ； In the formula, This refers to the initial corrosion current signal. Let W be the macroscopic evolution benchmark, W represent the time window length, t represent the sampling time, and i represent the sampling point index.

4. The method according to claim 1, characterized in that, The meteorological and environmental data and corresponding timestamp information are used to extract features and encode time periods to construct an environmental feature matrix, including: The current local ambient temperature of the transmission tower member to be monitored is coupled and multiplied with the moisture time within a preset monitoring period to construct a damp heat index feature; Based on the cumulative duration of the relative humidity of the monitored transmission tower members exceeding a preset humidity threshold during the historical monitoring period, a cumulative moisture damage memory feature is constructed. Transient environmental factors are constructed based on the ambient temperature, relative humidity, and dew point difference characteristics derived from the ambient temperature and relative humidity of the transmission tower members to be monitored. Based on the timestamp information, the time scale of the meteorological and environmental data is periodically encoded to construct a time periodic code; the time periodic code includes time sine coding and time cosine coding; Based on the duration of moisture exposure of the transmission tower member to be monitored within the preset monitoring time period, a moisture exposure time characteristic is constructed. An environmental feature matrix is ​​constructed based on the damp heat index features, the cumulative damp damage memory features, the transient environmental factors, the time period encoding, and the damp time features.

5. The method according to claim 1, characterized in that, The deep learning network that integrates the cumulative transferred charge conservation constraint includes: a multi-scale local feature embedding layer, a long-range physical evolution coding layer, and a fully connected linear mapping layer; wherein... The multi-scale local feature embedding layer includes a one-dimensional convolutional neural network; The long-range physical evolution coding layer includes a Transformer encoder that incorporates a time decay mask matrix; The loss function of the deep learning network that incorporates the cumulative transferred charge conservation constraint includes a joint objective function that incorporates SmoothL1 loss, physical area conservation loss, and total variation smoothing penalty.

6. The method according to claim 1, characterized in that, The method further includes: Acquire the weight loss data of the natural hanging plates deployed in the same position as the atmospheric corrosion monitor on the transmission tower member to be monitored; Based on the weightlessness data, a target mapping fitting curve is constructed to convert the cumulative transferred charge of the atmospheric corrosion monitor into the initial corrosion depth under stress-free conditions.

7. The method according to claim 1, characterized in that, The dynamic stress of the transmission tower member to be monitored is extracted based on the three-dimensional finite element model of the member, and the initial corrosion depth is corrected based on the corrected fitting curve between the dynamic stress and the corrosion acceleration factor to obtain the target corrosion depth, including: A three-dimensional finite element model is constructed based on the topology and dimensional parameters of the transmission tower members to be monitored. The historical wind speed and direction time history data of the area where the transmission tower member to be monitored is located are converted into dynamic wind pressure load according to the structural wind load calculation specification. The dynamic wind pressure load is input into the three-dimensional finite element model to extract the dynamic stress of the transmission tower member to be monitored; Based on accelerated experiments of metals under different stress conditions in the same environment, the first corrosion rate under different stress conditions and the second corrosion rate under no stress conditions were obtained. Based on the first corrosion rate and the second corrosion rate, the corrosion acceleration factor under different stress conditions is calculated; Nonlinear curve fitting was performed on the corrosion acceleration factor under the different stress conditions to obtain a corrected fitting curve between dynamic stress and corrosion acceleration factor. The initial corrosion depth is corrected based on the modified fitting curve and the dynamic stress of the transmission tower member to be monitored, so as to obtain the target corrosion depth.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images