Power transmission line galloping early warning method and device, computer device and storage medium

Abstract

The present application relates to the technical field of power transmission line state monitoring, and discloses a power transmission line galloping early warning method and device, computer equipment and a storage medium, the method comprising: acquiring meteorological monitoring data covering the power transmission line area; extracting galloping physical criterion features based on the meteorological monitoring data; training a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power transmission line galloping early warning model; inputting actual meteorological monitoring data into the power transmission line galloping early warning model to obtain an original galloping prediction result, and early warning the power transmission line galloping risk based on the original galloping prediction result. The present application solves the problem of poor interpretability of the power transmission line galloping early warning model in the prior art, which cannot be applied to large-scale, multi-scenario power transmission line galloping risk identification tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transmission line state monitoring, in particular to a power transmission line galloping early warning method and device, computer equipment and a storage medium. BACKGROUND

[0002] As a key infrastructure of the power system, the safe and stable operation of the power transmission line is directly related to the reliability of power transmission and the safety of the power supply system. In winter or the transition season, the conductor galloping phenomenon often occurs under complex weather conditions. The so-called galloping is a low-frequency, large-amplitude, quasi-periodic self-excited oscillation of the iced conductor under the action of crosswind. Once this phenomenon occurs, it will not only cause the conductor to contact each other, trip, damage the insulator, and fatigue the fittings, and in severe cases, it may even cause the tower to collapse, causing a large-scale power outage accident, which is one of the important disaster risks in the operation of the power system.

[0003] The physical cause of galloping is relatively clear. It is mainly caused by the non-symmetry of the cross section of the conductor caused by icing, and the unstable aerodynamic force formed under the action of lateral wind, thereby inducing structural self-excited vibration. The classic Den Hartog criterion points out that when the derivative of the lift force with respect to the angle of attack is negative and the drag coefficient is large, the system will have negative damping, thereby causing galloping to occur. This theory has a wide influence in wind engineering and aeroelasticity research and is considered to be the core criterion for the occurrence of galloping. However, the Den Hartog criterion is difficult to apply directly in actual engineering, and its key parameters such as the lift coefficient and the derivative of the angle of attack are usually obtained by wind tunnel test or CFD simulation, which is difficult to obtain in real time through conventional monitoring means, so the galloping judgment is mainly based on experience or subjective analysis, and there is a lack of unified, quantifiable and generalizable technical path.

[0004] In traditional practice, the galloping early warning method mainly relies on the combination of meteorological condition threshold judgment and manual inspection, such as setting the experience threshold of wind speed, temperature, humidity and other variables for pre-judgment. This method is highly dependent on local experience and is difficult to adapt to widespread deployment in multiple regions and under multiple weather conditions, and cannot effectively depict the coupling relationship between different weather factors. In recent years, with the improvement of weather monitoring and intelligent operation level, data-driven methods such as deep learning have begun to be introduced into the field of galloping prediction. Such methods construct prediction models (such as LSTM, Transformer, etc.) based on wind speed, wind direction, temperature, humidity, precipitation and other multivariate weather sequences, which to some extent improve the recognition accuracy and can realize automatic updating and batch deployment of the model.

[0005] However, the pure data-driven method has obvious limitations. On the one hand, the conductor galloping sample is extremely scarce in the monitoring data, often several orders of magnitude lower than the non-galloping sample, which easily leads to model bias to the main class and is difficult to identify the real galloping event. On the other hand, such method often ignores the physical formation mechanism of galloping, and the model has poor interpretability, which is difficult to make clear physical explanation to the results, and limits the credibility and practicality in engineering scenarios such as dispatching control and emergency response. SUMMARY

[0006] Therefore, the present application provides a power line galloping early warning method and device, computer equipment and storage medium, to solve the problem that the existing power line galloping early warning model has poor interpretability and cannot be applied to large-scale, multi-scenario power line galloping risk identification tasks.

[0007] In a first aspect, the present application provides a power line galloping early warning method, which comprises:

[0008] acquiring meteorological monitoring data covering the power line area;

[0009] extracting galloping physical criterion features based on the meteorological monitoring data;

[0010] training a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power line galloping early warning model;

[0011] inputting actual meteorological monitoring data into the power line galloping early warning model to obtain an original galloping prediction result, and warning the power line galloping risk based on the original galloping prediction result.

[0012] The application provides a power transmission line galloping early warning method, which comprises the following steps: obtaining meteorological monitoring data covering a power transmission line area, ensuring that the data is directly related to the actual environment of the power transmission line, and avoiding the problem that generalized meteorological data is disconnected with the actual working condition of the line; meanwhile, galloping physical criterion features are extracted based on the meteorological detection data, the core physical influencing factors of galloping are focused on, key features are screened from the principle level, irrelevant data interference is reduced, high-quality input is provided for subsequent model training, and the accuracy of the early warning result is improved. The deep learning model is jointly trained by using the meteorological monitoring data and the galloping physical criterion features, the data-driven mode learning ability and the interpretability of the physical mechanism are considered, the data noise fitting or overfitting problem that may occur in the pure data-driven model is avoided, the generalization ability of the early warning model obtained by training is stronger under different lines and different meteorological conditions, and the reliability is higher. By inputting the actual meteorological data into the early warning model, the original galloping prediction result can be obtained in advance, and the galloping risk is early warned. This process changes the traditional post-disposal into pre-warning, sufficient time is obtained for the operation and maintenance personnel to take prevention and control measures, the accident risks such as line tripping, wire breaking and tower damage caused by galloping are effectively reduced, the safe and stable operation of the power transmission line is ensured, the method is suitable for the galloping risk identification task of the power transmission line in a wide range and multiple scenes, and the problem that the existing galloping early warning model of the power transmission line has poor interpretability and cannot be applied to the galloping risk identification task of the power transmission line in a wide range and multiple scenes is solved.

[0013] In an optional implementation, before the galloping physical criterion features are extracted based on the meteorological monitoring data, the method further comprises the following steps:

[0014] The meteorological monitoring data is time-aligned, fused, missing value-filled, abnormal value-excluded and normalized, and standard meteorological feature data is obtained.

[0015] The galloping early warning method provided by the application can ensure the consistency, integrity and accuracy of the meteorological data through time alignment, missing value filling, abnormal value exclusion and normalization of data preprocessing, unify the feature dimension, provide a high-quality data basis for subsequent physical criterion feature extraction and model training, reduce the interference of data noise and deviation on the early warning result, and improve the early warning reliability.

[0016] In an optional implementation, the meteorological monitoring data comprises wind speed, wind direction, line direction and meteorological factors.

[0017] The galloping physical criterion features are extracted based on the meteorological monitoring data, and the method comprises the following steps:

[0018] The wind yaw angle is extracted based on the wind direction and the line direction.

[0019] The wind speed is decomposed to obtain a horizontal wind speed vector.

[0020] constructing an icing coefficient based on meteorological factors;

[0021] constructing a comprehensive galloping coefficient based on the wind yaw angle, the transverse wind speed vector and the icing coefficient;

[0022] taking the wind yaw angle, the transverse wind speed component, the icing coefficient and the comprehensive galloping coefficient as the galloping physical criterion features.

[0023] The galloping early warning method for the power transmission line provided by the application can convert the original meteorological data into core features that conform to the galloping mechanism by extracting the wind yaw angle, the transverse wind speed component, the icing coefficient and the comprehensive galloping coefficient as physical criterion features: the wind yaw angle and the transverse wind speed component accurately capture the direction and intensity of the wind force that induces galloping, the icing coefficient focuses on low-temperature icing as a key inducement, and the comprehensive galloping coefficient realizes the collaborative quantification of multiple physical factors. These features not only strengthen the relevance of data and galloping risk and reflect the physical inducements of galloping occurrence, but also provide the model with interpretable physical basis, reduce redundant information interference, and effectively improve the feature learning efficiency and early warning accuracy of the subsequent early warning model.

[0024] In an optional implementation, the preset deep learning model includes a first sub-channel, a second sub-channel, a double-layer fully connected network corresponding to the first sub-channel, an independent fully connected network corresponding to the second sub-channel, a fusion layer, a shared feature learning channel, a multilayer perceptron and a Dropout regularization layer.

[0025] The preset deep learning model is trained based on the meteorological monitoring data and the galloping physical criterion features to obtain the galloping early warning model for the power transmission line, including:

[0026] The standardized meteorological feature data is input into the first sub-channel and high-dimensional meteorological features are extracted through the corresponding double-layer fully connected network;

[0027] The galloping physical criterion features are input into the second sub-channel and galloping inducement expression features are extracted through the corresponding independent fully connected network;

[0028] The high-dimensional meteorological features and the galloping inducement expression features are subjected to feature fusion through the fusion layer;

[0029] The fused features are input into the shared feature learning channel for learning, and then output the original galloping prediction result after being processed by the multilayer perceptron and the Dropout regularization layer, and the galloping early warning model for the power transmission line is obtained.

[0030] In an optional implementation, the preset deep learning model is trained based on the meteorological monitoring data and the galloping physical criterion features to obtain the galloping early warning model for the power transmission line, further including:

[0031] The preset deep learning model is trained by taking a weighted binary cross-entropy loss function as an optimization objective function, and the formula of the weighted binary cross-entropy loss function is as follows:

[0032] ;

[0033] ;

[0034] Wherein, N is the total number of sample of meteorological feature data and dance physical criterion feature, 0 and 1 respectively, representing not dancing and dancing respectively; is a weight coefficient; is the number of samples of the category .

[0035] The power line dance early warning method provided by the application, in the aspect of model structure design, proposes a double-channel deep neural network structure fusing physical criteria and meteorological time series characteristics. The network includes two independent input channels, one of which is used to receive physical indicators based on the dance theory, and the other is used to process the original meteorological variable time series. Two channels respectively extract high-level semantic features through independent neural network structures, and form a unified joint feature representation in the middle, and then pass through the subsequent shared fully connected layer to predict the dance probability. This structure maintains the deep learning modeling capability, effectively fuses the physical prior information, improves the generalization and interpretability of the model, and is suitable for large-scale, multi-scenario power line dance risk identification tasks.

[0036] In an optional implementation, after obtaining the original dance prediction result, the method further includes:

[0037] A correction function is constructed based on the comprehensive dance coefficient, and the original dance prediction result is corrected based on the correction function to obtain a final dance prediction result; the correction function is represented by the following formula:

[0038] ;

[0039] Wherein, is the original dance prediction result, is the final dance prediction result, is the comprehensive dance coefficient, is a Sigmoid activation function, so that the value range is (0, 1), is a preset correction coefficient.

[0040] The application provides a power transmission line galloping early warning method, considers possible misjudgment and insufficient credibility of a deep learning model under extremely unbalanced samples, proposes a prediction result correction mechanism based on a galloping criterion coefficient, and uses the constructed galloping criterion coefficient (such as an index aggregation of a wind yaw angle, a crosswind speed, and a precipitation form) as an external physical reference to post-process a probability value output by the model. Specifically, the model output value is fused with the galloping coefficient through a nonlinear mapping function to form a final corrected galloping probability, thereby significantly enhancing the identification sensitivity of the model near the boundary value, and providing compensatory guidance for the case of low confidence but high physical risk. As a lightweight structure decoupled from the original model, the mechanism not only improves the reliability of the early warning model, but also enhances the physical interpretability of the result and the auxiliary value of manual review.

[0041] In an optional embodiment, the power transmission line galloping risk is early warned based on the galloping prediction result, comprising:

[0042] The final galloping prediction result is compared with a preset threshold value, if the final galloping prediction result is greater than or equal to the preset threshold value, it is determined that there is a power transmission line galloping risk, and if the final galloping prediction result is less than the preset threshold value, it is determined that there is no power transmission line galloping risk.

[0043] The power transmission line galloping early warning method provided by the application can convert the abstract prediction value output by the model into an intuitive and clear risk conclusion (“there is a risk” or “no risk”) through the comparison of the preset threshold value and the final prediction result, thereby providing clear decision basis for the operation and maintenance personnel. The standardized determination method ensures the consistency and operability of the early warning result, avoids ambiguous interpretation, facilitates quick identification of high-risk lines and timely initiation of prevention and control measures, and improves the practicality and response efficiency of the galloping risk early warning.

[0044] In a second aspect, the application provides a power transmission line galloping early warning device, which comprises:

[0045] A meteorological monitoring data acquisition module is configured to acquire meteorological monitoring data covering a power transmission line area.

[0046] A galloping physical criterion feature extraction module is configured to extract galloping physical criterion features based on the meteorological monitoring data.

[0047] A model training module is configured to train a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features, thereby obtaining a power transmission line galloping early warning model.

[0048] A galloping risk early warning module is configured to input actual meteorological monitoring data into the power transmission line galloping early warning model to obtain an original galloping prediction result, and early warn the power transmission line galloping risk based on the original galloping prediction result.

[0049] In a third aspect, the present application provides a computer device, comprising a memory and a processor, which are connected with each other in communication, and the memory stores computer instructions, and the processor executes the power line galloping early warning method of the first aspect or any of the corresponding embodiments thereof by executing the computer instructions.

[0050] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions for making a computer execute the power line galloping early warning method of the first aspect or any of the corresponding embodiments thereof.

[0051] In a fifth aspect, the present application provides a computer program product, which comprises computer instructions for making a computer execute the power line galloping early warning method of the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0053] Figure 1 is a flowchart of the power line galloping early warning method according to an embodiment of the present application;

[0054] Figure 2 is a flowchart of another power line galloping early warning method according to an embodiment of the present application;

[0055] Figure 3 is a flowchart of still another power line galloping early warning method according to an embodiment of the present application;

[0056] Figure 4 is a structural block diagram of the power line galloping early warning device according to an embodiment of the present application;

[0057] Figure 5 is a hardware structure schematic diagram of the computer device of an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0059] Currently, part of the research begins to try to introduce physical mechanism and priori knowledge to assist deep learning modeling, and converts the wind yaw angle, wind speed component, icing coefficient and the like in the dancing formation process into physical criterion features, so as to enhance the recognition ability of the model to the dancing boundary condition. This kind of method extracts priori information with physical meaning from meteorological data, constructs derived features such as "dancing coefficient", and uses them as model input or post-processing basis, which has shown good development prospects. However, at present, there is still a lack of a systematic and general dancing physical criterion and deep learning model fusion mechanism, especially in the aspects of feature engineering, structure design, sample imbalance processing and post-processing correction, there is still a lot of room for improvement. Therefore, it is urgent to propose a dancing warning technical solution fusing physical mechanism and deep neural network, to enhance the discrimination ability and interpretability of the model, to adapt to the actual application requirements under complex meteorological conditions. The embodiments of the present application provide a power transmission line dancing warning method, which forms a dancing warning technical solution by fusing physical mechanism and deep neural network, enhances the discrimination ability and interpretability of the model, and adapts to the actual application requirements under complex meteorological conditions.

[0060] According to the embodiments of the present application, a power transmission line dancing warning method embodiment is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0061] In the present embodiment, a power transmission line dancing warning method is provided, which can be used in the power transmission line dancing warning device described above, Figure 1 The flowchart of the power transmission line dancing warning method according to the embodiments of the present application is shown in FIG. 1, which includes the following steps: Figure 1 As shown in the figure, the flowchart includes the following steps:

[0062] In step S101, meteorological monitoring data covering the power transmission line area is obtained.

[0063] Specifically, the meteorological monitoring data covering the power transmission line area is obtained to provide high-quality input feature data for subsequent construction of dancing physical criterion features and deep learning prediction model, for example, the meteorological monitoring data involves core meteorological data such as wind speed, wind direction, line direction, temperature, humidity and precipitation.

[0064] Install ground meteorological stations, automatic meteorological monitoring terminals and other equipment along the key sections of the transmission line (such as areas prone to dancing, near towers), and collect basic meteorological parameters such as wind speed, wind direction, temperature, humidity, and precipitation in real time. In combination with the public data of regional meteorological departments (such as meteorological station network data, satellite remote sensing data, and radar monitoring data), supplement the large-scale meteorological information around the line, and expand the coverage.

[0065] Real-time data from each monitoring point can also be transmitted to the data center through wireless communication (such as 4G / 5G, LoRa) or wired network, while integrating historical meteorological databases to form a complete set of meteorological data covering the line area, providing raw data support for subsequent analysis.

[0066] This embodiment selects meteorological monitoring data covering the transmission line area in the past five years, and the selected core meteorological variables are shown in Table 1:

[0067] Table 1 Core Meteorological Variables

[0068]

[0069] Step S102, extracting dancing physical criterion features based on meteorological monitoring data.

[0070] Specifically, the dancing physical criterion features refer to the key physical indicators that can quantitatively reflect the possibility of dancing occurrence extracted from meteorological monitoring data based on the dancing mechanism of the transmission line. These features are directly related to the core causes of dancing occurrence (such as wind action, icing state, etc.), and are the physical basis for judging whether the line has dancing risk.

[0071] Transmission line dancing is a low-frequency, large-amplitude vibration phenomenon caused by aerodynamic instability under certain meteorological conditions (such as wind, icing, etc.), and its occurrence is directly related to physical factors such as wind direction and strength, line icing state, etc. Dancing physical criterion features are exactly the conversion of these abstract physical influences into quantifiable indicators, accurately capturing the key conditions for dancing occurrence.

[0072] Step S103, training a preset deep learning model based on meteorological monitoring data and dancing physical criterion features to obtain a transmission line dancing early warning model.

[0073] In particular, based on the meteorological monitoring data and the galloping physical criterion characteristics, a deep learning model for predicting the galloping risk of the power transmission line is designed and trained. The model adopts a double-channel structure, respectively processes the original meteorological monitoring data input and the galloping physical criterion characteristic input, and realizes joint feature learning through a fusion layer to improve the galloping discrimination ability. While maintaining the deep learning modeling ability, the structure effectively fuses the physical prior information, improves the generalization and interpretability of the model, and is suitable for the galloping risk identification task of the power transmission line in a large range and multiple scenes.

[0074] In step S104, the actual meteorological monitoring data is input into the power transmission line galloping early warning model to obtain an original galloping prediction result, and the power transmission line galloping risk is early warned based on the original galloping prediction result.

[0075] In particular, according to the comparison of the prediction result and the set threshold, the galloping risk determination result of the power transmission line is output.

[0076] The power transmission line galloping early warning method provided in the embodiment acquires meteorological monitoring data covering the area of the power transmission line, ensures that the data is directly related to the actual environment where the power transmission line is located, and avoids the problem that the generalized meteorological data is disconnected with the actual working condition of the line. Meanwhile, the galloping physical criterion characteristics are extracted based on the meteorological detection data, the core physical influencing factors of galloping are focused on, the key characteristics are selected from the principle level, irrelevant data interference is reduced, high-quality input is provided for subsequent model training, and the accuracy of the early warning result is improved. The deep learning model is trained by combining meteorological monitoring data and galloping physical criterion characteristics, the model learning ability of data-driven mode and the interpretability of physical mechanism are considered, the problems such as data noise fitting or overfitting that may occur in the pure data-driven model are avoided, the generalization ability of the early warning model obtained by training is stronger under different lines and different weather conditions, and the reliability is higher. By inputting the actual meteorological data into the early warning model, the original galloping prediction result can be obtained in advance, and the galloping risk is early warned based on this. This process changes the traditional after-treatment into early warning, can give the operation and maintenance personnel sufficient time to take preventive measures, effectively reduces the accident risks such as line trip-out, wire breakage, and tower damage caused by galloping, guarantees the safe and stable operation of the power transmission line, is suitable for the galloping risk identification task of the power transmission line in a large range and multiple scenes, and solves the problem that the existing galloping early warning model of the power transmission line has poor interpretability and cannot be applied to the galloping risk identification task of the power transmission line in a large range and multiple scenes.

[0077] In the embodiment, a power transmission line galloping early warning method is provided, which can be used for a power transmission line galloping early warning device, Figure 2 is a flowchart of the power transmission line galloping early warning method according to the embodiment of the present application, as Figure 2 shown, the flowchart includes the following steps:

[0078] Step S201, obtain meteorological monitoring data covering the area of the power transmission line. For details, please refer to Figure 1 Step S101 of the embodiment shown, which will not be described here.

[0079] Step S202, data preprocessing of time alignment, missing value filling, abnormal value elimination and normalization of meteorological monitoring data, to obtain standardized meteorological feature data.

[0080] Specifically, the above step S202 includes:

[0081] 1) Time alignment and data integration:

[0082] Because the sampling frequency and time stamp of different data sources may be different, time alignment processing needs to be unified. Time alignment and data integration include: resampling all meteorological monitoring data to daily scale; keep daily timing data (such as 9:00); when merging multi-site data, prefer to use the latest site data. For each power transmission line, construct a time series format meteorological monitoring data sample, as shown in Table 2:

[0083] Table 2 Time series format meteorological monitoring data sample

[0084]

[0085] 2) Missing value filling: missing measurement often occurs in actual sampling of meteorological monitoring data. The missing value filling strategy is as follows: if single-day data is missing, use the previous day's data to fill (forward filling); if the entire segment is missing for more than 3 consecutive days, directly eliminate the segment record; for nonlinear variables such as precipitation and wind speed, use interpolation method to fill.

[0086] 3) Abnormal value elimination: use rule threshold and statistical method to double judge abnormality: wind speed exceeding 60 m / s, temperature below -50°C or above 50°C, humidity being 0 or exceeding 100 are considered as abnormal values and are eliminated.

[0087] (1);

[0088] At the same time, use 3σ principle to eliminate extreme values If it is satisfied, it is considered as an extreme abnormal value and is eliminated, and the formula is as follows:

[0089] (2);

[0090] Wherein: is the average value of all meteorological monitoring data samples; σ is the standard deviation of the sample, indicating the dispersion degree of the data, is the meteorological monitoring data sample.

[0091] 4) Data normalization processing: in order to improve the training efficiency and convergence speed of the deep learning model, the continuous variable needs to be normalized. The normalization method is as follows:

[0092] (3);

[0093] wherein, is the normalized data, , respectively, the minimum and maximum values in the meteorological monitoring data sample.

[0094] All normalization parameters are calculated on the training set, and the same parameters are used for standardization on the test set to ensure consistency.

[0095] Step S203, extracting the dance physical criterion feature based on the meteorological monitoring data.

[0096] Specifically, the meteorological monitoring data includes wind speed, wind direction, line direction and meteorological factors, wherein the meteorological factors include temperature, humidity and precipitation, based on the physical mechanism of the formation of the power line dance, combined with wind speed, wind direction and meteorological factors, a plurality of quantifiable dance physical criterion features are constructed, which are used to enhance the understanding ability of the deep model to the actual dance mechanism. The above step S203 includes:

[0097] Step S2031, extracting the wind deflection angle based on the wind direction and the line direction.

[0098] Specifically, the wind deflection angle is the angle between the wind direction and the line direction, which is one of the basic physical quantities for judging whether the dance occurs. Dance often occurs under the action of transverse wind, so the angle feature needs to be extracted, and the formula is as follows:

[0099] (4);

[0100] wherein, is the wind deflection angle, is the wind direction, is the line direction, is the modulo operation, that is, the remainder operation.

[0101] Step S2032, decomposing the wind speed to obtain the transverse wind speed vector.

[0102] Specifically, since only the wind force component perpendicular to the conductor direction can cause dance, the wind speed needs to be decomposed, and only the transverse wind speed component is retained, and the formula is as follows:

[0103] (5);

[0104] wherein, is the transverse wind speed component, is the wind speed.

[0105] Step S2033, constructing icing coefficient based on meteorological factors. The icing coefficient is constructed based on temperature, humidity and precipitation.

[0106] Specifically, since icing is an important condition for inducing galloping, the embodiment of the present application constructs an icing coefficient considering temperature, humidity and precipitation. The coefficient is only non-zero at low temperature (such as <0°C), and the specific formula is as follows:

[0107] (6);

[0108] wherein, is the precipitation, is the relative humidity, is the temperature.

[0109] Step S2034, constructing comprehensive galloping coefficient based on wind yaw angle, transverse wind speed vector and icing coefficient.

[0110] Specifically, the present application fuses multiple physical factors to construct a comprehensive galloping coefficient for determining the possibility of galloping, as an auxiliary channel input or post-processing correction index of the subsequent deep learning model.

[0111] The formula of the comprehensive galloping coefficient is as follows:

[0112] (7);

[0113] wherein, and are the galloping weight coefficients set according to experience.

[0114] Step S2035, taking the wind yaw angle, transverse wind speed component, icing coefficient and comprehensive galloping coefficient as the galloping physical criterion features.

[0115] It should be noted that after the extraction of the galloping physical criterion features, they need to be added as new features to each meteorological monitoring data sample record to form fusion new features for subsequent direct use or independent analysis of the deep learning model, as shown in Table 3 below:

[0116] Table 3 Fusion new features

[0117]

[0118] Step S204, training the preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain the transmission line galloping early warning model. For details, please refer to the step S103 of the embodiment shown in Figure 1 , which will not be repeated here.

[0119] Step S205, input the actual meteorological monitoring data into the power transmission line galloping early warning model to obtain an original galloping prediction result, and perform early warning on the power transmission line galloping risk based on the original galloping prediction result. For details, please refer to Figure 1 Step S104 of the embodiment shown in FIG. 1 is not repeated here.

[0120] The power transmission line galloping early warning method provided in this embodiment can convert the original meteorological data into core features that conform to the galloping mechanism by extracting the wind yaw angle, the transverse wind speed component, the icing coefficient and the comprehensive galloping coefficient as physical criterion features. The wind yaw angle and the transverse wind speed component accurately capture the direction and intensity of the wind force that induces galloping, the icing coefficient focuses on low-temperature icing as a key inducement, and the comprehensive galloping coefficient realizes the collaborative quantification of multiple physical factors. These features not only strengthen the relevance of data and galloping risk and reflect the physical inducements of galloping, but also provide an interpretable physical basis for the model, reduce redundant information interference, and effectively improve the feature learning efficiency and early warning accuracy of the subsequent early warning model.

[0121] In this embodiment, a power transmission line galloping early warning method is provided, which can be used for a power transmission line galloping early warning device, Figure 3 is a flowchart of the power transmission line galloping early warning method according to an embodiment of the present application, as shown in Figure 3 The flowchart includes the following steps:

[0122] Step S301, obtain meteorological monitoring data covering the power transmission line area. For details, please refer to Figure 2 Step S201 of the embodiment shown in FIG. 1 is not repeated here.

[0123] Step S302, extract galloping physical criterion features based on the meteorological monitoring data. For details, please refer to Figure 2 Step S203 of the embodiment shown in FIG. 1 is not repeated here.

[0124] Step S303, train a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power transmission line galloping early warning model.

[0125] Specifically, based on the preprocessed standard meteorological feature data and the galloping physical criterion features, a deep learning model for predicting the power transmission line galloping risk is designed and trained. The model adopts a double-channel structure, processes the original meteorological input and the physical criterion input respectively, and realizes joint feature learning through a fusion layer to improve the galloping discrimination ability. The preset deep learning model includes a first sub-channel, a second sub-channel, a double-layer fully connected network corresponding to the first sub-channel, an independent fully connected network corresponding to the second sub-channel, a fusion layer, a shared feature learning channel, a multilayer perceptron and a Dropout regularization layer.

[0126] 1) Model input design:

[0127] The input of the model is divided into two sub-channels:

[0128] The first sub-channel: original meteorological data: wind speed, wind direction, line direction, temperature, humidity, precipitation;

[0129] The second sub-channel: wind yaw angle, transverse wind speed, icing coefficient, and dancing coefficient.

[0130] 2) Model structure design: In order to fully integrate the original meteorological features and physical criterion features, a double-channel neural network structure is constructed for dancing event identification.

[0131] One sub-channel is used to receive physical indicators constructed based on dancing theory, and the other sub-channel is used to process original time series of meteorological variables. Two sub-channels respectively extract high-level semantic features through independent neural network structures, and form a unified joint feature representation in the middle, and then through subsequent shared fully connected layers to predict the dancing probability. This structure maintains the ability of deep learning modeling, effectively integrates physical prior information, improves the generalization and interpretability of the model, and is suitable for large-scale, multi-scenario transmission line dancing risk identification tasks.

[0132] The above step S303 includes:

[0133] Step S3031, the standardized meteorological feature data is input into the first sub-channel, and high-dimensional meteorological features are extracted through the corresponding double-layer fully connected network.

[0134] Specifically, the first sub-channel takes wind speed, wind direction, line direction, temperature, humidity, precipitation and other conventional meteorological variables as input, and extracts high-dimensional features through two-layer fully connected network.

[0135] Step S3032, the dancing physical criterion features are input into the second sub-channel, and the dancing cause expression features are extracted through the corresponding independent fully connected network.

[0136] Specifically, the second sub-channel takes wind yaw angle, transverse wind speed, icing coefficient, and dancing coefficient as input, and extracts dancing cause expression features at the physical level through an independent fully connected network.

[0137] Step S3033, the high-dimensional meteorological features and dancing cause expression features are fused through the fusion layer.

[0138] Specifically, the output features of the two sub-channels are fused through vector splicing, and the specific fusion method can be obtained by referring to the related technology, which will not be described here.

[0139] Step S3034, input the fused features into the shared feature learning channel for learning, and output the original galloping prediction result after processing by the multi-layer perceptron and the Dropout regularization layer, and obtain the transmission line galloping early warning model.

[0140] Specifically, after the fused features are input into the shared feature learning channel and further processed by the multi-layer perceptron and the Dropout regularization layer, a probability value of galloping occurrence, i.e., the original galloping prediction result, is output, and the corresponding deep learning model is converted into the transmission line galloping early warning model.

[0141] Step S3035, the weighted binary cross-entropy loss function is used as the optimization objective function to train the preset deep learning model.

[0142] Specifically, the model uses the Sigmoid activation function for binary classification, the loss function uses the class weighted binary cross-entropy, and a sample imbalance processing mechanism is introduced to improve the sensitivity and recognition ability to the minority class (galloping sample). Specifically, it includes:

[0143] Considering that the galloping event belongs to a typical extreme minority class problem, the data presents a high imbalance (the galloping sample is much less than the non-galloping sample), therefore, the method uses the weighted binary cross-entropy loss function (Weighted Binary Cross Entropy) as the optimization objective function. By assigning a higher loss weight to the galloping class (positive class), the bias of the network to the majority class (non-galloping) is effectively suppressed, and the recognition ability to the minority class is improved. The weight ratio can be automatically calculated according to the total number of class samples, or it can be set and adjusted according to actual business needs. In addition, to prevent numerical instability and gradient explosion problems during training, the network uses the Sigmoid activation function to output the galloping probability, and dynamically adjusts the learning rate and regularization coefficient during training to improve the generalization ability and convergence stability of the model. The weight coefficient formula is represented as:

[0144] (8);

[0145] The formula of the weighted binary cross-entropy loss function constructed according to the weight coefficient is represented as:

[0146] (9);

[0147] Wherein, N is the total number of sample of meteorological feature data and galloping physical criterion features, 0 and 1 respectively represent non-galloping and galloping; is the weight coefficient; is the number of samples of class .

[0148] Step S304, a correction function is constructed based on the comprehensive galloping coefficient, and the original galloping prediction result is corrected based on the correction function to obtain a final galloping prediction result.

[0149] Specifically, although the deep learning model can realize galloping prediction by learning complex nonlinear relationships, due to factors such as the size of the training sample, class imbalance, meteorological uncertainty, etc., the model may miss reports or have low confidence prediction. To alleviate this problem, the galloping criterion coefficient constructed in the foregoing is used as an auxiliary physical quantity, a correction function is introduced to adjust the model output probability twice to form a corrected output. The correction function is constructed using a coefficient smoothing mechanism, combining the original output of the model and the galloping coefficient value, and the correction function is expressed by the following formula:

[0150] (10);

[0151] wherein, is the original galloping prediction result (between 0 and 1), is the final galloping prediction result, is the comprehensive galloping coefficient, is a Sigmoid activation function, so that the value range is (0, 1), is a preset correction coefficient, which controls the enhancement amplitude of the physical criterion on the model result (β can be set to 0.3).

[0152] Step S305, input the actual meteorological monitoring data into the galloping early warning model of the transmission line to obtain the original galloping prediction result, and based on the original galloping prediction result, the galloping risk of the transmission line is early warned.

[0153] Specifically, the corrected prediction result is used for the final galloping risk determination, and the above step S305 includes:

[0154] Step a, compare the final galloping prediction result with a preset threshold value, if the final galloping prediction result is greater than or equal to the preset threshold value, it is determined that there is a galloping risk of the transmission line; if the final galloping prediction result is less than the preset threshold value, it is determined that there is no galloping risk of the transmission line, and the formula of the galloping risk determination is:

[0155] (11);

[0156] wherein, is a preset threshold value, which is set according to the actual situation, and is not specifically limited here.

[0157] The power transmission line galloping early warning method provided by the embodiment considers the misjudgment and insufficient credibility of the deep learning model under extremely unbalanced samples, and proposes a prediction result correction mechanism based on the galloping criterion coefficient. The mechanism uses the constructed galloping criterion coefficient (such as the wind deflection angle, crosswind speed, and precipitation form) as an external physical reference to post-process the probability value of the model prediction output. The specific method is to fuse the model output value and the galloping coefficient through a nonlinear mapping function to form the final corrected galloping probability, which significantly enhances the identification sensitivity of the model near the boundary value, especially for cases with low confidence but high physical risk. This mechanism, as a lightweight structure decoupled from the original model, not only improves the reliability of the early warning model, but also enhances the physical interpretability of the results and the auxiliary value of manual review. By comparing the preset threshold with the final prediction result to determine the risk, the abstract prediction value output by the model can be converted into an intuitive and explicit risk conclusion ("risk exists" or "no risk"), providing clear decision-making basis for operation and maintenance personnel. This standardized determination method ensures the consistency and operability of the early warning results, avoids ambiguous interpretation, facilitates quick identification of high-risk lines and timely initiation of prevention and control measures, and improves the practicality and response efficiency of galloping risk early warning.

[0158] As one or more specific application embodiments of the embodiment of the present application, the power transmission line galloping early warning method provided by the present application is further described in detail as follows:

[0159] Step 1, collection and preprocessing of meteorological data:

[0160] This step aims to provide high-quality input feature data for subsequent construction of galloping coefficient criterion and deep learning prediction model. Specifically, it includes data source determination, variable selection, time alignment, missing value filling, outlier removal, and normalization processing.

[0161] 1) Select meteorological monitoring data covering the power transmission line area in the past five years. The core meteorological variables are shown in Table 1 above.

[0162] 2) Time alignment and data integration: Since the sampling frequency and time stamp of different data sources may be different, time alignment processing is needed. Methods include: resampling all data to daily scale; retain daily timed data (such as 9:00); when merging multi-site data, prefer to use the latest site data. For each power transmission line, construct sample data in time series format, as shown in Table 2 above.

[0163] 3) Missing value filling: Missing value filling strategy is as follows: If single-day data is missing, use the previous day's data to fill in (forward filling); If the missing data is more than 3 consecutive days, directly remove the record; For nonlinear variables such as precipitation and wind speed, interpolation method is used to fill in.

[0164] 4) Dual judgment of abnormality by rule threshold and statistical method: Data with wind speed exceeding 60 m / s, temperature below -50°C or above 50°C, humidity of 0 or more than 100 are considered as abnormal values and removed.

[0165] (1);

[0166] At the same time, 3σ principle is used to remove extreme values If it is satisfied, it is considered as extreme abnormal value and removed, the formula is as follows:

[0167] (2);

[0168] Among them: is the average value of all meteorological monitoring data samples; σ is the standard deviation of the sample, indicating the dispersion of the data, is the meteorological monitoring data sample.

[0169] 5) Data normalization: In order to improve the training efficiency and convergence speed of deep learning model, continuous variables need to be normalized. The normalization method is as follows:

[0170] (3);

[0171] Among them, is the normalized data, , are the minimum and maximum values in the meteorological monitoring data sample, respectively.

[0172] Step 2: Extraction and construction of dance physical criterion features: Based on the physical mechanism of transmission line dance, combined with wind speed, wind direction and meteorological factors, multiple quantifiable dance physical criterion features are constructed to enhance the understanding ability of deep model to actual dance mechanism.

[0173] 1) Wind yaw angle: Wind yaw angle is the angle between wind direction and transmission line direction, which is one of the basic physical quantities to judge whether dance occurs. Dance often occurs under the action of transverse wind, so this angle feature needs to be extracted, the formula is as follows:

[0174] (4);

[0175] Among them, is the wind yaw angle, is the wind direction, For line orientation, For modulo operation, that is, remainder operation.

[0176] 2) Transverse wind speed component: Since only the wind force component perpendicular to the conductor direction can cause galloping, the wind speed needs to be decomposed, and only the transverse wind speed component is retained, which is expressed as follows:

[0177] (5);

[0178] wherein, is the transverse wind speed component, is the wind speed.

[0179] 3) Icing coefficient: Since icing is an important condition for inducing galloping, the embodiment of the present application considers the icing coefficient of temperature, humidity, and precipitation. The coefficient is only non-zero at low temperature (such as <0°C), and the specific formula is as follows:

[0180] (6);

[0181] wherein, is the precipitation, is the relative humidity, is the temperature.

[0182] 4) Galloping coefficient: The present application integrates multiple physical factors to construct a comprehensive galloping coefficient , which is used to determine the possibility of galloping and serves as an auxiliary channel input or post-processing correction index for the subsequent deep learning model.

[0183] The formula of the comprehensive galloping coefficient is as follows:

[0184] (7);

[0185] wherein, and are the galloping weight coefficients set according to experience.

[0186] 5) Sample construction of fused new features: After completing the criterion construction, it needs to be added as new features to each processed meteorological monitoring data sample record for direct use by the model or independent analysis. The fused new features are shown in Table 3 above.

[0187] Step 3: Deep learning model construction and training with fused physical criteria:

[0188] Based on the pre-processed weather features and physical criterion features, a deep learning model for predicting the galloping risk of power transmission lines is designed and trained. The model adopts a dual-channel structure, which processes the original weather input and physical criterion input respectively, and realizes joint feature learning through the fusion layer to improve the galloping discrimination ability.

[0189] 1) Model input design:

[0190] The input of the model is divided into two sub-channels:

[0191] Channel 1: original weather data: wind speed, wind direction, line orientation, temperature, humidity, precipitation;

[0192] Channel 2: wind yaw angle, lateral wind speed, icing coefficient, galloping coefficient;

[0193] 2) Model structure design: In order to fully fuse the original weather features and physical criterion features, a dual-channel neural network structure is constructed for galloping event recognition. One channel takes wind speed, wind direction, line orientation, temperature, humidity, and precipitation as input, and extracts high-dimensional features through two fully connected networks; the other channel takes wind yaw angle, lateral wind speed, icing coefficient, and galloping coefficient as input, and extracts physical-level galloping inducement expression through an independent fully connected network. The output features of the two channels are fused through vector concatenation, and then enter the shared feature learning channel, which is further processed through multiple perceptrons and Dropout regularization layers, and finally outputs the probability value of galloping occurrence. The model uses Sigmoid activation function for binary classification, and uses class-weighted binary cross-entropy loss function, and introduces sample imbalance processing mechanism to improve the sensitivity and recognition ability of the minority class (galloping samples).

[0194] 3) Loss function: Considering that galloping events belong to typical extreme minority class problems, the data presents high imbalance (galloping samples are much less than non-galloping samples), therefore this method adopts weighted binary cross-entropy loss function (Weighted Binary Cross Entropy) as the optimization objective function. By assigning higher loss weight to the galloping class (positive class), the bias of the network to the majority class (non-galloping) is effectively suppressed, and the recognition ability of the minority class is improved. The weight ratio can be automatically calculated according to the total number of class samples, or it can be set and adjusted according to actual business needs. In addition, to prevent numerical instability and gradient explosion problems during training, the network uses Sigmoid activation function to output the galloping probability, and dynamically adjusts the learning rate and regularization coefficient during training to improve the generalization ability and convergence stability of the model, and the weight coefficient formula is represented as:

[0195] (8);

[0196] The formula of the weighted binary cross-entropy loss function constructed according to the weight coefficient is:

[0197] (9);

[0198] wherein N is the total number of samples of the meteorological feature data and the flutter physical criterion feature, respectively take 0 and 1, respectively representing no flutter and flutter; is the weight coefficient; is the number of samples of the category .

[0199] Step 4: Flutter coefficient guided model correction mechanism: Although the deep learning model can realize flutter prediction by learning complex nonlinear relationships, due to factors such as limited training sample size, class imbalance, meteorological uncertainty, etc., the model may have false negatives or low confidence predictions. To alleviate this problem, the flutter criterion coefficient constructed in the foregoing is used as an auxiliary physical quantity, a correction function is introduced to adjust the model output probability twice, and a corrected output is formed.

[0200] 1) Correction function construction: A coefficient smoothing mechanism is used to construct the correction function, which combines the model original output and the flutter coefficient value. The correction function is represented by the following formula:

[0201] (10);

[0202] wherein is the original flutter prediction result (between 0 and 1), is the final flutter prediction result, is the comprehensive flutter coefficient, is a Sigmoid activation function, so that the value range is (0, 1), is a preset correction coefficient, which controls the enhancement amplitude of the physical criterion on the model result (β can be set to 0.3).

[0203] 2) Flutter determination strategy: The corrected prediction result is used for the final flutter risk determination. The formula of the flutter risk determination is:

[0204] (11);

[0205] wherein is a preset threshold value, which is set according to the actual situation and is not specifically limited here.

[0206] The power transmission line flutter early warning method provided by the embodiment has the following innovations:

[0207] I) Meteorological feature construction method based on flutter physical criterion:

[0208] To address the issue of traditional deep learning models relying on raw meteorological variables and lacking physical mechanism constraints, a feature construction strategy that integrates galloping physical mechanisms is proposed. By introducing typical galloping theories such as the Den Hartog criterion (one of the classic theories explaining the mechanism of transmission line icing galloping), the raw meteorological data such as wind speed, wind direction, temperature, humidity, and precipitation are processed to construct criterion-based feature variables with clear physical meanings, such as wind deflection parameter, lateral wind speed, ice coverage coefficient, and galloping coefficient. These features not only retain the dynamic characteristics of input meteorological variables, but also reflect the physical causes of galloping, providing a more discriminative and interpretable input data basis for subsequent model training.

[0209] II) Dual-channel neural network structure integrating physical indicators and meteorological time series deep representation:

[0210] In terms of model structure design, a dual-channel deep neural network structure integrating physical criteria and meteorological time series features is proposed. The network contains two independent input channels, one of which receives physical indicators constructed based on galloping theory, and the other processes the original meteorological variable time series. The two channels extract high-level semantic features through independent neural network structures and form a unified joint feature representation in the middle, and then pass through subsequent shared fully connected layers for galloping probability prediction. This structure maintains the deep learning modeling capability while effectively integrating physical prior information, improving the model's generalization and interpretability, and is suitable for large-scale, multi-scenario transmission line galloping risk identification tasks.

[0211] III) Galloping coefficient-guided model correction mechanism:

[0212] Considering the potential misjudgment and credibility issues of deep learning models under extremely imbalanced samples, a prediction result correction mechanism based on galloping criterion coefficients is proposed. This mechanism uses the constructed galloping criterion coefficients (such as wind deflection angle, lateral wind speed, and precipitation form) as external physical references to post-process the probability values output by the model. Specifically, the model output value is fused with the galloping coefficient through a nonlinear mapping function to form the final corrected galloping probability, significantly enhancing the identification sensitivity of the model near the boundary value, especially for cases with low confidence but high physical risk. This mechanism, as a decoupled lightweight structure from the original model, not only improves the reliability of the early warning model, but also enhances the physical interpretability of the results and the auxiliary value of manual review.

[0213] A power line galloping early warning device is also provided in the embodiment, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware is also possible and contemplated.

[0214] The embodiment provides a power line galloping early warning device, as shown in the accompanying drawings, comprising: Figure 4

[0215] A meteorological monitoring data acquisition module 401 is configured to acquire meteorological monitoring data covering a power line area.

[0216] A galloping physical criterion feature extraction module 402 is configured to extract galloping physical criterion features based on the meteorological monitoring data.

[0217] A model training module 403 is configured to train a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power line galloping early warning model.

[0218] A galloping risk early warning module 404 is configured to input actual meteorological monitoring data into the power line galloping early warning model to obtain an original galloping prediction result, and early warn a power line galloping risk based on the original galloping prediction result.

[0219] In some optional embodiments, the power line galloping early warning device further comprises:

[0220] A data preprocessing module is configured to perform data preprocessing of time alignment and data fusion, missing value filling, outlier removal, and normalization on the meteorological monitoring data before extracting the galloping physical criterion features based on the meteorological monitoring data to obtain standardized meteorological feature data.

[0221] In some optional embodiments, the meteorological monitoring data comprises wind speed, wind direction, line direction, and meteorological factors; and the galloping physical criterion feature extraction module 402 comprises:

[0222] A wind yaw angle extraction unit is configured to extract a wind yaw angle based on the wind direction and the line direction.

[0223] A wind speed decomposition unit is configured to decompose the wind speed to obtain a horizontal wind speed vector.

[0224] An icing coefficient construction unit is configured to construct an icing coefficient based on the meteorological factors.

[0225] A comprehensive galloping coefficient construction unit is configured to construct a comprehensive galloping coefficient based on the wind yaw angle, the horizontal wind speed vector, and the icing coefficient.

[0226] ​The dancing physical criterion feature determination unit is configured to take the wind yaw angle, the transverse wind speed component, the icing coefficient, and the comprehensive dancing coefficient as dancing physical criterion features.

[0227] In some optional embodiments, the preset deep learning model comprises a first sub-channel, a second sub-channel, a double-layer full connection network corresponding to the first sub-channel, an independent full connection network corresponding to the second sub-channel, a fusion layer, a shared feature learning channel, a multi-layer perceptron, and a Dropout regularization layer; the model training module 403 comprises:

[0228] The first sub-channel processing unit is configured to input the standardized meteorological feature data into the first sub-channel and extract high-dimensional meteorological features through the corresponding double-layer full connection network.

[0229] The second sub-channel processing unit is configured to input the dancing physical criterion features into the second sub-channel and extract dancing cause expression features through the corresponding independent full connection network.

[0230] The feature fusion unit is configured to perform feature fusion on the high-dimensional meteorological features and the dancing cause expression features through the fusion layer.

[0231] The prediction unit is configured to input the fused features into the shared feature learning channel for learning, and then output an original dancing prediction result after processing by the multi-layer perceptron and the Dropout regularization layer, and obtain a power transmission line dancing early warning model.

[0232] In some optional embodiments, the model training module 403 further comprises:

[0233] The loss function determination unit is configured to train the preset deep learning model by using a weighted binary cross-entropy loss function as an optimization objective function, and the formula of the weighted binary cross-entropy loss function is as follows:

[0234] ;

[0235] ;

[0236] wherein N is the total number of samples of the meteorological feature data and the dancing physical criterion features, 0 and 1 respectively, representing no dancing and dancing respectively; is a weight coefficient; is the number of samples of the category .

[0237] In some optional embodiments, the power transmission line dancing early warning device further comprises:

[0238] The correction module is configured to construct a correction function based on the comprehensive dance coefficient after obtaining the original dance prediction result, and correct the original dance prediction result based on the correction function to obtain a final dance prediction result; the correction function is expressed by the following formula:

[0239] ;

[0240] wherein, is the original dance prediction result, is the final dance prediction result, is the comprehensive dance coefficient, is a Sigmoid activation function, and the value range is (0, 1), is a preset correction coefficient.

[0241] In some optional embodiments, the dance risk early warning module 404 comprises:

[0242] The dance risk judgment unit is configured to compare the final dance prediction result with a preset threshold value, and determine that there is a power transmission line dance risk if the final dance prediction result is greater than or equal to the preset threshold value, or determine that there is no power transmission line dance risk if the final dance prediction result is less than the preset threshold value.

[0243] Further function descriptions of the above-mentioned various modules and units are the same as those of the corresponding embodiments, and will not be described here.

[0244] The power transmission line dance early warning device in the embodiment is presented in the form of a functional unit, and the unit herein refers to an ASIC (Application Specific Integrated Circuit, Application Specific Integrated Circuit) circuit, a processor and a memory executing one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0245] The embodiment of the present application also provides a computer device with the above-mentioned Figure 4 power transmission line dance early warning device.

[0246] Please refer to Figure 5 , Figure 5 is a structural schematic diagram of a computer device provided by an optional embodiment of the present application, as Figure 5As shown, the computer device includes one or more processors 10, memory 20, and interfaces 30 for external devices such as a keyboard and a mouse and a disk drive. One or more of the interfaces 30 enable a user to interact with the computer device. In some embodiments, the interface 30 also includes an input device, such as a microphone, or output device, such as a speaker. Figure 5 The processor 10 is used in the description as an example.

[0247] The processor 10 can be a central processing unit, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a general array logic, or any combination thereof.

[0248] The memory 20 stores instructions that can be executed by the at least one processor 10 to cause the at least one processor 10 to perform the methods described in the above embodiments.

[0249] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system, application programs, and the like for use by the at least one processor 10. The data storage area can store data created by the computer device, etc. Additionally, the memory 20 can include a volatile memory, such as a random access memory, and a non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid state storage device. In some embodiments, the memory 20 can optionally include a memory that is remote from the processor 10, such as a network storage device connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and a combination thereof.

[0250] The memory 20 can include a volatile memory, such as a random access memory, and a non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid state storage device.

[0251] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 can be connected by a bus or other means, Figure 5 The bus connection is taken as an example.

[0252] The input device 30 can receive inputted digital or character information, and generate key signal input related to user settings and function control of the computer device, such as touch screen, keypad, mouse, trackpad, touchpad, pointing stick, one or more mouse buttons, trackball, joystick, etc. The output device 40 can include display device, auxiliary lighting device (e.g. LED), and tactile feedback device (e.g. vibration motor), etc. The display device includes but is not limited to liquid crystal display, light emitting diode, display and plasma display. In some alternative embodiments, the display device can be a touch screen.

[0253] The embodiments of the present application also provide a computer readable storage medium, the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium by network downloading of computer code, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor or programmable or special hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the method shown in the above embodiments is implemented.

[0254] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can invoke or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc., and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer executes the corresponding compiled program after compiling the instructions, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0255] While embodiments of the application have been described in connection with the preferred embodiments of the various figures, those of ordinary skill in the art will appreciate that various modifications and changes can be made without departing from the spirit and scope of the application, and that such modifications and changes fall within the scope of the appended claims.

Claims

1. A galloping warning method for a power transmission line, characterized by, The method comprises the following steps: acquiring meteorological monitoring data covering the area of the power transmission line; extracting galloping physical criterion features based on the meteorological monitoring data; training a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power transmission line galloping early warning model; inputting actual meteorological monitoring data into the power transmission line galloping early warning model to obtain an original galloping prediction result, and warning of the galloping risk of the power transmission line based on the original galloping prediction result; before the step of extracting galloping physical criterion features based on the meteorological monitoring data, the method further comprises the following steps: performing data preprocessing on the meteorological monitoring data, including time alignment and data fusion, missing value filling, abnormal value elimination and normalization to obtain standardized meteorological feature data; the meteorological monitoring data comprises wind speed, wind direction, line direction and meteorological factors; the step of extracting galloping physical criterion features based on the meteorological monitoring data comprises the following steps: extracting a wind yaw angle based on the wind direction and the line direction; decomposing the wind speed to obtain a horizontal wind speed vector; constructing an icing coefficient based on meteorological factors; constructing a comprehensive galloping coefficient based on the wind yaw angle, the horizontal wind speed vector and the icing coefficient; taking the wind yaw angle, the horizontal wind speed component, the icing coefficient and the comprehensive galloping coefficient as the galloping physical criterion features; the preset deep learning model comprises a first sub-channel, a second sub-channel, a double-layer fully connected network corresponding to the first sub-channel, an independent fully connected network corresponding to the second sub-channel, a fusion layer, a shared feature learning channel, a multilayer perceptron and a Dropout regularization layer; the step of training the preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain the power transmission line galloping early warning model comprises the following steps: inputting the standardized meteorological feature data into the first sub-channel and extracting high-dimensional meteorological features through the corresponding double-layer fully connected network; inputting the galloping physical criterion features into the second sub-channel and extracting galloping inducement expression features through the corresponding independent fully connected network; performing feature fusion on the high-dimensional meteorological features and the galloping inducement expression features through the fusion layer; inputting the fused features into the shared feature learning channel for learning, and then outputting an original galloping prediction result through the multilayer perceptron and the Dropout regularization layer, and simultaneously obtaining the power transmission line galloping early warning model.

2. The method of claim 1, wherein, the step of training the preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain the power transmission line galloping early warning model further comprises the following step: training the preset deep learning model by taking a weighted binary cross-entropy loss function as an optimization objective function, and the formula of the weighted binary cross-entropy loss function is as follows: ； ； Wherein, N is the total number of samples of meteorological characteristic data and dance physical criterion characteristics, 0 and 1 are taken respectively, and represent no dance and dance respectively; is a weight coefficient; is the number of samples of the category .

3. The method of claim 1, wherein, after obtaining the original galloping prediction result, the method further comprises the following steps: constructing a correction function based on the comprehensive galloping coefficient, and correcting the original galloping prediction result based on the correction function to obtain a final galloping prediction result; the correction function is expressed by the following formula: ； wherein, is the original dance prediction result, is the final dance prediction result, is the comprehensive dance coefficient, is a Sigmoid activation function, such that the value range is (0, 1), is a preset correction coefficient.

4. The method of claim 3, wherein, warning of the galloping risk of the power transmission line based on the galloping prediction result comprises the following steps: The final galloping prediction result is compared with a preset threshold value, if the final galloping prediction result is greater than or equal to the preset threshold value, it is determined that there is a risk of galloping of the power transmission line, and if the final galloping prediction result is less than the preset threshold value, it is determined that there is no risk of galloping of the power transmission line.

5. A galloping warning device for a power transmission line, characterized by The device comprises: a meteorological monitoring data acquisition module for acquiring meteorological monitoring data covering a power transmission line area; a galloping physical criterion feature extraction module for extracting galloping physical criterion features based on the meteorological monitoring data; a model training module for training a preset deep learning model based on the meteorological monitoring data and the galloping physical criterion features to obtain a power transmission line galloping early warning model; a galloping risk early warning module for inputting actual meteorological monitoring data into the power transmission line galloping early warning model to obtain an original galloping prediction result and early warning of a power transmission line galloping risk based on the original galloping prediction result; The device further comprises: a data preprocessing module for performing data preprocessing of time alignment and data fusion, missing value filling, outlier removal and normalization on the meteorological monitoring data before extracting the galloping physical criterion features based on the meteorological monitoring data to obtain standardized meteorological feature data; The meteorological monitoring data includes wind speed, wind direction, line direction and meteorological factors; the galloping physical criterion feature extraction module comprises: a wind yaw angle extraction unit for extracting a wind yaw angle based on the wind direction and the line direction; a wind speed decomposition unit for decomposing the wind speed to obtain a transverse wind speed vector; an icing coefficient construction unit for constructing an icing coefficient based on meteorological factors; a comprehensive galloping coefficient construction unit for constructing a comprehensive galloping coefficient based on the wind yaw angle, the transverse wind speed vector and the icing coefficient; a galloping physical criterion feature determination unit for determining the wind yaw angle, the transverse wind speed component, the icing coefficient and the comprehensive galloping coefficient as the galloping physical criterion features; The preset deep learning model comprises a first sub-channel, a second sub-channel, a double-layer fully connected network corresponding to the first sub-channel, an independent fully connected network corresponding to the second sub-channel, a fusion layer, a shared feature learning channel, a multi-layer perceptron and a Dropout regularization layer; the model training module comprises: a first sub-channel processing unit for inputting the standardized meteorological feature data into the first sub-channel and extracting high-dimensional meteorological features through the corresponding double-layer fully connected network; a second sub-channel processing unit for inputting the galloping physical criterion features into the second sub-channel and extracting galloping inducement expression features through the corresponding independent fully connected network; a feature fusion unit for performing feature fusion on the high-dimensional meteorological features and the galloping inducement expression features through the fusion layer; a prediction unit for inputting the fused features into the shared feature learning channel for learning, and then outputting the original galloping prediction result after processing by the multi-layer perceptron and the Dropout regularization layer, and simultaneously obtaining the power transmission line galloping early warning model.

6. The apparatus of claim 5, wherein, The model training module further comprises: a loss function determination unit for training the preset deep learning model by using a weighted binary cross-entropy loss function as an optimization objective function, and the formula of the weighted binary cross-entropy loss function is: ； ； Wherein, N is the total number of samples of weather feature data and dance physical criterion features, 0 and 1 are taken respectively, and represent no dance and dance respectively; is a weight coefficient; is the number of samples of the category .

7. The apparatus of claim 6, wherein, The device also comprises: The correction module is configured to, after obtaining the original galloping prediction result, construct a correction function based on the comprehensive galloping coefficient, and correct the original galloping prediction result based on the correction function to obtain a final galloping prediction result; the correction function is expressed by the following formula: ； wherein, is the original dance prediction result, is the final dance prediction result, is the comprehensive dance coefficient, is a Sigmoid activation function, such that the value range is (0, 1), is a preset correction coefficient.

8. The apparatus of claim 7, wherein, The galloping risk early warning module comprises: The galloping risk judgment unit is configured to compare the final galloping prediction result with a preset threshold value, and determine that there is a galloping risk of the power transmission line if the final galloping prediction result is greater than or equal to the preset threshold value, or determine that there is no galloping risk of the power transmission line if the final galloping prediction result is less than the preset threshold value.

9. A computer device, comprising: The device comprises: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the power transmission line galloping early warning method according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the power transmission line galloping early warning method according to any one of claims 1 to 4.

Images