Imbalanced sample galloping prediction method and system based on meteorological data

Abstract

The invention relates to the technical field of power transmission engineering, and discloses an unbalanced sample galloping prediction method and system based on meteorological data, and the method comprises the steps: collecting the historical monitoring data of each monitoring point, and synchronously collecting the meteorological information of the corresponding monitoring point; monitoring devices and dates are taken as basic units, one piece of sample data is selected from daily data of each device, non-galloping samples in the sample data are grouped according to line numbers, and only samples corresponding to different spans are reserved in each group; according to line span characteristics, galloping high-incidence time periods and typical weather combinations, differential loss weights are given to sample data, and a weighted cross entropy loss function is constructed; and constructing a daily scale galloping prediction model, and training the model by using a weighted cross entropy loss function to realize daily scale prediction of a galloping event. By constructing a scientific sample extraction strategy, introducing a category weight mechanism and adopting a deep learning classification model, the recognition capability of the model on minority class events is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power transmission engineering, and in particular to a method and system for predicting unbalanced sample galloping based on meteorological data. Background Art

[0002] Galloping is the large-amplitude, low-frequency mechanical oscillation of transmission line conductors under certain meteorological conditions, particularly after ice or rime has formed, induced by crosswinds. Galloping typically exhibits pendulum, elliptical, or torsional patterns, with frequencies as low as 0.1Hz to 3Hz and amplitudes reaching tens of centimeters or even meters. Its mechanism is complex and closely related to various meteorological factors, including conductor tension, ice formation, wind speed and direction, temperature and humidity. Due to the strong nonlinear characteristics of galloping, it can quickly cause conductor skipping, phase collisions, insulation breakdown, and even serious accidents such as conductor breakage and tower damage.

[0003] Conductor galloping has long been a major hazard to the safe operation of power systems. Galloping is particularly common in cold, icy regions such as Northeast, Northwest, and North my country, where it frequently occurs during winter, placing significant pressure on power grid operations and emergency repairs. To address this, power systems have deployed numerous online monitoring devices along transmission lines to detect conductor displacement, acceleration, and galloping in real time. While monitoring technology is relatively mature, galloping rarely occurs in most time periods. Consequently, the number of galloping samples in actual data collection is far less than that of non-galloping samples, resulting in a typically highly unbalanced data structure.

[0004] Faced with such large-scale unbalanced data, traditional machine learning and statistical analysis methods often have the problem of "bias towards the majority class" during training, resulting in low recognition accuracy of dancing categories and high missed reporting rate, which makes it difficult to meet the accurate warning requirements of engineering applications. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for predicting dance movements based on unbalanced samples of meteorological data to solve the problem that the dance sample structure is extremely unbalanced, resulting in low recognition accuracy and high false negative rate of dance categories.

[0006] In a first aspect, the present invention provides a method for predicting unbalanced sample galloping based on meteorological data, the method comprising: Multiple monitoring devices are deployed along the transmission lines to collect historical monitoring data from each monitoring point and simultaneously collect meteorological information from the corresponding monitoring points; Taking the monitoring device and date as basic units, select a sample data from the daily data of each device, group the non-galloping samples in the sample data by line number, retain only samples corresponding to different gear spans in each group, and retain all galloping samples in the sample data. The sample data contains the complete time series data of the day; According to the line span characteristics, high-incidence periods of galloping and typical meteorological combinations, differentiated loss weights are assigned to the sample data to construct a weighted cross entropy loss function; A daily-scale dancing prediction model is constructed using an LSTM network. A time series input matrix is ​​constructed based on the filtered sample data. The time series input matrix is ​​input into the daily-scale dancing prediction model. The weighted cross-entropy loss function is used to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

[0007] The present invention provides a method for predicting dance movements based on unbalanced samples of meteorological data. In view of the scarcity of dance samples and the discrete temporal and spatial distribution, a sample extraction strategy based on the monitoring device-time dimension is designed, and the category weight factor is combined with the deep learning model to improve the model's recognition ability for minority class events.

[0008] In an optional embodiment, differentiated loss weights are assigned to sample data based on line span characteristics, high-incidence periods of galloping, and typical meteorological combinations, including: Extracting categorical features from the meteorological information, and calculating a category frequency weighting factor based on the frequency of each category in the sample data; According to the differences in galloping risks in different gear spans, corresponding structural sensitivity factors are set for different gear spans; According to the probability of dancing occurring in different time periods, corresponding time period sensitivity factors are set for different time periods; For sample data that meets the preset meteorological combination, set the meteorological condition adjustment factor.

[0009] In an optional embodiment, the weighted cross entropy loss function is as follows:

[0010] in, is the weighted cross entropy loss function, is the final sample weight of multi-factor synthesis, is the true label of the i,j,t,kth sample, is the probability of galloping events predicted by the daily-scale galloping prediction model.

[0011] In an optional embodiment, the method further includes: Normalizing numerical features in the meteorological information, the numerical features including wind speed, temperature, humidity, and precipitation; The categorical features in the meteorological information are converted into one-hot encoding features, where the categorical features include wind direction category and precipitation category.

[0012] In an optional embodiment, all sample data are jointly indexed by device number + timestamp.

[0013] In an optional embodiment, the daily scale dancing prediction model uses a one-layer or two-layer LSTM network to process the time series input matrix, extracts the time features, maps them to a single output node through a fully connected layer, and then outputs the probability of dancing occurring on that day through a Sigmoid activation function.

[0014] In an optional embodiment, the daily scale dance prediction model is expressed as follows:

[0015] in, is the Sigmoid activation function, is the weight matrix of the fully connected layer, is the hidden state of the last time step of the LSTM network, is the bias term.

[0016] In a second aspect, the present invention provides a device for predicting unbalanced sample galloping based on meteorological data, the device comprising: The data acquisition module is used to deploy multiple monitoring devices along the transmission line, collect historical monitoring data from each monitoring point, and simultaneously collect meteorological information from the corresponding monitoring points; A data screening module is used to select a sample data from the daily data of each device based on the monitoring device and date as basic units, group the non-galloping samples in the sample data by line number, retain only samples corresponding to different gear spans in each group, and retain all galloping samples in the sample data. The sample data contains the complete time series data of the day; The weight assignment module is used to assign differentiated loss weights to sample data based on line span characteristics, high-incidence periods of galloping, and typical meteorological combinations, and to construct a weighted cross-entropy loss function; The model prediction module is used to use the LSTM network to build a daily-scale dancing prediction model, construct a time series input matrix based on the filtered sample data, input the time series input matrix into the daily-scale dancing prediction model, and use the weighted cross-entropy loss function to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

[0017] The present invention provides an unbalanced sample dance prediction device based on meteorological data, which aims to solve the problem of poor prediction performance in existing dance prediction due to extreme sample imbalance. By constructing a scientific sample extraction strategy, introducing a category weight mechanism, and adopting a deep learning classification model, while ensuring the authenticity and representativeness of the data, the recognition accuracy and early warning capabilities of dance events are improved, and the operational safety and intelligence level of the power grid system under complex meteorological conditions are enhanced.

[0018] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to thereby execute the unbalanced sample dance prediction method based on meteorological data according to the first aspect or any corresponding embodiment thereof.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the unbalanced sample dance prediction method based on meteorological data according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 is a flow chart of a method for predicting unbalanced sample galloping based on meteorological data according to an embodiment of the present invention; Figure 2 is a structural block diagram of a device for predicting unbalanced sample galloping based on meteorological data according to an embodiment of the present invention; Figure 3 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0023] An embodiment of the present invention provides a method for predicting galloping of unbalanced samples based on meteorological data. Through technical means such as device stratified sampling, category weighting, and deep learning modeling, it effectively improves the accuracy of galloping prediction under conditions of severe category imbalance and enhances the power system's response capability to galloping risks.

[0024] According to an embodiment of the present invention, an embodiment of a method for predicting unbalanced sample galloping based on meteorological data is provided. It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0025] In this embodiment, a method for predicting the unbalanced sample oscillation based on meteorological data is provided. Figure 1 is a flow chart of a method for predicting unbalanced sample oscillation based on meteorological data according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps: Step S1: deploy multiple monitoring devices along the transmission line, collect historical monitoring data of each monitoring point, and simultaneously collect meteorological information of the corresponding monitoring point.

[0026] Specifically, multiple online monitoring devices were deployed along the transmission lines. Each device was equipped with the ability to identify windshield movement and transmit its status in real time. Five consecutive years of daily monitoring data were collected to form a raw sample database. Meteorological information corresponding to the monitoring points was also collected simultaneously, including but not limited to wind speed, temperature, humidity, precipitation type, and precipitation intensity. All data was indexed using a "device number + timestamp" system to ensure temporal and spatial consistency.

[0027] Step S2, taking the monitoring device and date as the basic units, select a sample data from the daily data of each device, group the non-dancing samples in the sample data by line number, retain only the samples corresponding to different gear spans in each group, retain all the dancing samples in the sample data, and the sample data contains the complete time series data of the day.

[0028] Specifically, in order to solve the problem of scarce dancing samples and serious imbalance in category ratios in monitoring data, this application adopts stratified sampling and structural balance strategies to finely screen and control training samples.

[0029] 1) First, using the monitoring device and date as the basic units, only one sample is retained for each monitoring device every day, and duplicate records of the same device on the same day are removed to avoid model deviation caused by data stacking.

[0030] 2) Secondly, among the non-dancing samples, we further group them by route number and retain only samples corresponding to different spans within each route to avoid a large concentration of samples with the same span and ensure balanced data coverage in the spatial structure. Suppose the set of all spans of a certain route is D = {d1, d2, …, dn}, then only retain the samples corresponding to each span in the set:

[0031] 3) All dance samples are retained without any deletion to ensure that the model fully learns the dance characteristics.

[0032] In response to the characteristics of dancing data, which have large sample sizes and extremely unbalanced categories, this application proposes a sample extraction method based on "monitoring device-date". Specifically, a representative data point is selected from the daily data of each device for modeling, effectively reducing the proportion of redundant data and maintaining a balanced global spatial and temporal distribution. By limiting the number of non-dancing samples and maintaining the full utilization of dancing samples, this strategy significantly alleviates the sample imbalance problem, while improving model training efficiency and data representativeness. It is suitable for sample management and modeling needs in large-scale heterogeneous monitoring scenarios.

[0033] In step S3, differentiated loss weights are assigned to the sample data according to the line span characteristics, the high-incidence period of galloping, and the typical meteorological combination, and a weighted cross entropy loss function is constructed.

[0034] Specifically, based on the traditional category weighting method, this application innovatively introduces structural sensitivity factors, time period adjustment factors, and meteorological condition adjustment factors to construct a category weighting mechanism with a multi-factor combination. This mechanism assigns differentiated loss weights to samples based on line span characteristics, high-incidence periods of gyrations, and typical meteorological combinations, thereby focusing on learning samples that are easily missed and improving the model's ability to identify small gyration samples. This method has good scalability and engineering adaptability, significantly improving the robustness and practicality of deep learning models under actual operating conditions.

[0035] Step S4, using the LSTM network to build a daily-scale dancing prediction model, constructing a time series input matrix based on the filtered sample data, inputting the time series input matrix into the daily-scale dancing prediction model, and using the weighted cross entropy loss function to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

[0036] Specifically, this application employs a model based on a long short-term memory (LSTM) network to analyze daily meteorological data and predict whether dancing events will occur that day. Taking into account the sensitivity of dancing behavior to meteorological changes, this application uses an LSTM network to model daily meteorological time-series data, enabling daily-scale prediction of dancing events. Compared to traditional methods that rely solely on statistical features, this application preserves both temporal patterns and mutation characteristics, improving the advance and accuracy of dancing predictions and better meeting the practical engineering requirement of predicting the risk of dancing the next day from the previous day.

[0037] In one specific example, over 3,000 online monitoring devices were deployed in a power grid area, collecting over 10 million data points over five years. Meteorological features (such as wind speed, humidity, and precipitation) were extracted, and trained using a weighted cross-entropy neural network model, increasing the detection rate of pulsation on the test set from 60% with the original model to over 87%. Compared to existing technologies, this invention offers the following advantages: it is suitable for scenarios with highly unbalanced samples, significantly improving the detection rate of minority classes (pulsation); it balances accuracy and generalization, with stable performance across multiple monitoring points; it is easy to deploy and expand, and can be connected to any meteorological monitoring system or online device; it enhances the security and operation of the power grid, and can serve as an intelligent support for pulsation prevention and control and strategy formulation.

[0038] This invention provides a method for predicting pulsation based on imbalanced samples of meteorological data. Addressing the scarcity and discrete spatiotemporal distribution of pulsation samples, a sample extraction strategy based on the monitoring device-time dimension is designed. This strategy, combined with category weighting factors and a deep learning model, enhances the model's ability to identify minority events. This method leverages the massive data resources accumulated over five years from over 3,000 existing monitoring devices. Without introducing false samples, it effectively mitigates the class imbalance problem and significantly improves the accuracy and robustness of pulsation prediction. This method possesses significant engineering value and potential for widespread adoption.

[0039] In an optional embodiment, the above step S3 includes: Step S31: extracting categorical features from meteorological information and calculating category frequency weighting factors based on the frequency of each category in the sample data.

[0040] Specifically, the basic category weight is calculated according to the frequency of each category in the training set :

[0041] in is the weight of category i, f i is the frequency of occurrence of the category, and c is the adjustment constant to prevent the denominator from being zero and to prevent extreme scaling. Generally, c is 1.01.

[0042] In the dancing prediction task, there are two categories: category 0, no dancing (large number of samples); category 1, dancing (very few samples).

[0043] Assume that the statistics in the current sample are as follows: Number of non-dancing samples: 800,000, accounting for approximately ; Number of dance samples: 5,000, accounting for approximately .

[0044] Let c = 1.01, then: Class 0 (no dancing) weights:

[0045] Category 1 (Dancing) Weights:

[0046] In this way, the loss of the dancing sample is multiplied by the larger w1, allowing the model to pay more attention to this rare data, thereby improving the recognition rate of dancing and avoiding being ignored due to too few samples.

[0047] Step S32: setting corresponding structural sensitivity factors for different gear spans according to the differences in galloping risks of different gear spans.

[0048] Specifically, considering the differences in galloping risks in different gear segments, the structural factor , give discrimination weights to non-dancing samples:

[0049] in, is the structural factor corresponding to the j-th gear pitch segment, which is set based on historical dancing statistics or experience. The common long gear pitch setting is >1, normal gear spacing is set to =1.

[0050] Step S33: setting corresponding time period sensitivity factors for different time periods according to the probability of dancing occurring in different time periods.

[0051] Specifically, considering that dancing is more likely to be ignored by the model during the period of high frequency (such as night and early morning), a time weighting term is specially set up. :

[0052] in, is the sensitivity coefficient of the tth time period, such as night =1.5, daytime =1.

[0053] Step S34: setting a meteorological condition adjustment factor for the sample data that meets the preset meteorological combination.

[0054] Specifically, for samples under specific meteorological combinations (such as high wind speed and high humidity), the meteorological adjustment factor is further multiplied. :

[0055] If the sample meets certain meteorological conditions (such as wind speed > 12m / s∧humidity > 90%), then set >1, otherwise 1.

[0056] In an optional embodiment, the method further includes: Step S5: normalize the numerical features in the meteorological information, where the numerical features include wind speed, temperature, humidity, and precipitation.

[0057] Step S6: converting the categorical features in the meteorological information into one-hot encoding features, where the categorical features include wind direction category and precipitation category.

[0058] Specifically, meteorological factors influencing power line galloping are extracted as model input features, primarily including wind speed (m / s), wind direction, temperature (°C), humidity (%), precipitation (mm), and precipitation type (encoded). All numerical features, such as wind speed, temperature, humidity, and precipitation, are normalized. Categorical features, such as wind direction and precipitation type, are converted into one-hot encoded features. This allows the model to more clearly distinguish between different wind directions and precipitation types, thereby more accurately learning the relationship between these features and power line galloping.

[0059] In an optional implementation, this application uses a model based on a long short-term memory (LSTM) network to perform time-series analysis on daily meteorological data and predict whether a dancing event will occur that day. This model serves only as a tool for implementing the service, focusing on improving prediction results in conjunction with the aforementioned multi-factor loss weighting mechanism. Its basic structure and training process are described below: 1) Input data construction For each monitoring device, a time series input matrix containing 24-hour meteorological data is constructed every day, and the corresponding label y∈{0,1} indicates whether a dancing event occurs on that day.

[0060] 2) Model structure The model uses a one- or two-layer LSTM network to process the time series input matrix, extracts the time features, maps them to a single output node through a fully connected layer, and then uses a Sigmoid activation function to output the probability of dancing occurring on that day. The specific expression is:

[0061] in, is the Sigmoid activation function; is the weight matrix of the fully connected layer, which is usually of dimension 1×d, where d is the dimension of the hidden state vector output by the LSTM (e.g. 64 or 128); is the hidden state of the last time step of the LSTM network; is the bias term, which is a scalar.

[0062] 3) Model training and loss function This application uses the weighted cross entropy loss function constructed by the aforementioned multi-factor weight mechanism to train the model. The loss function is:

[0063] in, is the weighted cross entropy loss function, is the final sample weight of multi-factor synthesis, is the true label (0 or 1) of the i, j, t, kth sample, is the probability of galloping events predicted by the daily-scale galloping prediction model.

[0064] The model is trained using a conventional optimizer (such as Adam) and an early stopping strategy. During the training process, various model metrics on the validation set are recorded to ensure that the model can achieve high recall and F1 scores under imbalanced sample conditions.

[0065] This embodiment also provides a device for predicting unbalanced sample galloping based on meteorological data. This device is used to implement the above-mentioned embodiments and preferred implementations, and details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0066] This embodiment provides an unbalanced sample dance prediction device based on meteorological data, such as Figure 2 As shown, including: The data acquisition module 21 is used to deploy multiple monitoring devices along the transmission line, collect historical monitoring data of each monitoring point, and simultaneously collect meteorological information of the corresponding monitoring point.

[0067] The data screening module 22 is used to select a sample data from the daily data of each device based on the monitoring device and date as the basic unit, group the non-dancing samples in the sample data by line number, retain only the samples corresponding to different gear spans in each group, retain all the dancing samples in the sample data, and the sample data contains the complete time series data of the day.

[0068] The weight assignment module 23 is used to assign differentiated loss weights to sample data according to line span characteristics, high-incidence periods of dancing and typical meteorological combinations, and construct a weighted cross entropy loss function.

[0069] The model prediction module 24 is used to construct a daily-scale dancing prediction model using an LSTM network, construct a time series input matrix based on the filtered sample data, input the time series input matrix into the daily-scale dancing prediction model, and use a weighted cross-entropy loss function to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

[0070] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0071] The unbalanced sample galloping prediction device based on meteorological data in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0072] The present invention provides an unbalanced sample dance prediction device based on meteorological data, which aims to solve the problem of poor prediction performance in existing dance prediction due to extreme sample imbalance. By constructing a scientific sample extraction strategy, introducing a category weight mechanism, and adopting a deep learning classification model, while ensuring the authenticity and representativeness of the data, the recognition accuracy and early warning capabilities of dance events are improved, and the operational safety and intelligence level of the power grid system under complex meteorological conditions are enhanced.

[0073] The embodiment of the present invention also provides a computer device having the above Figure 2 The unbalanced sample galloping prediction device based on meteorological data is shown.

[0074] See also Figure 3 , Figure 3 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 3As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 3 A processor 10 is taken as an example.

[0075] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0076] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0077] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0078] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0079] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0080] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a 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 a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0081] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for predicting unbalanced sample galloping based on meteorological data, characterized in that: The method comprises: Multiple monitoring devices are deployed along the transmission lines to collect historical monitoring data from each monitoring point and simultaneously collect meteorological information from the corresponding monitoring points; Taking the monitoring device and date as basic units, select a sample data from the daily data of each device, group the non-galloping samples in the sample data by line number, retain only samples corresponding to different gear spans in each group, and retain all galloping samples in the sample data. The sample data contains the complete time series data of the day; According to the line span characteristics, high-incidence periods of galloping and typical meteorological combinations, differentiated loss weights are assigned to the sample data to construct a weighted cross entropy loss function; A daily-scale dancing prediction model is constructed using an LSTM network. A time series input matrix is ​​constructed based on the filtered sample data. The time series input matrix is ​​input into the daily-scale dancing prediction model. The weighted cross-entropy loss function is used to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

2. The unbalanced sample galloping prediction method based on meteorological data according to claim 1 is characterized in that: Differentiated loss weights are assigned to sample data based on line span characteristics, high-incidence periods of galloping, and typical weather conditions, including: Extracting categorical features from the meteorological information, and calculating a category frequency weighting factor based on the frequency of each category in the sample data; According to the differences in galloping risks in different gear spans, corresponding structural sensitivity factors are set for different gear spans; According to the probability of dancing occurring in different time periods, corresponding time period sensitivity factors are set for different time periods; For sample data that meets the preset meteorological combination, set the meteorological condition adjustment factor.

3. The unbalanced sample galloping prediction method based on meteorological data according to claim 2 is characterized in that: The weighted cross entropy loss function is as follows: in, is the weighted cross entropy loss function, is the final sample weight of multi-factor synthesis, is the true label of the i,j,t,kth sample, is the probability of galloping events predicted by the daily-scale galloping prediction model.

4. The unbalanced sample galloping prediction method based on meteorological data according to claim 2 is characterized in that: The method further comprises: Normalizing numerical features in the meteorological information, the numerical features including wind speed, temperature, humidity, and precipitation; The categorical features in the meteorological information are converted into one-hot encoding features, where the categorical features include wind direction category and precipitation category.

5. The unbalanced sample galloping prediction method based on meteorological data according to claim 1 is characterized in that: All sample data are indexed by device number + timestamp.

6. The unbalanced sample galloping prediction method based on meteorological data according to claim 1 is characterized in that: The daily scale dance prediction model uses a one-layer or two-layer LSTM network to process the time series input matrix, extracts the time features, maps them to a single output node through a fully connected layer, and then outputs the probability of dance occurring on that day through a Sigmoid activation function.

7. The unbalanced sample galloping prediction method based on meteorological data according to claim 6 is characterized in that: The daily scale dancing prediction model is expressed as follows: in, is the Sigmoid activation function, is the weight matrix of the fully connected layer, is the hidden state of the last time step of the LSTM network, is the bias term.

8. A device for predicting unbalanced sample oscillation based on meteorological data, characterized in that: The device comprises: The data acquisition module is used to deploy multiple monitoring devices along the transmission line, collect historical monitoring data from each monitoring point, and simultaneously collect meteorological information from the corresponding monitoring points; A data screening module is used to select a sample data from the daily data of each device based on the monitoring device and date as basic units, group the non-galloping samples in the sample data by line number, retain only samples corresponding to different gear spans in each group, and retain all galloping samples in the sample data. The sample data contains the complete time series data of the day; The weight assignment module is used to assign differentiated loss weights to sample data based on line span characteristics, high-incidence periods of galloping, and typical meteorological combinations, and to construct a weighted cross-entropy loss function; The model prediction module is used to use the LSTM network to build a daily-scale dancing prediction model, construct a time series input matrix based on the filtered sample data, input the time series input matrix into the daily-scale dancing prediction model, and use the weighted cross-entropy loss function to train the daily-scale dancing prediction model to achieve daily-scale prediction of dancing events.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the unbalanced sample dance prediction method based on meteorological data according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the unbalanced sample galloping prediction method based on meteorological data according to any one of claims 1 to 7.

Images