Overhead line icing prediction method based on physical constraint neural network and monitoring data correction

Abstract

The embodiment of the invention provides an overhead line icing prediction method and device based on a physical constraint neural network and monitoring data correction, and a storage medium. The method comprises the following steps: acquiring initial icing state data and weather forecast data in a future preset time period; inputting the initial icing state data and weather forecast data into a physical constraint neural network model to obtain an initial prediction sequence; in a future preset time period, collecting real-time monitoring data of overhead line icing according to a time interval; whenever real-time monitoring data is collected, determining a data deviation value between the real-time monitoring data at the moment and the icing growth rate and freezing coefficient prediction value at the same moment in the initial prediction sequence, and dynamically correcting the prediction values at all future time points after the moment in the initial prediction sequence according to the data deviation value; and determining the icing thickness of the overhead line at the end of the future preset time period based on the dynamically corrected prediction sequence.

Description

Technical Field

[0001] This application relates to the field of overhead line icing prediction technology, such as power grid / rail transit contact network, and specifically to an overhead line icing prediction method, device and storage medium based on physical constraint neural network and monitoring data correction. Background Technology

[0002] Ice accumulation on overhead power transmission lines and rail transit contact networks can easily lead to serious faults such as line tripping and line breaks, requiring advance prediction of future icing conditions. Existing technologies mostly rely on single monitoring data for short-term forecasts or are based solely on theoretical deductions from weather forecasts, which presents two major problems: First, medium- and long-term forecasts lack dynamic correction based on real-time monitoring data, making it easy for forecast deviations to cause prediction results to deviate from reality; second, they lack constraints from physical laws, resulting in noisy data or inaccurate forecasts with poor robustness, and the fixed forecast duration and update frequency cannot adapt to changes in meteorological elements, making it difficult to meet the precise requirements of flexible operation and maintenance decision-making.

[0003] Relying solely on monitoring data is insufficient for medium- to long-term forecasts of N hours (e.g., N≥72), and relying solely on weather forecasts is prone to distortion due to forecast bias. Furthermore, forecasting models without physical constraints are easily detached from actual patterns, exhibiting poor accuracy and robustness when data is noisy or forecasts fluctuate. Existing solutions suffer from fixed forecast durations and rigid update frequencies, failing to dynamically adapt to meteorological elements, resulting in insufficient forecast flexibility and scenario adaptability. Moreover, the core parameter derivation logic for medium- to long-term forecasts is not clearly defined, leading to incomplete forecasting processes and low operability. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, and storage medium for predicting icing of overhead power lines based on physical constraint neural networks and monitoring data correction.

[0005] To achieve the above objectives, the first aspect of this application provides a method for predicting icing of overhead power lines based on a physical constraint neural network and monitoring data correction, comprising: Acquire the initial icing status data of overhead lines at the predicted start time and the meteorological forecast data for the future preset time period; Based on the changes in meteorological elements in the meteorological forecast data, the time interval between monitoring data collection and forecast value correction is dynamically determined; The initial icing status data and meteorological forecast data are input into the physical constraint neural network model to obtain the initial prediction sequence of the icing growth rate and freezing coefficient within a preset time period, with the time interval as the step size. The physical constraint neural network model is trained by introducing constraints that characterize the physical laws of icing formation. Real-time monitoring data on icing of overhead power lines will be collected at time intervals within a predetermined time period in the future. Whenever real-time monitoring data is collected, the data deviation between the real-time monitoring data at that moment and the predicted values ​​of the ice cover growth rate and freezing coefficient at the same moment in the initial prediction sequence is determined, and the predicted values ​​for all future time points after that moment in the initial prediction sequence are dynamically corrected based on the data deviation value. Based on the dynamically corrected prediction sequence, the icing thickness of the overhead line at the end of a preset time period in the future is determined.

[0006] In this embodiment of the application, dynamically determining the time interval between monitoring data collection and forecast value correction based on the changes in meteorological elements in the meteorological forecast data includes: extracting the change rate of at least one target meteorological element from the meteorological forecast data; comparing the extracted change rate of the target meteorological element with a preset fluctuation threshold corresponding to the target meteorological element; and determining the time interval of the target meteorological element from a plurality of predefined candidate time intervals based on the comparison result, wherein the larger the change rate, the smaller the corresponding time interval value.

[0007] In this embodiment of the application, the target meteorological element includes at least one of temperature, precipitation intensity, and wind speed; the candidate time interval ranges from 1 hour to 6 hours.

[0008] In this embodiment, the physical constraint neural network model is trained by minimizing the total loss function. The total loss function includes a weighted sum of a data fitting loss term and a physical constraint loss term. The physical constraint loss term is used to ensure that the model's prediction results conform to the physical laws of icing formation during training. The physical constraint loss term includes a kinetic constraint term and a thermal balance constraint term. The kinetic constraint term is constructed based on a kinetic model of icing growth and is used to constrain the difference between the predicted icing growth rate and the theoretical value calculated by the kinetic model from the predicted freezing coefficient, wind speed, and icing profile parameters. The thermal balance constraint term is constructed based on the principle of energy conservation and is used to constrain the balance relationship between heat fluxes related to the icing process.

[0009] In this embodiment, the kinetic model is the Makkonen model; the heat flux includes latent heat of freezing, convective heat transfer, evaporative heat dissipation, and radiative heat dissipation.

[0010] In this embodiment of the application, the predicted values ​​for all future time points after a given moment in the initial prediction sequence are dynamically corrected based on the data deviation value. This includes correcting the predicted icing growth rate value in the initial prediction sequence according to the following formula:

[0011]

[0012] Where k is the number of corrections (k=1,2,...,N / Δt - 1), corresponding to the monitoring nodes for Δt, 2Δt, ..., N-Δt hours; Deviation between actual and predicted growth rates; (Deviation between measured and predicted freezing coefficient); β is the correction weight; The revised forecast only updates the prediction results for subsequent time periods (t>kΔt), while retaining the reasonable trends extrapolated from the previous period.

[0013] In this embodiment of the application, determining the icing thickness of the overhead line in a future preset time period based on the dynamically corrected prediction sequence includes calculating the icing thickness of the overhead line at the end of the future preset time period according to the following formula:

[0014] in, This refers to the icing thickness of the overhead line at the end of a future preset time period, where the future preset time period refers to the next N hours. The initial icing thickness is given by the initial icing data; the integration interval is the entire preset future time period. This is the predicted icing growth rate at time τ after dynamic correction; This refers to the mass-to-thickness conversion factor obtained from the calibration of the conductor physical parameters of overhead lines.

[0015] In this embodiment of the application, the real-time monitoring data includes icing profile parameters obtained by a visual sensor and / or equivalent load data obtained by a mechanical sensor.

[0016] A second aspect of this application provides an overhead line icing prediction device based on a physical constraint neural network and monitoring data correction, comprising: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the above-described overhead line icing prediction method based on physical constraint neural networks and monitoring data correction.

[0017] A third aspect of this application provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the above-described overhead line icing prediction method based on a physically constrained neural network and monitoring data correction.

[0018] This solution employs a physically constrained neural network as a robust theoretical deduction engine to ensure that predictions do not violate fundamental physical laws; and a dynamic monitoring and correction engine as a sensitive feedback calibration engine to ensure that predictions closely follow real-world changes. Furthermore, through dynamic frequency adjustments driven by meteorological factors, the system possesses intelligent resource scheduling and risk response capabilities, and provides a complete prediction information chain from process parameters (growth rate, freezing coefficient) to outcome parameters (final thickness), achieving a leap from determining thickness quantity to determining how it grows. This solution effectively addresses the key issues pointed out in the background technology, such as poor mid-to-long-term prediction accuracy, weak model robustness, and rigid update strategies, providing a powerful intelligent tool for precise anti-icing and de-icing decisions for overhead power lines.

[0019] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The schematic diagram illustrates a flowchart of an overhead line icing prediction method based on a physically constrained neural network and monitoring data correction according to an embodiment of this application. Figure 2 The diagram illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] Figure 1 The illustration shows a schematic flowchart of an overhead line icing prediction method based on a physically constrained neural network and monitoring data correction according to an embodiment of this application. Figure 1 As shown in one embodiment of this application, an overhead line icing prediction method based on physical constraint neural networks and monitoring data correction is provided. This embodiment mainly illustrates the application of this method to a processor or controller, including the following steps: Step 102: Obtain the initial icing status data of the overhead line at the predicted start time and the meteorological forecast data for the future preset time period; Step 104: Based on the changes in meteorological elements in the meteorological forecast data, dynamically determine the time interval between monitoring data collection and forecast value correction; Step 106: Input the initial icing state data and meteorological forecast data into the physical constraint neural network model to obtain the initial prediction sequence of the icing growth rate and freezing coefficient within a preset time period with time intervals as steps. The physical constraint neural network model is trained by introducing constraint conditions that characterize the physical laws of icing formation. Step 108: Collect real-time monitoring data on icing of overhead lines at time intervals within a future preset time period; Step 110: Whenever real-time monitoring data is collected, determine the data deviation between the real-time monitoring data at that moment and the predicted values ​​of the ice cover growth rate and freezing coefficient at the same moment in the initial prediction sequence, and dynamically correct the predicted values ​​of all future time points after that moment in the initial prediction sequence based on the data deviation value. Step 112: Based on the dynamically corrected prediction sequence, determine the icing thickness of the overhead line at the end of a future preset time period.

[0023] Overhead lines refer to power transmission lines or rail transit contact networks exposed to the atmosphere, and are the primary carriers of icing disasters. Initial icing status data refers to the quantitative data on the icing condition of the line obtained through on-site sensors at the predicted start time (t=0), mainly including: the initial icing thickness of the contact network at the predicted start time. Initial icing mass (This data can be obtained through on-site sensor measurements). The initial icing thickness is the radial thickness of the icing layer on the conductor surface. The initial icing mass is the mass of icing per unit length of conductor. The preset future time period refers to the desired future time length for forecasting, such as 72 hours (3 days), 120 hours (5 days), etc. This duration can be flexibly set according to operational needs, reflecting the method's adaptability to medium- and long-term forecasts. Meteorological forecast data refers to forecast data released at a specific time resolution (e.g., hourly) within the next N hours from authoritative meteorological platforms. Meteorological elements include: wind speed v(t), ambient temperature T(t), relative humidity H(t), precipitation type and intensity P(t), etc. Simultaneously, the fluctuation characteristics of meteorological elements are extracted, such as the temperature change rate ΔT / Δt and precipitation intensity level. These characteristics are crucial for determining the severity of weather.

[0024] In this scheme, the processor can acquire two types of information: first, the current condition of the line, i.e., the initial icing thickness / mass; and second, the weather forecast for the future. It can be seen that this embodiment not only considers the absolute value of the weather forecast (e.g., the temperature), but also focuses on its changing trend (e.g., whether the temperature is rapidly decreasing or slowly increasing). By analyzing the rate of change of wind speed, temperature, and precipitation intensity, the system can predict in advance whether the future weather will be stable or subject to drastic fluctuations. The time interval (Δt) between monitoring data acquisition and forecast correction refers to the cycle of collecting actual icing data and using that data to correct the forecast result within the forecast period. This time interval is not fixed but dynamically changing. Based on the forward-looking judgment in the first step, the processor can adaptively formulate monitoring and correction strategies. If meteorological elements change slowly (e.g., temperature changes very slowly, no heavy precipitation), the processor can determine that the icing process is relatively stable, and then adopt a lower monitoring frequency (e.g., once every 6 hours) to reduce resource consumption while ensuring accuracy. If meteorological elements fluctuate drastically (such as a sudden drop in temperature or intensified freezing rain), the system determines that the icing conditions are changing rapidly and automatically increases the monitoring frequency (such as once every 1 or 3 hours) to capture the actual situation at a higher frequency, preventing the forecast from becoming inaccurate due to sudden weather changes.

[0025] Next, the processor can input the initial icing status data and weather forecast data into the physically constrained neural network model to obtain an initial prediction sequence of the icing growth rate and freezing coefficient within a preset future time period, with time intervals as steps. The physically constrained neural network model (Ice-PINN) is a neural network model specifically designed for this task. This model is trained by introducing constraints that characterize the physical laws governing icing formation. Its special features are: 1. Physical constraints: During the model training phase, its loss function not only requires the predicted value to be close to the training data (data-driven), but also adds an additional physical loss, which forces the model's output to conform to the known physical laws of ice formation (such as mass conservation and energy conservation). 2. Dual output: The model simultaneously outputs two key physical quantity sequences: icing growth rate and freezing coefficient.

[0026] The icing growth rate refers to the rate at which the mass of ice on a unit length of conductor increases per unit time. The freezing coefficient is a dimensionless parameter between 0 and 1, representing the proportion of supercooled water droplets that actually freeze upon impact with the conductor. The freezing coefficient comprehensively reflects the influence of various microphysical factors such as temperature, droplet size, and surface condition, and is a key physical parameter connecting meteorological conditions and icing growth. The initial prediction sequence refers to the sequence of predicted icing growth rate and freezing coefficient values, covering the next N hours and with a step size of Δt, directly obtained after inputting initial state and meteorological forecast data into a physically constrained neural network model. Real-time monitoring information has not yet been incorporated at this stage.

[0027] Next, the processor can collect real-time monitoring data on overhead line icing at time intervals within a preset future time period. Real-time monitoring data refers to data actually measured by sensors deployed on the line during the prediction period. In a specific embodiment, real-time monitoring data includes icing profile parameters D(t) obtained through a visual sensor and / or equivalent load data W(t) obtained through a mechanical sensor. Whenever real-time monitoring data is collected, the data deviation value between the real-time monitoring data at that moment and the predicted values ​​of the icing growth rate and freezing coefficient at the same moment in the initial prediction sequence is determined, and the predicted values ​​for all future time points after that moment in the initial prediction sequence are dynamically corrected based on the data deviation value. The data deviation value (Δ) refers to the numerical difference between the icing state value inverted or calculated from the real-time monitoring data at a certain monitoring moment and the predicted value at the same moment in the initial prediction sequence. For example, the difference between the measured growth rate and the predicted growth rate quantifies the error between the current prediction and the actual state. Dynamic correction refers to the algorithmic process of instantly correcting the predicted values ​​for future time points that have not yet arrived using the latest calculated data deviation value. The core logic is: assuming that the prediction deviation at the current moment will continue to have a certain impact in the future, this deviation is added to the future prediction value according to a certain proportion (weight β), thereby pulling back the prediction curve and making it closer to the actual evolution trend.

[0028] Through the aforementioned dynamic and rolling correction process, a continuously optimized icing growth rate prediction sequence is obtained until the end of the prediction period. Finally, by integrating this corrected growth rate sequence over time and combining it with the physical conversion coefficient from mass to thickness, the final icing thickness on the line from the initial moment to the end of N hours can be calculated.

[0029] This solution employs a physically constrained neural network as a robust theoretical deduction engine to ensure that predictions do not violate fundamental physical laws; and a dynamic monitoring and correction engine as a sensitive feedback calibration engine to ensure that predictions closely follow real-world changes. Furthermore, through dynamic frequency adjustments driven by meteorological factors, the system possesses intelligent resource scheduling and risk response capabilities, and provides a complete prediction information chain from process parameters (growth rate, freezing coefficient) to outcome parameters (final thickness), achieving a leap from determining thickness quantity to determining how it grows. This solution effectively addresses the key issues pointed out in the background technology, such as poor mid-to-long-term prediction accuracy, weak model robustness, and rigid update strategies, providing a powerful intelligent tool for precise anti-icing and de-icing decisions for overhead power lines.

[0030] In one embodiment, dynamically determining the time interval between monitoring data collection and forecast value correction based on changes in meteorological elements in weather forecast data includes: extracting the rate of change of at least one target meteorological element from the weather forecast data; comparing the extracted rate of change of the target meteorological element with a preset fluctuation threshold corresponding to the target meteorological element; and determining the time interval of the target meteorological element from a plurality of predefined candidate time intervals based on the comparison result, wherein the larger the rate of change, the smaller the corresponding time interval value. In a specific embodiment, the target meteorological element includes at least one of temperature, precipitation intensity, and wind speed; the candidate time interval ranges from 1 hour to 6 hours.

[0031] Different meteorological elements have different fluctuation thresholds. By setting fluctuation thresholds for meteorological elements and dynamically adjusting the monitoring and correction frequency Δt, we can adapt to different meteorological scenarios. 1. Stable meteorological conditions (meeting any of the following conditions): temperature change rate |ΔT / Δt|≤0.5℃ / h, precipitation intensity≤5mm / h, wind speed fluctuation≤2m / s; at this time, the time interval Δt=6h is set, and monitoring and correction are triggered once every 6 hours to balance forecast efficiency and accuracy; 2. Meteorological element fluctuation condition (meeting any of the following conditions): temperature change rate |ΔT / Δt|>0.5℃ / h, precipitation intensity>5mm / h, wind speed fluctuation>2m / s; at this time, the time interval Δt is automatically reduced to 3h or 1h to increase the monitoring and correction frequency and avoid the expansion of forecast deviation caused by meteorological changes. The adjustment range of Δt is 1h≤Δt≤6h, and the threshold range can be preset according to the micro-meteorological characteristics of different regions and operation and maintenance needs.

[0032] In one embodiment, the physical constraint neural network model is trained by minimizing the total loss function, which is a weighted sum of a data fitting loss term and a physical constraint loss term. The physical constraint loss term is used to ensure that the model's predictions conform to the physical laws of icing formation during training. The physical constraint loss term includes a kinetic constraint term and a thermal balance constraint term. The kinetic constraint term is constructed based on a kinetic model of icing growth and is used to constrain the difference between the predicted icing growth rate and the theoretical value calculated by the kinetic model from the predicted freezing coefficient, wind speed, and icing profile parameters. The thermal balance constraint term is constructed based on the principle of energy conservation and is used to constrain the balance relationship between heat fluxes related to the icing process.

[0033] In this embodiment, a physical constraint neural network (Ice-PINN) can be constructed, in which the prediction function and the correction mechanism are deeply coupled, adapting to flexible duration and dynamic frequency.

[0034] The expression for the initial N-hour prediction function (based on forecast + initial state) is as follows:

[0035] in, This is a dual-output neural network with a 4-layer fully connected hidden layer structure (256, 128, 64, and 32 neurons). The activation function is a combination of ReLU and Swish, supporting flexible adaptation to different N and Δt values. The superscript "0" represents the initial prediction result, and t represents a future time (t = Δt, 2Δt, ..., Nh). The function incorporates physical constraint logic to ensure that the deduction conforms to objective laws. Correction is triggered based on the frequency set by Δt, adjusting the prediction results for subsequent time periods using the latest monitoring data. The correction formula is as follows:

[0036]

[0037] Where k is the number of corrections (k=1,2,...,N / Δt - 1), corresponding to the monitoring nodes for Δt, 2Δt, ..., N-Δt hours; Deviation between actual and predicted growth rates; (Deviation between measured and predicted freezing coefficient); β is the correction weight, for example, 0.7; The revised forecast only updates the prediction results for subsequent time periods (t>kΔt), while retaining the reasonable trends extrapolated from the previous period.

[0038] The total loss function of the physically constrained neural network model takes into account both data fitting accuracy and conformity to physical laws. Its formula is as follows:

[0039] in, This is the function value of the total loss function; For data loss items; This is the weighting coefficient, and its specific value can be 0.85; This is the physical constraint loss term.

[0040] Among them, data loss items The calculation formula is as follows:

[0041] Where M = N / Δt, that is, the total number of nodes every N hours over Δt hours, and the subscript ref is the reference value for the fusion of forecast data and monitoring data; where, The box on the right refers to its cumulative bias, and the purpose of the loss function is to reduce the bias.

[0042] Physical constraint loss term The calculation formula is as follows:

[0043] In one embodiment, the kinetic model is the Makkonen model; the heat flux includes latent heat of freezing, convective heat transfer, evaporative heat dissipation, and radiative heat dissipation. The expression for the Makkonen kinetic constraints is as follows:

[0044] Among them, α1=0.02 and α2=0.05 are environmental adaptation coefficients, which can be finely adjusted according to the micro-meteorological characteristics of different regions.

[0045] The expression for the thermal equilibrium constraint is as follows:

[0046] in, (Latent heat of freezing) includes item, (Viscous heat dissipation) (Heat conduction) (Meteorological environment heat exchange) (Convection cooling) (Evaporative cooling) (Radiative heat dissipation) are all known quantities calculated based on sensor data and classical thermal formulas.

[0047] In one embodiment, determining the icing thickness of the overhead line in a future preset time period based on the dynamically corrected prediction sequence includes calculating the icing thickness of the overhead line at the end of the future preset time period according to the following formula:

[0048] in, This refers to the icing thickness of the overhead line at the end of a future preset time period, where the future preset time period refers to the next N hours. The initial icing thickness is given by the initial icing data; the integration interval is the entire preset future time period. This is the predicted icing growth rate at time τ after dynamic correction; This refers to the mass-to-thickness conversion factor obtained based on the physical parameters of the overhead line conductors. Specifically, =0.03.

[0049] Figure 1 This is a flowchart illustrating an overhead power line icing prediction method based on a physical constraint neural network and monitoring data correction in one embodiment. It should be understood that, although... Figure 1The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0050] In one embodiment, an overhead line icing prediction device (not shown in the figure) based on physical constraint neural network and monitoring data correction is provided. The device includes: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the overhead line icing prediction method based on physical constraint neural networks and monitoring data correction as described in any of the above embodiments.

[0051] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and by adjusting kernel parameters, an overhead power line icing prediction method based on physical constraint neural networks and monitoring data correction can be implemented.

[0052] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0053] This application provides a storage medium storing a program that, when executed by a processor, implements the above-described overhead line icing prediction method based on physical constraint neural networks and monitoring data correction.

[0054] This application provides a processor for running a program, wherein the program executes the above-described overhead line icing prediction method based on physical constraint neural network and monitoring data correction.

[0055] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 2As shown. The computer device includes a processor A01, a network interface A02, memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A04. The network interface A02 is used for communication with external terminals via a network connection. When the processor A01 executes the computer program B02, it implements an overhead power line icing prediction method based on a physically constrained neural network and monitoring data correction.

[0056] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0057] This application provides a computer (electronic) device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the above-mentioned overhead line icing prediction methods based on physical constraint neural networks and monitoring data correction.

[0058] This application also provides a computer program product that, when executed on a data processing device, is suitable for executing a program that initializes an overhead line icing prediction method based on a physically constrained neural network and monitoring data correction.

[0059] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0064] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0065] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0066] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0067] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for predicting icing on overhead power lines based on physical constraint neural networks and monitoring data correction, characterized in that, The method includes: Acquire initial icing status data of overhead lines at the predicted start time and meteorological forecast data for the future preset time period; Based on the changes in meteorological elements in the meteorological forecast data, the time interval between monitoring data collection and forecast value correction is dynamically determined; The initial icing state data and the weather forecast data are input into a physical constraint neural network model to obtain an initial prediction sequence of the icing growth rate and freezing coefficient within the preset future time period, with the time interval as the step size. The physical constraint neural network model is trained by introducing constraints that characterize the physical laws of icing formation. During the predetermined future time period, real-time monitoring data on icing of the overhead power line will be collected at the specified time intervals. Whenever the real-time monitoring data is collected, the data deviation value between the real-time monitoring data at that moment and the predicted values ​​of the icing growth rate and freezing coefficient at the same moment in the initial prediction sequence is determined, and the predicted values ​​of all future time points after that moment in the initial prediction sequence are dynamically corrected based on the data deviation value. Based on the dynamically corrected prediction sequence, the icing thickness of the overhead line at the end of the preset future time period is determined.

2. The method according to claim 1, characterized in that, Based on the changes in meteorological elements in the aforementioned meteorological forecast data, the time interval between monitoring data collection and forecast value correction is dynamically determined, including: Extract the rate of change of at least one target meteorological element from the meteorological forecast data; The rate of change of the extracted target meteorological element is compared with the preset fluctuation threshold corresponding to the target meteorological element; The time interval of the target meteorological element is determined from a predefined pool of candidate time intervals based on the comparison results, wherein the greater the rate of change, the smaller the corresponding time interval value.

3. The method according to claim 2, characterized in that, The target meteorological elements include at least one of temperature, precipitation intensity, and wind speed; the candidate time interval ranges from 1 hour to 6 hours.

4. The method according to claim 1, characterized in that, The physical constraint neural network model is trained by minimizing the total loss function, which is a weighted sum of a data fitting loss term and a physical constraint loss term. The physical constraint loss term is used during training to ensure that the model's prediction results conform to the physical laws of ice formation. The physical law constraint loss term includes a kinetic constraint term and a thermal balance constraint term. The kinetic constraint term is constructed based on the kinetic model of icing growth and is used to constrain the difference between the predicted value of the icing growth rate and the theoretical value calculated by the kinetic model from the predicted value of the freezing coefficient, wind speed and icing profile parameters. The thermal balance constraint term is constructed based on the principle of energy conservation and is used to constrain the balance relationship between heat fluxes related to the icing process. The formula for the total loss function of the physically constrained neural network model is as follows: ； in, This is the function value of the total loss function; For data loss items; These are the weighting coefficients; This is the physical constraint loss term.

5. The method according to claim 4, characterized in that, The kinetic model is the Makkonen model; the heat flux includes latent heat of freezing, convective heat transfer, evaporative heat dissipation, and radiative heat dissipation.

6. The method according to claim 1, characterized in that, The step of dynamically correcting the predicted values ​​for all future time points after that moment in the initial prediction sequence based on the data deviation value includes correcting the predicted icing growth rate in the initial prediction sequence according to the following formula: Where k is the number of corrections (k=1,2,...,N / Δt - 1), corresponding to the monitoring nodes at hours Δt, 2Δt, ..., N-Δt; Deviation between actual and predicted growth rates; (Deviation between measured and predicted freezing coefficient); β is the correction weight; The revised forecast only updates the forecast results for subsequent time periods (t > kΔt), while retaining the reasonable trends extrapolated from earlier periods.

7. The method according to claim 1, characterized in that, Based on the dynamically corrected prediction sequence, the icing thickness of the overhead line in the future preset time period is determined, including the icing thickness at the end of the future preset time period calculated according to the following formula: in, The icing thickness of the overhead line at the end of the future preset time period, where the future preset time period refers to the next N hours; The ice thickness is defined in the initial icing state data; the integration interval is the entire preset future time period. This is the predicted icing growth rate at time τ after dynamic correction; This refers to the mass-to-thickness conversion factor obtained by calibrating the physical parameters of the overhead line conductors.

8. The method according to claim 1, characterized in that, The real-time monitoring data includes icing profile parameters obtained through a visual sensor and / or equivalent load data obtained through a mechanical sensor.

9. An overhead power line icing prediction device based on physical constraint neural network and monitoring data correction, characterized in that, include: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the overhead line icing prediction method based on a physical constraint neural network and monitoring data correction as described in any one of claims 1 to 8.

10. A machine-readable storage medium storing instructions thereon, characterized in that, When executed by a processor, the instruction causes the processor to be configured to perform the overhead line icing prediction method based on physical constraint neural networks and monitoring data correction as described in any one of claims 1 to 8.

Images