Physical constraint-based power line icing prediction method, system, device and medium

Abstract

The present application belongs to the technical field of electric wire icing prediction, and discloses a physical constraint-based electric wire icing prediction method, system, device and medium, to solve the technical problems of poor space adaptability of the existing pure physical model and high false negative rate caused by gradient collapse of the pure deep learning model in sparse icing events. The method comprises: acquiring historical meteorological and terrain data of a target area, performing logarithmic smoothing processing on the skew distribution variables to construct a multi-dimensional environmental input tensor; inputting the input tensor into a backbone network to extract bottleneck features; performing meteorological prediction output on the bottleneck features to obtain a continuous meteorological field, while predicting dynamic physical parameters constrained by a physical interval; replacing a hard threshold with a smoothing function to construct a differentiable physical layer, inputting the meteorological field and the dynamic parameters into the layer, and calculating a physical theoretical icing increment based on the Makkonen equation; and outputting a nonlinear icing residual error by using a residual decoder, and fusing the residual error with the physical increment after scale transformation to obtain a final prediction result.

Description

Technical Field

[0001] This invention belongs to the field of power line icing prediction technology, specifically relating to a method, system, device, and medium for power line icing prediction based on physical constraints. Background Technology

[0002] In the field of power transmission, ice and snow accumulation on transmission lines during extreme weather significantly increases the mechanical load on conductors, potentially leading to accidents such as conductor galloping, insulator flashover, and even tower collapse and line breakage, posing a serious threat to the safe operation of the power grid. Therefore, achieving high-precision prediction of power line icing is of great significance for disaster early warning and emergency dispatch.

[0003] Currently, power line icing prediction mainly relies on two technical approaches. The first is a purely physical model, which is based on thermodynamics and kinetics, using meteorological parameters such as wind speed, temperature, and liquid water content to calculate the collision, capture, and freezing processes of supercooled water droplets. However, this type of model has significant limitations: key physical parameters (such as collision efficiency) are usually set as globally fixed empirical constants, which cannot adapt to the micro-meteorological differences in different spatial locations under complex terrain; at the same time, physical formulas are difficult to accurately characterize the complex effects of nonlinear factors such as turbulence in nature, resulting in large prediction errors in specific terrain areas. The second type is the purely data-driven deep learning model that has emerged in recent years. This type of model attempts to directly learn the mapping relationship between meteorological input fields and icing increments through neural networks. However, icing events are extremely sparse samples in both time and space (the icing increment is zero in most spatiotemporal grids). Without the constraints of physical laws, deep learning models are prone to getting trapped in local optima that always output zero values ​​in order to quickly reduce global errors, resulting in a high false negative rate for real icing events.

[0004] In summary, existing pure physics models suffer from poor spatial adaptability, while pure deep learning models are prone to gradient collapse when dealing with sparse icing events. How to incorporate physical thermodynamic mechanisms into deep learning frameworks in an embedded differentiable form to achieve a deep integration of physical laws and data-driven approaches is a pressing technical problem in this field. Summary of the Invention

[0005] Based on the aforementioned shortcomings and deficiencies in the existing technology, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the existing technology. In other words, one of the objectives of this invention is to provide a method, system, device, and medium for predicting power line icing based on physical constraints that meets one or more of the aforementioned requirements, so as to achieve the goal of deeply integrating physical laws with data-driven approaches, improving the accuracy of icing prediction under complex terrain, and reducing the false negative rate of extreme sparse icing events.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting icing on power lines based on physical constraints, comprising the following steps: S1. Obtain historical surface meteorological field data and high-resolution topographic data of the target area, and perform log smoothing on variables with extreme skewed distributions in the data to construct a multidimensional environmental input tensor; S2. Input the multidimensional environment input tensor into the backbone network to extract bottleneck features; S3. Perform meteorological and parameter prediction on the bottleneck characteristics, including: S31. Restore the bottleneck features to the original high-resolution space and output the continuous meteorological field prediction results for future periods; S32. Based on the bottleneck characteristics, predict the dynamic physical parameters constrained by the physical interval; S4. An optimized differentiable physical layer is constructed by replacing the hard threshold with a smooth function. The prediction results of the continuous meteorological field and the dynamic physical parameters are input into the optimized differentiable physical layer, and the physical theoretical ice accumulation increment is calculated based on the Makkonen equation. S5. The bottleneck features are processed using a residual decoder to output nonlinear icing residuals, and the theoretical icing increment is scaled and fused with the nonlinear icing residuals to obtain the final icing prediction result.

[0007] In a second aspect, the present invention provides a power line icing prediction system based on physical constraints, for implementing the power line icing prediction method as described in the first aspect.

[0008] Thirdly, the present invention provides an electronic device, the electronic device including a memory, a processor and a computer program, wherein when the computer program is executed by the processor, it implements the wire icing prediction method as described in the first aspect.

[0009] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the wire icing prediction method as described in the first aspect.

[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. Compared to purely data-driven models, this invention transforms the Makkonen physical equations into a differentiable physics layer embedded in a deep network. This allows the model to be guided by physical laws during training, rather than simply relying on data fitting. Simultaneously, a physical consistency loss term is introduced into the multi-task composite loss function, penalizing the model when it predicts positive icing at temperatures above the icing critical point, thus effectively curbing false predictions that violate thermodynamic principles. Compared to purely physical models, this invention, through differentiable design, enables the physics layer to participate in backpropagation, achieving a deep integration of physical laws and data-driven approaches, thereby improving the accuracy and reliability of predicting extreme icing events.

[0011] 2. To address the problem of extremely sparse icing events and the tendency of pure deep learning models to get trapped in local optima with zero predicted values, this invention truncates the gradients of the meteorological prediction branch and the icing prediction branch during the end-to-end optimization process. This ensures that the backpropagation of the two gradients does not interfere with each other, preventing the large gradient of the meteorological prediction task from overwhelming the effective gradient of the icing prediction task. Simultaneously, a false negative penalty loss term is introduced into the multi-task composite loss function. When meteorological conditions meet the icing conditions and icing actually exists, an additional penalty is imposed on the model's tendency to underreport. A residual regularization loss term is also introduced to constrain the amplitude of the nonlinear icing residual, encouraging the model to prioritize the theoretical icing amount output by the differentiable physics layer. The synergistic effect of these mechanisms enables the model to accurately capture extreme icing events even in a large number of zero-icing scenarios, significantly reducing the false negative rate compared to existing deep learning methods.

[0012] 3. Compared to traditional physical models that set key parameters such as collision efficiency as globally fixed constants, this invention predicts a collision efficiency field that varies with spatial location based on bottleneck characteristics. This field serves as a dynamic parameter in the Makkonen equation, and the predicted values ​​are limited to a reasonable range through physical interval constraints (such as the Sigmoid function). This design allows physical parameters to be dynamically adjusted according to terrain and micro-meteorological environment, overcoming the limitation of fixed constants in adapting to spatial heterogeneity under complex terrain. Compared to traditional physical models, this design offers superior local forecast accuracy and micro-terrain adaptability.

[0013] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0015] Figure 1This is a schematic diagram of the process for predicting icing on power lines as described in Embodiment 1 of the present invention.

[0016] Figure 2 This is a schematic diagram of the neural network model described in Embodiment 1 of the present invention.

[0017] Figure 3 This is a schematic diagram of the wire icing prediction system described in Embodiment 2 of the present invention.

[0018] Figure 4 This is a structural diagram of the electronic device described in Embodiment 3 of the present invention.

[0019] Icon labels: 400. Electronic devices; 401. Processor; 402. Communication bus; 403. User interface; 404. Network interface; 405. Memory. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0022] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0023] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0024] The power line icing prediction method described in the embodiments of this specification is applied to the field of meteorological disaster early warning in power systems, specifically applicable to the prediction of ice thickness on transmission lines under extreme weather conditions such as freezing rain and rime in winter. In these scenarios, the application of the power line icing prediction method aims to provide early warning information to power grid dispatching departments, enabling timely implementation of anti-icing measures such as DC de-icing and AC short-circuit de-icing, reducing the risk of accidents such as conductor galloping, insulator flashover, tower collapse, and line breakage, and ensuring the safe and stable operation of the power system.

[0025] The following is a brief explanation of the bottleneck features, smoothing function, hard threshold, differentiable physics layer, Makkonen equation, residual decoder, physical theory ice accumulation increment, and collision efficiency involved in several embodiments of this specification: Bottleneck features refer to low-dimensional, high-semantic feature representations obtained by downsampling and feature compression of the original high-dimensional input data (multidimensional environmental input tensor) through a backbone network. These features retain the key spatial topological relationships and temporal evolution trends in the original data while removing redundant information, serving as the common input basis for subsequent weather forecasts, parameter predictions, and residual predictions.

[0026] A smoothing function is a continuously differentiable function used to replace the hard threshold gating conditions in traditional physical equations. In this invention, the smoothing function specifically employs a Sigmoid function with learnable parameters to handle freezing temperature thresholds (e.g., determining freezing at 0°C) and precipitation occurrence thresholds (e.g., whether precipitation has occurred). By replacing the stepped hard thresholds with an S-shaped smooth curve, the gradient of the physical layer can smoothly pass through the threshold boundaries during backpropagation, thereby achieving end-to-end joint training of the physical formulas and the neural network.

[0027] Hard thresholding refers to a step function based on conditional judgments in traditional physical models, such as "freezing occurs if the temperature is below 0°C, otherwise it does not freeze." These functions are non-differentiable at the threshold boundaries, preventing gradients from passing through and thus hindering the embedding of physical formulas into neural networks for backpropagation training. This invention uses a smooth function instead of hard thresholding in its differentiable physical layer, thereby solving the gradient breakage problem.

[0028] The differentiable physics layer is an implementation that transforms the classic Makkonen icing physics equations into a neural network module. This layer uses a smoothing function to replace the hard threshold gating condition in the traditional equations, making the entire physical calculation process differentiable throughout the model's backpropagation. The differentiable physics layer receives meteorological forecast results and dynamic physical parameters as inputs and outputs the theoretical icing increment as the physical benchmark for subsequent residual fusion.

[0029] The Makkonen equation, proposed by Finnish scientist Makkonen, is a classical thermodynamic and kinetic model describing the process of icing on power lines. Based on the law of conservation of mass, this equation calculates the rate of increase in ice mass per unit time and per unit length of power line using parameters such as collision efficiency, adhesion efficiency, freezing efficiency, liquid water content, wind speed, and wire diameter. This invention constructs a differentiable physical layer based on the Makkonen equation, serving as the theoretical core of the physical constraints.

[0030] The residual decoder is a neural network upsampling module with a structure similar to that of a weather decoder. It is used to learn from bottleneck features the influence of nonlinear factors (such as conductor torsion, irregular turbulence, and uneven water droplet adhesion) on icing that cannot be accurately characterized by physical models. The residual decoder outputs nonlinear icing residuals, which are used to compensate for and correct the theoretical icing increment.

[0031] The theoretical icing increment refers to the theoretical icing amount calculated based on the Makkonen equation after inputting the predicted meteorological field (temperature, precipitation, wind speed, etc.) and dynamic physical parameters (such as spatially varying collision efficiency fields) into a differentiable physical layer. This increment serves as a physical benchmark, providing a foundational value for subsequent residual fusion.

[0032] Collision efficiency is a key physical parameter in the Makkonen equation, representing the probability of a supercooled water droplet colliding with a conductor under the influence of airflow, with a value range of [0,1]. In traditional physical models, collision efficiency is usually set as a globally fixed empirical constant, which cannot adapt to the differences in airflow disturbances under different terrains. In this invention, collision efficiency is predicted by bottleneck characteristics as a dynamic parameter field that varies with spatial location, replacing the fixed constant, thereby achieving spatial adaptation to complex terrains and micro-meteorological environments.

[0033] Example 1: like Figure 1 As shown, this embodiment provides a method for predicting power line icing based on physical constraints, including the following steps: Step S1, Data Acquisition and Preprocessing: Historical surface meteorological field data and high-resolution topographic data of the target area are acquired. The historical surface meteorological field data includes high-resolution surface meteorological variables such as near-surface temperature, relative humidity, precipitation, zonal wind speed, and meridional wind speed. The high-resolution topographic data is digital elevation model (DEM) data, used to provide spatial geographic information of the target area to characterize the impact of topography on wind field and precipitation distribution.

[0034] Because neural networks are highly sensitive to data with different dimensions, continuous variables need to be standardized before input. Specifically, Z-score standardization is performed on continuous variables such as temperature and wind speed to make their mean 0 and standard deviation 1. Furthermore, since meteorological variables such as precipitation have extremely skewed distribution characteristics (i.e., zero values ​​most of the time, with large values ​​appearing only occasionally), this embodiment performs logarithmic smoothing on variables with extremely skewed distributions in the data. Specifically, for precipitation variables, logarithmic transformation is used to compress their numerical range, making the data distribution more uniform and facilitating neural network learning. After processing, meteorological data and topographic data are spatiotemporally aligned and stitched together to construct a multidimensional environmental input tensor.

[0035] Step S2, Bottleneck Feature Extraction: The multidimensional environment input tensor is input into the backbone network to extract bottleneck features.

[0036] In this embodiment, the backbone network adopts a SwingTransformer architecture based on a self-attention mechanism of local windows and moving windows. Specifically: First, the multidimensional environment input tensor is divided into non-overlapping local windows, and feature interaction is performed within each local window through a self-attention mechanism. Second, feature interaction between windows is achieved through a moving window self-attention mechanism. Finally, multi-scale bottleneck features containing spatial topological relationships and temporal evolution trends are extracted through hierarchical downsampling and block merging operations.

[0037] The bottleneck feature is the common input for subsequent weather forecasts, parameter forecasts, and residual forecasts.

[0038] Step S3, Weather Forecast and Parameter Prediction: Meteorological and parameter predictions are performed on the bottleneck characteristics.

[0039] Step S31: The weather forecast includes restoring the bottleneck features to their original high-resolution spatial representation and outputting the continuous weather field forecast results for future time periods. In this embodiment, the Pixel Shuffle upsampling module is used to restore the spatial resolution step by step, and the output weather field includes the temperature field, precipitation field, humidity field, and wind field for future time periods.

[0040] Step S32, the parameter prediction includes: predicting dynamic physical parameters constrained by the physical interval based on the bottleneck characteristics.

[0041] In this embodiment, the dynamic physical parameters include the collision efficiency α1. The collision efficiency in the traditional Makkonen equation is a fixed constant, which cannot adapt to the differences in airflow disturbance under different terrains. This embodiment uses an independent convolutional prediction head to implicitly predict the collision efficiency field that varies with spatial location from the bottleneck features, and uses the Sigmoid function to constrain the predicted value within a physically reasonable range of [0,1].

[0042] Step S4, Differentiable physical layer calculation: An optimized differentiable physical layer is constructed by replacing the hard threshold with a smooth function. The prediction results of the continuous meteorological field and the dynamic physical parameters are input into the optimized differentiable physical layer, and the physical theoretical ice accumulation increment is calculated based on the Makkonen equation.

[0043] In this embodiment, the smoothing function is a sigmoid function with learnable parameters, used to replace the hard threshold gating conditions in traditional physical equations, including the freezing temperature threshold (if T < 0°C then freezing else no freezing) and the precipitation threshold (if precipitation > 0 then precipitation else no precipitation). By replacing the hard thresholds with a differentiable smoothing function, the physical layer becomes differentiable throughout the model backpropagation process, allowing the gradient to flow smoothly across the threshold boundaries.

[0044] Before inputting the differentiable physical layer, the prediction results of the continuous meteorological field are denormalized to restore them to real physical units, namely, the unit of temperature T is set to ℃, the unit of relative humidity RH is set to %, the unit of precipitation TP is set to mm / h, and the unit of wind speed WS is set to m / s, for use in physical layer calculations.

[0045] To incorporate rigorous physical constraints into the deep learning framework, this embodiment constructs a fully differentiable icing operator based on the classic Makkonen theory. This module, as a deterministic theoretical regularization term, maps the meteorological state variables (the bottleneck features) predicted by the backbone network in step S2 to icing increment estimates that conform to physical laws. Specifically: First, using the Chaine-Castonguay empirical formula, precipitation TP is parameterized as the liquid water content in the air. The formula is as follows: , In the formula, This indicates the amount of precipitation.

[0046] The icing dynamics are derived based on the law of conservation of mass for a cylindrical conductor. According to the classic Makkonen model, the rate of increase in icing mass per unit time and unit length is... Defined by the following standard formula: , In the formula, Indicates wind speed. D The current diameter of the conductor; the three coefficients in the formula represent the three core physical stages of the supercooled water droplet capture process. Collision efficiency represents the probability of a supercooled water droplet colliding with a wire, with a value range of [0,1]. The adhesion efficiency represents the probability that a water droplet will adhere to the surface of the wire after a collision, and its value ranges from [0,1]. The freezing efficiency represents the probability that a water droplet will freeze into ice after adhesion, and its value ranges from [0,1].

[0047] Among them, collision efficiency In this embodiment, the spatial variation dynamic physical parameters predicted by the bottleneck features replace the fixed constants in the traditional Makkonen equation, so as to achieve adaptation to different terrains and micro-meteorological environments.

[0048] To solve for the ice layer thickness (i.e., radius) R The increment of ) is compared with the radial geometric relationship. (Assuming the ice layer uniformly covers the surface of the conductor), the diameter term is then... D It is naturally eliminated. From this, the direct differential equation for radial icing thickness can be derived: , In the formula, This is the standard ice density.

[0049] By combining the above formulas, this invention constructs a deterministic ideal physical model and outputs the theoretical physical ice accumulation increment, which provides a rigorous thermodynamic and kinetic theoretical benchmark for neural networks.

[0050] Step S5, Residual Prediction and Fusion: The bottleneck features are processed using a residual decoder to output a nonlinear icing residual. The theoretically predicted icing increment, after scaling, is then fused with this nonlinear icing residual to obtain the final icing prediction result. Specifically: First, the residual decoder adopts an upsampling structure similar to that of the meteorological decoder, and learns the influence of nonlinear factors (such as conductor torsion, irregular turbulence, etc.) on ice accumulation from the bottleneck features that cannot be characterized by the physical model, and outputs nonlinear ice accumulation residuals.

[0051] Secondly, since the theoretical ice accumulation increment has real physical units (such as mm), while the nonlinear ice accumulation residual is an eigenvalue of the latent space of the neural network, the two have different dimensions and scales and cannot be directly added. In this embodiment, the theoretical ice accumulation increment is subjected to a signed logarithmic mapping and normalization process: first, a Sign-Log1p transformation is performed, and then the transformed value is linearly normalized to the latent space scale to obtain the normalized physical increment.

[0052] Finally, the normalized physical increment and the nonlinear icing residual are added pixel by pixel to obtain the final icing prediction result.

[0053] Furthermore, this embodiment also includes a neural network model training step: using a multi-task composite loss function to perform end-to-end optimization of the trainable parameters in the above steps.

[0054] It is understood that the key structures of the neural network model include: the backbone network (corresponding to step S2), the meteorological decoder (corresponding to step S31), the parameter prediction branch (corresponding to step S32), the differentiable physical layer (corresponding to step S4), and the residual decoder (corresponding to step S5). Figure 2 As shown, the backbone network is used to extract bottleneck features; the meteorological decoder is used to restore the bottleneck features to the original high-resolution space and output continuous meteorological field prediction results; the parameter prediction branch is used to predict dynamic physical parameters constrained by physical intervals based on the bottleneck features; the differentiable physics layer is used to calculate the theoretical ice accumulation increment based on the Makkonen equation; and the residual decoder is used to process the bottleneck features to output nonlinear ice accumulation residuals. The neural network model takes the multidimensional environment input tensor as input and the final ice accumulation prediction result as output, and constitutes a complete forward computation path through the aforementioned steps S1 to S5.

[0055] The multi-task composite loss function (Total Loss) is defined as follows: , In the formula, the meanings of each loss term are as follows: Losses from meteorological field reconstruction This loss term is used to constrain the continuous weather field predictions for future periods, minimizing the error between the model's output weather field and actual weather observations. This loss term ensures the model's ability to accurately predict future weather conditions, providing reliable weather input for subsequent physical layer calculations.

[0056] Icing-weighted error loss The error is calculated based on the difference between predicted and actual ice accumulation. Considering the sparsity of ice accumulation events, this loss term assigns higher weights to positive samples (those with ice accumulation) to alleviate the sample imbalance problem.

[0057] Physical consistency loss This is used to penalize the model for violating thermodynamic laws by predicting positive icing when the temperature is significantly above the freezing threshold. Specifically, if the temperature in the continuous meteorological field prediction is higher than a preset freezing threshold (e.g., 0°C), and the final icing prediction is positive, a penalty is imposed on that prediction. This ensures that the model output adheres to fundamental thermodynamic laws and avoids physical errors.

[0058] Penalty for underreporting losses This term penalizes the model's tendency to underreport when meteorological conditions meet the criteria for icing and icing actually exists. It is specifically designed to address the extreme sparsity of icing events, forcing the model to focus on rare icy samples and effectively reducing the false negative rate.

[0059] Residual regularization loss This loss term constrains the magnitude of the nonlinear icing residual, encouraging the model to preferentially rely on the theoretical icing amount output by the differentiable physics layer, rather than over-relying on the residual branch. This loss term ensures the dominance of the physics layer throughout the prediction framework, with the residual serving only as fine-tuning compensation.

[0060] Of the above-mentioned loss items, , , , and These are the corresponding hyperparameter weights, used to balance the contribution of each loss term to the total loss. Through the joint constraints of the above multi-task composite loss function, the model not only numerically approximates the actual observed ice accumulation, but also strictly follows the physical laws of nature in its internal logic.

[0061] During the end-to-end optimization process, the gradients corresponding to the continuous meteorological field prediction results output in step S31 and the gradients corresponding to the icing prediction results output in steps S4 to S5 are truncated to prevent the backpropagation of the two gradients from interfering with each other and to prevent the gradients of the meteorological prediction task from dominating the backpropagation and drowning out the effective gradients of sparse icing events.

[0062] Example 2: like Figure 3 As shown, this embodiment provides a power line icing prediction system based on physical constraints, used to implement the power line icing prediction method as described in Embodiment 1, including: The data processing module is used to acquire historical surface meteorological field data and high-resolution topographic data of the target area, and to perform log smoothing on variables with extreme skewed distributions in the data to construct a multidimensional environmental input tensor. The feature extraction module is used to input the multidimensional environment input tensor into the backbone network to extract bottleneck features; The weather forecasting and parameter prediction module is used to perform weather forecasting and parameter prediction on the bottleneck feature, including: restoring the bottleneck feature to the original high-resolution spatial output continuous weather field prediction result, and predicting dynamic physical parameters constrained by physical intervals based on the bottleneck feature. The differentiable physics layer module is used to construct an optimized differentiable physics layer by replacing the hard threshold with a smooth function. The continuous meteorological field prediction results and the dynamic physical parameters are input into the optimized differentiable physics layer, and the physical theoretical ice accumulation increment is calculated based on the Makkonen equation. The residual fusion module is used to process the bottleneck features using the residual decoder to output nonlinear icing residuals, and to fuse the physical theoretical icing increment with the nonlinear icing residuals after scaling transformation to obtain the final icing prediction result.

[0063] Example 3: like Figure 4 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0064] The communication bus can be used to enable communication between the various components mentioned above.

[0065] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0066] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0067] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0068] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a prediction application. The processor can be used to call the prediction application stored in the memory and execute the steps of the wire icing prediction method mentioned in the foregoing embodiments.

[0069] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0070] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0071] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0072] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0073] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0074] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A method for predicting icing on power lines based on physical constraints, characterized in that, Including the following steps: S1. Obtain historical surface meteorological field data and high-resolution topographic data of the target area, and perform log smoothing on variables with extreme skewed distributions in the data to construct a multidimensional environmental input tensor; S2. Input the multidimensional environment input tensor into the backbone network to extract bottleneck features; S3. Perform meteorological and parameter prediction on the bottleneck characteristics, including: S31. Restore the bottleneck features to the original high-resolution space and output the continuous meteorological field prediction results for future periods; S32. Based on the bottleneck characteristics, predict the dynamic physical parameters constrained by the physical interval; S4. An optimized differentiable physical layer is constructed by replacing the hard threshold with a smooth function. The prediction results of the continuous meteorological field and the dynamic physical parameters are input into the optimized differentiable physical layer, and the physical theoretical ice accumulation increment is calculated based on the Makkonen equation. S5. The bottleneck features are processed using a residual decoder to output nonlinear icing residuals, and the theoretical icing increment is scaled and fused with the nonlinear icing residuals to obtain the final icing prediction result.

2. The method for predicting icing on power lines according to claim 1, characterized in that, In step S4, a smooth function is used to replace the hard threshold to construct an optimized differentiable physical layer, specifically as follows: A sigmoid function with learnable parameters is used as the smoothing function to replace the hard threshold gating conditions in the freezing temperature threshold and precipitation occurrence threshold.

3. The method for predicting icing on power lines according to claim 1, characterized in that, The dynamic physical parameters include collision efficiency. Specifically, the predicted dynamic physical parameters constrained by the physical interval in step S32 are as follows: Based on the bottleneck characteristics, the collision efficiency field that varies with spatial location is predicted and used as a dynamic parameter in the calculation of the physical theory of ice accumulation increment using the Makkonen equation.

4. The method for predicting icing on power lines according to claim 1, characterized in that, Step S5 involves fusing the theoretically measured ice accumulation increment with the nonlinear ice accumulation residual after scaling, including: The physical theory of ice accumulation increment is normalized by performing a sign logarithmic mapping, resulting in a normalized physical increment. The normalized physical increment is added pixel by pixel to the nonlinear icing residual.

5. The method for predicting icing on power lines according to claim 1, characterized in that, Also includes: The trainable parameters in steps S2 to S5 are optimized end-to-end using a multi-task composite loss function. The multi-task composite loss function includes: A physical consistency loss term is used to penalize the tendency of the final icing prediction result to be positive when the temperature in the continuous meteorological field prediction result is higher than a preset icing threshold. The residual regularization loss term is used to constrain the magnitude of the nonlinear icing residual.

6. The method for predicting icing on power lines according to claim 5, characterized in that: In the end-to-end optimization process, the gradients corresponding to the continuous meteorological field prediction results output in step S31 and the gradients corresponding to the ice accumulation prediction results output in steps S4 to S5 are truncated so that the back propagation of the two gradients does not interfere with each other.

7. The method for predicting icing on power lines according to claim 1, characterized in that, Step S2 extracts bottleneck features, including: The multidimensional environment input tensor is divided into non-overlapping local windows, and feature interaction is performed within each local window through a self-attention mechanism. Feature interaction between windows is achieved through a self-attention mechanism for moving windows; Multi-scale bottleneck features are extracted through hierarchical downsampling and block merging operations.

8. A power line icing prediction system based on physical constraints, characterized in that, Used to implement the wire icing prediction method as described in any one of claims 1 to 7.

9. An electronic device, the electronic device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by the processor, it implements the wire icing prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the wire icing prediction method as described in any one of claims 1 to 7.

Images