Transmission line hierarchical reclosing decision method and system suitable for winter faults

Abstract

The application relates to a power transmission line hierarchical reclosing decision method and system suitable for winter faults, wherein the method comprises the following steps: acquiring a fault transient waveform, extracting time-frequency features based on a time-frequency network classification model and outputting a fault cause probability distribution; constructing a reclosing success rate prior probability distribution of each type of fault cause based on historical reclosing data, and performing parameter estimation on historical reclosing prior information by using a beta-binomial distribution to represent the uncertainty; fusing the probability distribution and the prior distribution by using a probability fusion model to calculate the current fault reclosing success probability; comparing the reclosing success probability with a preset threshold to generate a hierarchical reclosing decision instruction. The application realizes fine identification of fault causes and explicit quantification of historical prior uncertainty, deeply fuses multi-source information through a Bayesian probability framework, significantly improves the reclosing decision accuracy, effectively improves the reclosing success rate, and reduces the impact of blind reclosing on the power grid.

Description

Technical Field

[0001] This invention relates to the field of transmission line protection and automation technology, and in particular to a graded reclosing decision method and system for transmission lines applicable to winter faults. Background Technology

[0002] Against the backdrop of global climate change, extreme weather events are becoming more frequent, posing a severe challenge to the safe operation of power systems. Especially during winter, severe weather conditions such as ice, snow, low temperatures, and strong winds significantly increase the failure rate of transmission lines. Reclosing, as a key means of rapidly restoring power supply and improving system stability, relies on the accurate identification of the nature of the fault for its correct operation. However, the physical mechanisms and recoverability of the aforementioned faults differ greatly, posing a serious challenge to the correct operation of automatic reclosing for rapid power restoration. Existing technological solutions have systemic shortcomings in addressing this challenge, mainly reflected in the following three directions of evolution:

[0003] Firstly, there are analytical methods based on the electrical transient characteristics after a fault. Existing adaptive reclosing research mainly relies on fault transient electrical signals, using the characteristics of the primary arc, secondary arc, and recovery voltage stages to distinguish between transient and permanent faults. For example, criteria are formed by analyzing the time-shifted voltage energy ratio, the voltage fundamental frequency component, or the current of the parallel reactor, but there is little further subdivision of the specific fault causes. This type of discrimination method may lead to misjudgment in complex operating environments. For example, an ice-breaking jump fault in its initial stage is easily identified as a permanent fault and reclosing is mistakenly blocked, while in reality, after a brief ice-breaking jump and recovery, the line often recovers its insulation and is ready for reclosing.

[0004] Secondly, decision-making methods incorporating historical data and multi-source auxiliary information: To compensate for the limitations of relying solely on electrical signal criteria, existing research has attempted to introduce prior information such as historical operating data as auxiliary basis. Examples include classification based on historical trip rates, simulation using statistical reclosing success rates, or designing secondary reclosing schemes by combining historical action information. However, a key drawback is that it ignores the inherent "uncertainty" in historical statistics (such as confidence issues due to limited samples), using only simple point estimations (average success rates), and failing to scientifically integrate real-time fault characteristics with prior knowledge within a probabilistic framework. This results in unreliable decision-making in boundary cases. In other words, these methods are mostly at the stage of summarizing post-fault patterns and have not yet achieved deep dynamic coupling with real-time fault characteristics, thus their application in practical online decision-making remains insufficient.

[0005] Thirdly, deep learning-based intelligent fault identification methods are developing towards more refined and intelligent research on fault cause identification to support accurate decision-making in adaptive reclosing. Specifically, image recognition methods, leveraging deep learning technology, are widely used in transmission line fault identification, not only distinguishing short-circuit types but also extending to the mechanism analysis and identification of specific fault causes such as foreign object intrusion, wildfires, wind deflection, and de-icing tripping. Specific technical approaches include using three-phase fault voltage and current waveform images as input and constructing a fault classifier using a transfer learning-AlexNet neural network; converting waveform images into Gram angle field maps as input and using an improved ResNet model for fault cause identification; and using current trajectory images as input and employing an improved convolutional neural network based on a channel attention mechanism for fault identification.

[0006] Despite progress in fault feature input methods and deep identification models, existing research lacks more efficient methods for extracting and representing the time-frequency domain features inherent in transient processes. This hinders the full revelation of the essential differences between various faults and limits the accurate identification capability in complex operating scenarios. The bottlenecks are specifically manifested in: 1) The general convolutional neural networks used are insufficiently efficient at extracting key time-frequency joint features of power transient signals, limiting the upper limit of identification accuracy; 2) The "black box" nature of the models leads to poor interpretability, which is detrimental to applications in high-reliability power scenarios; 3) More importantly, most existing studies are "classification for the sake of classification," and their output probability results do not form a closed loop with subsequent reclosing success rate assessment and decision-making processes. The identification results are merely real-time "likelihoods," failing to be Bayesianly integrated with historical "prior" data containing uncertainty, thus unable to directly output robust probabilities for refined risk decision-making.

[0007] In summary, current technological advancements in addressing the complex fault reclosing decision-making problem in winter transmission lines exhibit a fragmented approach across three stages: front-end identification, mid-stage evaluation, and back-end decision-making. At the front-end identification stage, while progress is towards refinement, feature extraction methods lack specificity for key time-frequency transient characteristics of the power system, and the output is poorly coupled with decision-making requirements. At the mid-stage evaluation stage, there is a severe lack of a quantitative evaluation model that scientifically integrates real-time observation uncertainty (identification probability) with historical statistical uncertainty (prior distribution). Existing methods' neglect or simplistic handling of uncertainty makes decision-making exceptionally fragile in data-sparse or boundary-based scenarios. At the back-end decision-making stage, the decision logic is generally a rigid "single threshold - binary output" model, failing to implement differentiated, flexible, and tiered handling strategies (such as delayed reclosing and condition-triggered reclosing) closely linked to specific fault physical mechanisms and real-time dynamic scenarios.

[0008] Therefore, a systematic and innovative solution is urgently needed to achieve the integration of refined probability identification of faults, quantitative fusion of uncertainty, and scenario-based hierarchical decision-making, thereby significantly improving the accuracy and reliability of reclosing operations under complex faults in winter. Summary of the Invention

[0009] In view of this, to address the problems of imprecise fault identification and blind reclosing decisions due to a lack of prior information support in winter transmission line faults, this invention proposes a hierarchical reclosing decision-making method and system for winter faults. The overall process can be summarized as follows: First, acquire transient waveform data of transmission line faults and extract waveform image features using a time-frequency network to achieve probabilistic identification of typical fault causes in winter; second, collect historical reclosing operation data and fit the prior reclosing power distribution for each fault cause using a beta-binomial distribution; then, calculate the reclosing success probability under the current fault by comparing the real-time fault cause identification probability with the sampled values ​​of the corresponding prior distribution using a Bayesian conditional probability model; finally, based on the comparison between the calculated probability and a preset threshold, and combined with different operating scenarios, a hierarchical reclosing decision is formed.

[0010] The specific technical solution is as follows:

[0011] The first aspect of this invention discloses a graded reclosing decision method for transmission lines applicable to winter faults, comprising the following steps: S1, acquiring transient waveform data when a fault occurs in the transmission line; S2, extracting time-frequency features from the transient waveform data based on a pre-constructed time-frequency network classification model, and probabilistically identifying the fault cause based on the time-frequency features, outputting a fault cause probability distribution, which represents the probability that the current fault belongs to each of the preset fault cause categories; S3, acquiring historical reclosing operation data, and constructing a priori probability distribution of reclosing power for each preset fault cause category based on the historical reclosing operation data; S4, calculating the reclosing success probability under the current fault using a probability fusion model based on the fault cause probability distribution and the reclosing power prior probability distribution; S5, comparing the calculated reclosing success probability with a preset threshold, and generating a graded reclosing decision instruction based on the comparison result.

[0012] In step S1, when a short-circuit fault occurs in the transmission line, the relay protection device quickly issues a trip command. At the same time, the fault recorder or the recording function inside the protection device is triggered, thereby recording electrical waveform data for a period of time before and after the fault occurs. The required transient waveform data can be obtained from the fault recorder or protection device. This data records the subtle process of voltage and current changes over time at a high sampling rate (usually several kilohertz), which contains rich information such as fault type, fault distance, transition resistance, arc characteristics, and physical causes of the fault.

[0013] Traditional methods for converting or processing transient waveform data involve directly stacking one-dimensional time-series waveforms into a two-dimensional image or using a fixed time-frequency transform (such as STFT) as preprocessing. A common drawback is that the feature extraction method is separated from the subsequent classification task, making it impossible to adaptively optimize the time-frequency transform parameters according to the fault identification task objective. In step S2 of this invention, a time-frequency network classification model is employed. Essentially, this is an end-to-end learnable joint framework for time-frequency feature extraction and classification. This model receives the original one-dimensional multi-channel transient waveform, maps the time-domain signal into a two-dimensional time-frequency distribution map through an internal differentiable time-frequency transform operation, and then uses a deep convolutional neural network to extract hierarchical abstract features from this time-frequency map. Finally, a fully connected classification layer outputs probability vectors belonging to each preset fault cause category. This process achieves joint optimization of "time-frequency representation learning" and "fault classification decision," with feature extraction parameters directly updated with gradients for the fault identification task, thus achieving stronger discriminative power than a fixed time-frequency transform.

[0014] Furthermore, existing technologies, when processing historical reclosing statistics, typically only calculate the average reclosing power for various fault types and use it as a fixed constant. This point estimation method implicitly assumes that historical statistics are a perfect estimate of the true success rate and that this estimate is error-free. However, in actual power systems, the number of historical samples for various fault types varies significantly, and some rare faults may only be recorded a few times, resulting in highly random statistical averages. In addition, reclosing success rates fluctuate significantly across different years, regions, and weather conditions. Ignoring these uncertainties and directly using point estimates as the basis for decision-making can easily lead to uncontrolled decision-making risks.

[0015] In step S3 of this invention, a "prior probability distribution of recombinant power" is used instead of the traditional "point estimation of recombinant power." Specifically, the recombinant power of each type of failure cause is treated as a random variable, rather than a fixed constant. Based on historical statistical data, a probability distribution is assigned to this random variable. The expected value of this distribution corresponds to the traditional point estimation, while the variance and confidence interval of the distribution explicitly quantify the uncertainty of the estimation. This feature uses the beta-binomial distribution as the prior distribution model. Its core advantages are: First, the beta distribution is a continuous distribution defined on the interval [0,1], which naturally matches the mathematical properties of the success rate as a proportion of data; Second, the beta distribution is the conjugate prior of the binomial distribution, which makes the parameter estimation have an analytical form and high computational efficiency; Third, by adjusting the shape parameter, different confidence levels can be flexibly represented—the larger the historical sample size, the more concentrated the distribution (the smaller the uncertainty); the smaller the historical sample size, the more dispersed the distribution (the greater the uncertainty).

[0016] Furthermore, step S4 of this invention employs a "probabilistic fusion model," the mathematical essence of which is the total probability formula within a Bayesian framework. This feature can further utilize Monte Carlo sampling to achieve the aforementioned expected numerical calculation: independently extracting a large number of success rate samples from the prior distribution of each type of fault, weighting and summing the sample matrix with the real-time probability vector to obtain a set of predicted recombinant power values, and taking their arithmetic mean as the final estimate. The advantages of this method are: first, it fully preserves the morphological information of the prior distribution, rather than only using its expected value; second, it simulates "how the uncertainty of historical experience is transmitted to the current decision" through the sampling process; and third, the calculation process has clear probabilistic semantics and strong interpretability.

[0017] Existing reclosing control logic typically follows a binary decision-making model: "close if conditions are met, otherwise do not close." Step S5 of this invention introduces a hierarchical decision-making mechanism, classifying operation instructions into different levels based on the probability of successful reclosing. The engineering significance of this design is that: for faults with extremely high success probabilities, reclosing is decisively implemented to quickly restore power; for faults with extremely low success probabilities, resolute blocking is enforced to prevent secondary impacts caused by reclosing into a permanent fault; for faults with success probabilities in the intermediate range, a simple "black and white" approach is not adopted, but rather a refined confirmation and differentiated handling process is initiated to explore opportunities for conditional reclosing.

[0018] The decision-making method provided by this invention organically connects four functional modules: "time-frequency deep learning probability identification," "prior distribution uncertainty quantification," "Bayesian probability fusion," and "hierarchical threshold decision-making," constructing a complete intelligent reclosing decision-making closed loop. Compared with existing technologies, its beneficial effects are reflected in the following aspects: First, it is the first to achieve probabilistic fusion of real-time waveform features and historical statistical patterns in the field of reclosing decision-making, changing the situation where the two types of information are separated in traditional methods; Second, it is the first to explicitly incorporate statistical uncertainty into the prior knowledge representation, enabling the decision-making system to automatically identify small sample scenarios and adopt more conservative strategies, thus improving the robustness of the system; Third, it is the first to propose using the reclosing success probability as the core decision variable, transforming the reclosing problem into a probability prediction problem, and providing a unified mathematical framework for subsequent threshold optimization and strategy adaptation.

[0019] Furthermore, the time-frequency network classification model includes: a time-frequency transformation layer, constructed to perform time-frequency transformation on the input one-dimensional multi-channel transient waveform data to generate a two-dimensional time-frequency distribution map; and a backbone feature extraction network, which is connected to the time-frequency transformation layer, for performing feature extraction and classification on the two-dimensional time-frequency distribution map and outputting the probability distribution of the fault causes.

[0020] The function of the time-frequency transformation layer is to convert a one-dimensional time-series signal into a two-dimensional time-frequency distribution map. Unlike traditional time-frequency analysis methods (such as Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT), the time-frequency transformation layer defined by this feature is differentiable and embedded within a neural network architecture. This means that its time-frequency transformation parameters (such as window function, scaling factor, modulation frequency, etc.) can be optimized through backpropagation algorithms, thereby adaptively learning the time-frequency representation method most conducive to fault classification. The backbone feature extraction network receives the two-dimensional time-frequency distribution map output by the time-frequency transformation layer. Through multi-layer convolution, pooling, nonlinear activation, and other operations, it gradually abstracts the low-level time-frequency texture features into high-level semantic features, and finally maps them to the probability distribution of fault cause categories through a fully connected layer. That is, its input is a time-frequency map, and its output is a probability vector.

[0021] By implementing a two-stage architecture of "time-frequency transformation layer + backbone network" for the time-frequency network classification model, end-to-end joint optimization of time-frequency analysis and deep classification is achieved, overcoming the shortcomings of the traditional "fixed time-frequency transformation + independent classifier" scheme where feature extraction is disconnected from the task objective. In addition, it reduces the network parameter size and training difficulty. The time-frequency transformation layer, as a pre-feature extractor, compresses one-dimensional long sequences into two-dimensional compact representations, reducing the processing burden on the backbone network. Moreover, it also enhances the interpretability of the model. The output of the time-frequency transformation layer is a visualized two-dimensional time-frequency plot, which makes it easier for operators to understand which time-frequency regions the model focuses on for decision-making.

[0022] Furthermore, the time-frequency transformation layer is a time-frequency convolutional layer including a real part convolution kernel and an imaginary part convolution kernel. It convolves the transient waveform data through the real part convolution kernel and the imaginary part convolution kernel to generate real part feature maps and imaginary part feature maps respectively, and calculates the modulus of the real part feature maps and the imaginary part feature maps to synthesize the two-dimensional time-frequency distribution map.

[0023] By specifying the time-frequency transformation layer with "complex convolution kernel + modulus synthesis", task-driven adaptive optimization of the time-frequency transformation parameters is achieved. This frees the time-frequency resolution from being limited by a fixed window function, enabling precise separation of faults with similar spectra but different time-varying characteristics, such as icing-de-icing jumps and icing-skirt bridging. Moreover, the design of the complex convolution kernel preserves the phase information of the signal. The real part mapping and imaginary part mapping capture the projection of the signal onto the orthogonal basis, respectively, and the modulus calculation extracts the energy envelope. This process is consistent with the definition of instantaneous signal energy in physics and has a solid theoretical foundation in signal processing. In addition, the convolution operation has high computational efficiency and can be implemented using highly optimized convolution operators in modern deep learning frameworks, meeting the millisecond-level response requirements of real-time decision-making in power systems.

[0024] Additionally, the backbone feature extraction network is a convolutional neural network, which sequentially includes: a first convolutional module for preliminary feature extraction from the two-dimensional time-frequency distribution map; the first convolutional module contains 16 convolutional kernels of size 15×8; a second convolutional module connected to the first convolutional module for deep feature extraction and downsampling; the second convolutional module contains 32 convolutional kernels of size 3×16; a third convolutional module and a fourth convolutional module connected sequentially for extracting high-level semantic features; the third convolutional module contains 64 convolutional kernels of size 3×32; the fourth convolutional module contains 128 convolutional kernels of size 3×64; an adaptive pooling layer connected to the fourth convolutional module for adjusting the feature map to a fixed size; and a fully connected classification layer connected to the adaptive pooling layer for outputting the probability distribution of the fault causes.

[0025] Furthermore, based on historical reclosing operation data, a prior probability distribution of reclosing power for each preset fault cause category is constructed, including: for each preset fault cause category, the total number of historical reclosing actions and the number of historical successful reclosings are counted from its historical fault event records; the reclosing power is treated as a random variable, and a beta-binomial distribution is used as the prior probability distribution model for this random variable; based on the historical successful reclosings and the total number of historical reclosing actions, two shape parameters of the beta-binomial distribution are determined, one of which is positively correlated with the historical successful reclosings and the other is positively correlated with the historical failed reclosings; the expected value and confidence interval of the beta-binomial distribution are calculated according to the shape parameters, the expected value is used as a point estimate of the reclosing power, and the confidence interval is used to characterize the uncertainty of the estimate.

[0026] It should be noted that "total number of reclosing events that should have occurred under this fault condition" refers to the total number of events for which reclosing should have occurred (usually the historical number of occurrences of this type of fault), while "number of successful reclosing events" is the number of successful reclosing events. The ratio of the two is the empirical reclosing success rate.

[0027] By constructing the prior probability distribution through the method provided by this invention, the historical reclosing success rate is upgraded from a "fixed value" to a "distribution," giving prior knowledge an uncertainty dimension. The decision-making system can automatically identify scenarios with "scarce historical data" and adopt more prudent strategies. Moreover, the beta-binomial distribution model is highly compatible with the statistical characteristics of success rate data, and the parameter estimation has an analytical solution, eliminating the need for iterative solutions and resulting in high computational efficiency, making it suitable for online system deployment. In addition, the confidence interval provides operators with an intuitive measure of decision risk, which can indicate that a certain estimated value should not be trusted, and the system will automatically reduce the weight of that prior.

[0028] Furthermore, the probability of successful re-merging under the current fault is calculated using a probabilistic fusion model, including: for each preset fault cause, multiple re-merging power samples are independently extracted from the corresponding prior probability distribution of re-merging power using the Monte Carlo method, as a successful re-merging sample matrix; the fault cause probability distribution of the current fault belonging to each preset fault cause category and the extracted re-merging power sample matrix are weighted and synthesized based on the full probability formula to obtain multiple predicted re-merging power values; the arithmetic mean of the multiple predicted re-merging power values ​​is calculated as the successful re-merging probability under the current fault.

[0029] Through the design of the probabilistic fusion model in this invention, the organic fusion of real-time identification information and historical prior information is achieved under a strict probabilistic framework. The fusion result has clear mathematical semantics and no black box components. Moreover, the Monte Carlo sampling method completely preserves the non-Gaussian shape of the prior distribution. For skewed distributions caused by small samples, the sampling can automatically capture its actual shape, rather than simply approximating it with a Gaussian distribution. In addition, its calculation process naturally supports parallelization. The weighted summation of M samplings can be transformed into matrix operations, and millisecond-level response can be achieved under modern computing architectures, meeting the real-time requirements of protection devices.

[0030] Furthermore, the transient waveform data includes time-series sampled values ​​of at least the following electrical quantity channels obtained from the fault recording device: phase A voltage, phase B voltage, phase C voltage, zero-sequence voltage, phase A current, phase B current, phase C current, and zero-sequence current. In the step of acquiring transient waveform data when a fault occurs in the transmission line, transient waveform data within a time window containing at least one cycle before the fault and at least two cycles after the fault is extracted, based on the fault occurrence time. Specifically, the sampled values ​​of each channel are resampled to a unified standard sampling frequency, and data within a fixed time window is extracted based on the fault occurrence time to form a standardized input matrix. The eight-channel joint input provides multi-view fault features for the time-frequency convolutional layer. For example, icing-skirt bridging faults are often accompanied by zero-sequence current features, while wind deflection faults may exhibit asymmetrical features in the three-phase currents. The complementary information from multiple channels helps improve identification accuracy. Moreover, the fixed time window extraction strategy eliminates the alignment error of the fault occurrence time, ensuring the spatiotemporal consistency of the time-frequency network input data.

[0031] Furthermore, the preset fault cause categories include at least two of the following: foreign object short circuit, wildfire, wind deflection, icing-de-icing jump, and icing-insulator skirt bridging.

[0032] Furthermore, the preset thresholds include a high threshold and a low threshold; wherein the high threshold is determined based on the average reclosing power of all preset fault cause categories in winter, and the low threshold is determined based on the lowest reclosing power among all preset fault cause categories in winter; the step of generating graded reclosing decision instructions based on the comparison results includes: if the reclosing success probability is greater than the high threshold, it is determined as a Level I decision, and an immediate reclosing instruction is generated; if the reclosing success probability is between the low threshold and the high threshold, it is determined as a Level II decision, and a fault confirmation and differentiated handling instruction is generated; if the reclosing success probability is less than or equal to the low threshold, it is determined as a Level III decision, and a reclosing blocking instruction is generated.

[0033] The high threshold is defined as the weighted (or arithmetic) average of the average re-closing power across all preset fault cause categories. This represents the average expected re-closing level of the current power grid under winter conditions. A threshold above this level indicates a significantly higher probability of successful re-closing than the historical average, warranting immediate execution. The low threshold is defined as the lowest re-closing power among all fault causes, representing the minimum success rate under the worst-case scenario. A threshold below this level means that even with the most optimistic estimate (identifying it as the fault with the highest success rate), the probability of successful re-closing is still extremely low, and it should be blocked. Furthermore, this invention maps continuous probability values ​​to a discrete decision instruction space. Level I decisions correspond to "high-confidence re-closing," executing immediate re-closing to maximize power restoration speed; Level III decisions correspond to "high-confidence blocking," avoiding re-closing at permanent faults that could cause equipment damage and system impact; Level II decisions correspond to the "uncertain region," where conclusions are not rushed, but rather the subsequent fault confirmation and differentiated handling process is initiated.

[0034] Through the specific design of this hierarchical decision-making mechanism, the following advantages are achieved: the dual threshold settings are automatically generated based entirely on historical statistical data, requiring no manual adjustment and possessing adaptability; as historical data accumulates, the thresholds can automatically reflect changes in operating conditions; moreover, the three-level decision structure avoids the defects of black-and-white binary decision-making, providing an opportunity for "reconfirmation" of faults with conditional reclosing, significantly improving the reclosing success rate; in addition, the introduction of Level II decision-making organically connects decision-making with diagnosis, upgrading the reclosing device from a simple actuator to a decision system with intelligent judgment capabilities.

[0035] Furthermore, the fault confirmation and differentiated handling instructions include: determining the dominant fault cause based on the probability distribution of fault causes belonging to each preset fault cause category; and executing the reclosing control strategy corresponding to the dominant fault cause based on the preset fault cause-handling strategy mapping relationship, including: if the dominant fault cause is icing-de-icing jump, then the reclosing control strategy is delayed reclosing, with the delay time set according to the empirical value of conductor stability; if the dominant fault cause is icing-insulator skirt bridging, then the reclosing control strategy is delayed reclosing. Slightly rapid reclosing; if the primary fault cause is a foreign object short circuit, the reclosing control strategy is: based on the assessment result of the possibility of the fault arc clearing the foreign object, decide whether to perform reclosing or block reclosing; if the primary fault cause is a wildfire, the reclosing control strategy is: based on whether the fire situation is under control, decide whether to perform reclosing or block reclosing; if the primary fault cause is wind deviation, the reclosing control strategy is: based on whether the real-time wind speed in the fault section is lower than the line design wind speed, decide whether to perform reclosing or block reclosing.

[0036] The assessment of the possibility of foreign object removal includes: obtaining the status of foreign objects by detecting the recovery time of the fault phase voltage after tripping, the attenuation characteristics of the high-frequency component of the zero-sequence current, or by obtaining the status of foreign objects through online image monitoring devices of transmission lines; the judgment of whether the fire situation is under control includes: obtaining information on the distance to the fire point and the fire intensity from the power grid disaster prevention and mitigation platform or the fire monitoring system; the acquisition of real-time wind speed includes: obtaining wind speed data of the fault section from online micro-meteorological monitoring devices of the line or numerical weather prediction services.

[0037] The second aspect of this invention also discloses a system for implementing the method disclosed in the first aspect of this invention, comprising: a data acquisition module for acquiring transient waveform data and historical reclosing operation data when a fault occurs in a transmission line; a fault identification module for extracting time-frequency features from the transient waveform data based on a pre-constructed time-frequency network classification model, and probabilistically identifying the cause of the fault based on the time-frequency features, and outputting a fault cause probability distribution, which represents the probability that the current fault belongs to each of the preset fault cause categories; a priori modeling module for constructing a priori probability distribution of reclosing power for each preset fault cause category based on the historical reclosing operation data; a probability fusion module for calculating the reclosing success probability under the current fault based on the fault cause probability distribution of the current fault belonging to each preset fault cause category and the priori probability distribution of reclosing power for each preset fault cause category through a probability fusion model; and a decision execution module for comparing the calculated reclosing success probability with a preset threshold, generating a graded reclosing decision instruction based on the comparison result, and outputting it.

[0038] A third aspect of the present invention also discloses an electronic device comprising: at least one processor; and at least one memory storing a computer program that, when executed by the at least one processor, causes the electronic device to perform the method disclosed in the first aspect of the present invention.

[0039] The fourth aspect of the present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method disclosed in the first aspect of the present invention.

[0040] Compared with existing technologies, the transmission line graded reclosing decision-making method and system applicable to winter faults of the present invention can achieve the following beneficial effects:

[0041] 1) A prior distribution fitting method based on the beta-binomial distribution for historical reclosing power is proposed. This method determines the shape parameters of the beta-binomial distribution by statistically analyzing historical reclosing action data corresponding to different fault causes, thereby constructing a probabilistic model capable of quantifying the estimation uncertainty. Compared to simple historical success rate point estimation, this method more scientifically characterizes the estimation uncertainty caused by limited historical samples and differences in operating conditions, providing a robust prior knowledge base for subsequent decision-making.

[0042] 2) A Bayesian prediction framework that integrates real-time identification results with historical prior knowledge was constructed. That is, the probability distribution vector of fault causes identified by the time-frequency network and the re-closing power sample sampled from the corresponding prior beta-binomial distribution in Monte Carlo were substituted into the Bayesian conditional probability model for fusion calculation, and the expected value of the reclosing success probability under the current fault was output. This enabled a probabilistic and accurate prediction of the success probability of reclosing operation.

[0043] 3) A tiered reclosing decision-making system based on threshold comparison and fault scenario correlation is proposed. Specifically, a tiered criterion is set, with the overall average reclosing power in winter as the high threshold and the reclosing power due to the lowest fault cause as the low threshold. This forms a three-level decision-making mechanism of "direct reclosing - fault confirmation - direct blocking." For operating conditions entering Level II, a differentiated handling process closely related to the physical mechanism of the fault is preset. This enables dynamic optimization of reclosing execution and blocking strategies, improving the actual reclosing power and reducing the secondary impact on the power grid caused by reclosing to permanent faults.

[0044] The following describes in detail the transmission line graded reclosing decision method and system applicable to winter faults, with reference to the embodiments shown in the accompanying drawings and the reference numerals. Attached Figure Description

[0045] Figure 1 This is a flowchart of the steps in the transmission line graded reclosing decision method applicable to winter faults in this invention.

[0046] Figure 2 This is a flowchart of the transmission line graded reclosing decision-making method and system applicable to winter faults in this invention.

[0047] Figure 3 This is a flowchart of the probability prediction process for successful overlap of fault events in this invention.

[0048] Figure 4 This is a flowchart of the graded reclosing decision-making process in a specific implementation of the present invention.

[0049] Figure 5 This is a comparison chart showing the resynthetic power enhancement effect of the test sample set in a specific implementation of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0052] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0053] Figure 1 This is a flowchart of the steps in the transmission line graded reclosing decision method applicable to winter faults in this invention. Figure 2 This is a flowchart of the transmission line graded reclosing decision-making method and system applicable to winter faults in this invention.

[0054] Combination Figure 1 and Figure 2As shown, the first aspect of the present invention discloses a graded reclosing decision method for transmission lines applicable to winter faults, comprising the following steps: S1, acquiring transient waveform data when a fault occurs in the transmission line; S2, extracting time-frequency features from the transient waveform data based on a pre-constructed time-frequency network classification model, and probabilistically identifying the fault cause based on the time-frequency features, outputting a fault cause probability distribution, which represents the probability that the current fault belongs to each of the preset fault cause categories; S3, acquiring historical reclosing operation data, and constructing a priori probability distribution of reclosing power for each preset fault cause category based on the historical reclosing operation data; S4, calculating the reclosing success probability under the current fault through a probability fusion model based on the fault cause probability distribution and the reclosing power prior probability distribution; S5, comparing the calculated reclosing success probability with a preset threshold, and generating a graded reclosing decision instruction based on the comparison result.

[0055] The implementation process of the method disclosed in this invention can be summarized as follows: First, acquire transient waveform data when a fault occurs in the transmission line; then, extract and identify features from the transient waveform data to output a fault cause probability vector, which represents the probability of the current fault belonging to each of multiple predefined fault cause categories; before, after, or simultaneously with the aforementioned steps, based on historical reclosing operation data, for each type of fault cause among the multiple predefined fault causes, construct a prior probability distribution of its recombining power using a beta-binomial distribution to form a prior knowledge base; then, based on the aforementioned steps, for each type of fault cause identified by the time-frequency network classification model, extract multiple recombining power samples from its corresponding prior probability distribution; weight and synthesize the fault cause probability vector and the extracted recombining power samples using a Bayesian conditional probability model to obtain an estimated value of the recombining power under the current fault; finally, compare the obtained estimated value of the recombining power with a preset threshold, and perform a graded reclosing decision based on the comparison result.

[0056] In one embodiment, in step S1, the transient waveform data includes timing sampled values ​​of at least the following electrical quantity channels obtained from the fault recording device: phase A voltage, phase B voltage, phase C voltage, zero-sequence voltage, phase A current, phase B current, phase C current, and zero-sequence current. Furthermore, in step S1, based on the fault occurrence time, transient waveform data within a time window encompassing at least one cycle before the fault and at least two cycles after the fault are extracted. Specifically, the sampled values ​​of each channel are resampled to a uniform standard sampling frequency, and data within a fixed time window is extracted based on the fault occurrence time to form a standardized input matrix.

[0057] Specifically, fault transient waveform data is acquired from the line protection fault recording device. A data window is extracted, centered on the fault initiation time, with a total duration of one power frequency cycle before the fault and two power frequency cycles after the fault. That is, using the fault initiation time as the time reference point, waveform data of one power frequency cycle (20ms) is extracted before the fault and two power frequency cycles (40ms) are extracted after the fault, forming an analysis data window with a total duration of approximately 60ms. The three-phase voltage () within this time window is collected. ), zero-sequence voltage ( ), three-phase current ( ) and zero-sequence current ( The original sampling sequence consists of eight channels. To ensure consistency in subsequent time-frequency analysis, the original sampling sequence is standardized by resampling all channel data to a preset standard sampling frequency. In a preferred embodiment, the standard sampling frequency is set to 4300 Hz.

[0058] After the above processing, the resampled data from the eight channels are stacked according to the channel dimension, ultimately forming a standardized input matrix with a dimension of 8×L. Here, the number 8 represents the number of electrical quantity channels, and L represents the total number of sampling points for each channel within a 60ms time window at the standard sampling frequency.

[0059] In embodiments of the present invention, typical causes of transmission line faults in winter are categorized into a specific set of categories, which includes at least foreign objects, wildfires, wind deflection, icing-de-icing jumps, and icing-insulator skirt bridging faults. The classification is based on a systematic analysis of the physical mechanisms of transmission line faults under severe winter weather conditions, aiming to refine the characterization of the correlation between different external causes and internal electrical and mechanical responses.

[0060] In an embodiment of the present invention, the time-frequency network classification model includes: a time-frequency transformation layer, configured to perform time-frequency transformation on the input one-dimensional multi-channel transient waveform data to generate a two-dimensional time-frequency distribution map; and a backbone feature extraction network, which is connected to the time-frequency transformation layer, for performing feature extraction and classification on the two-dimensional time-frequency distribution map and outputting the probability distribution of the fault causes.

[0061] In an embodiment of the present invention, the time-frequency transformation layer is a time-frequency convolution layer including a real part convolution kernel and an imaginary part convolution kernel. It convolves the transient waveform data through the real part convolution kernel and the imaginary part convolution kernel to generate real part feature maps and imaginary part feature maps respectively, and calculates the modulus of the real part feature maps and the imaginary part feature maps to synthesize the two-dimensional time-frequency distribution map.

[0062] That is, the time-frequency transformation layer is a time-frequency convolutional layer used to receive the input one-dimensional multi-channel fault waveform signal and generate a two-dimensional time-frequency distribution map through differentiable operations. In a specific embodiment, the time-frequency convolutional layer is used as a preprocessing layer to receive the input standardized fault waveform signal with a dimension of 8×L and generate a two-dimensional time-frequency distribution map through differentiable operations.

[0063] The time-frequency convolutional layer contains real-part convolutional kernels and imaginary-part convolutional kernels, which are used to convolve the input signal to extract real-part and imaginary feature maps, respectively, and synthesize the two-dimensional time-frequency distribution map by calculating the modulus. The output of the time-frequency convolutional layer is defined by the following formula:

[0064]

[0065] In the formula, This is the time-frequency distribution output for the k-th channel; Let be the complex kernel function of the k-th channel; and These are the real part convolution kernel and the imaginary part convolution kernel, respectively; x is the input signal, and * represents the convolution operation.

[0066] In an embodiment of the present invention, the backbone feature extraction network is a convolutional neural network, used to perform deep feature extraction and classification on the two-dimensional time-frequency distribution map output by the preprocessing layer, outputting the probability distribution of fault causes, and obtaining the fault cause prediction probability distribution vector. This will be the final identification result output.

[0067] Specifically, the backbone feature extraction network includes, in sequence:

[0068] The first convolutional module is used to receive the two-dimensional time-frequency distribution map and perform preliminary feature extraction and nonlinear activation.

[0069] The second convolutional module, connected to the first convolutional module, is used for deeper feature extraction and includes downsampling operations;

[0070] The third and fourth convolutional modules are connected in sequence to further extract advanced semantic features;

[0071] An adaptive pooling layer, connected to the fourth convolutional module, is used to adjust the feature map to a fixed size.

[0072] A fully connected classification layer, connected to the adaptive pooling layer, is used to map the extracted features to a probability distribution of fault cause categories.

[0073] Preferably, the specific structure of the backbone feature extraction network is as follows:

[0074] The first convolutional module contains a convolutional layer with 16 kernels of size 15×8, followed by a batch normalization layer and a ReLU activation function;

[0075] The second convolutional module contains a convolutional layer with 32 kernels of size 3×16, followed by a batch normalization layer, a ReLU activation function, and a max pooling layer with a stride of 2.

[0076] The third convolutional module contains a convolutional layer with 64 kernels of size 3×32, followed by a batch normalization layer and a ReLU activation function;

[0077] The fourth convolutional module contains a convolutional layer with 128 kernels of size 3×64, followed by a batch normalization layer, a ReLU activation function, and an adaptive pooling layer;

[0078] A fully connected classification layer consists of an operation that flattens the features into a vector, two fully connected layers with ReLU activation functions connected in sequence, and an output layer for classification.

[0079] In an embodiment of the present invention, a prior probability distribution of reclosing power for each preset fault cause category is constructed based on historical reclosing operation data. This includes: for each preset fault cause category, counting the total number of historical reclosing actions and the number of historical successful reclosings from its historical fault event records; treating the reclosing power as a random variable and using a beta-binomial distribution as the prior probability distribution model for this random variable; determining two shape parameters of the beta-binomial distribution based on the historical successful reclosings and the total number of historical reclosing actions, where one shape parameter is positively correlated with the historical successful reclosings and the other shape parameter is positively correlated with the historical failed reclosings; calculating the expected value and confidence interval of the beta-binomial distribution based on the shape parameters, using the expected value as a point estimate of the reclosing power, and using the confidence interval to characterize the uncertainty of the estimate.

[0080] That is, in step S2, historical transmission line reclosing operation data are collected and organized, and the prior reclosing power distribution of various fault causes is fitted using a beta-binomial distribution, specifically including the following:

[0081] By analyzing historical reclosing operation records corresponding to different fault causes, the causes of each fault can be obtained. Number of successful historical overlaps Total number of actions And calculate its historical resynthesis power. .

[0082] The recombination power for the nth type of fault is considered as a random variable. The beta-binomial distribution is used as its prior probability distribution, and its probability density function is:

[0083] ;

[0084] In the formula: For beta function, shape parameter and Based on historical statistical data, they are represented as follows:

[0085] ;

[0086] ;

[0087] Furthermore, the expected value of the distribution is calculated based on the shape parameters. With variance The calculation formulas are as follows:

[0088] ;

[0089] .

[0090] Preferably, the expected value of the beta-binomial distribution is used as a point estimate of the recombinant power; its variance and confidence interval at a 95% confidence level are calculated to characterize the estimation uncertainty caused by the limited historical samples and differences in operating conditions.

[0091] In an embodiment of the present invention, the probability of successful re-combination under the current fault is calculated by a probabilistic fusion model, including: for each preset fault cause, independently extracting multiple re-combination power samples from the corresponding prior probability distribution of re-combination power using the Monte Carlo method, as a successful re-combination sample matrix; combining the fault cause probability distribution of the current fault belonging to each preset fault cause category with the extracted re-combination power sample matrix based on the full probability formula to obtain multiple predicted re-combination power values; and calculating the arithmetic mean of the multiple predicted re-combination power values ​​as the successful re-combination probability under the current fault.

[0092] like Figure 3 As shown, in step S4, the probability of successful reclosing under the current fault is calculated using a Bayesian conditional probability model by combining the fault cause identification probability result with the sampled value of the reclosing success rate distribution. Specifically, this includes the following:

[0093] First, for each cause of failure Independent samples were drawn from the corresponding beta-binomial distribution using the Monte Carlo sampling method. A sample, forming sample matrix , represented as:

[0094] ;

[0095] Then, for each sampling Calculate the predicted probability of successful overlap. It satisfies the law of total probability in the Bayesian conditional probability framework, and the specific calculation formula is as follows:

[0096] ;

[0097] In the formula: This represents the probability of the nth type of fault occurring, as output by the fault identification model. This represents the recombination power of the nth fault cause during the mth sampling.

[0098] Finally, M The arithmetic mean of the sample is used as the optimal point estimate, and the specific calculation formula is as follows:

[0099] ;

[0100] In the formula: This represents the probability of successful re-coincidence under the current fault. This value reflects the expected probability of the re-coincidence operation success rate under the condition of comprehensively considering historical prior uncertainties and real-time fault identification results.

[0101] like Figure 4 As shown, in step S5, the calculated reclosing success probability is compared with a priori set threshold, and a graded reclosing decision is formed according to different operating scenarios, specifically including the following:

[0102] 1) Set a priori threshold

[0103] High threshold Determined based on the average reconnection power across all fault causes during winter, representing the generally expected level of reconnection success; low threshold. Determined based on the lowest resynthesis power among various causes of failure in winter, representing the bottom line for success under the most unfavorable conditions.

[0104] 2) Establish a hierarchical decision-making mechanism

[0105] The calculated probability of successful fault re-coincidence A three-level decision-making mechanism is formed by comparing the results with a threshold, as follows:

[0106] like Greater than The decision is classified as Level I, and an immediate reclosing command is executed.

[0107] like Between and If the situation is between these points, it is classified as a Level II decision and the process for fault confirmation and differentiated handling begins.

[0108] like Less than or equal to The decision was classified as Level III, and the interlocking reclosing command was executed, along with the activation of the emergency response plan.

[0109] Preferably, the differentiated handling process in Level II decision-making is as follows:

[0110] Based on the specific fault cause determined by the fault cause identification results, a corresponding preset reclosing strategy is executed, the strategy including but not limited to:

[0111] For icing-de-icing jump faults, a time-delayed re-coincidence is implemented, with the delay time set based on the empirical value of 3-6 seconds for conductor stability.

[0112] For icing-insulator skirt bridging faults, perform fast reclosing;

[0113] In response to foreign object faults, the possibility of clearing the foreign object is assessed based on the fault arc. If the arc causes the foreign object to burn or drift away, reclosing is performed; otherwise, if the foreign object remains suspended or attached to the equipment, reclosing is blocked.

[0114] In response to wildfire malfunctions, depending on whether the fire situation is under control, if the fire has been effectively controlled or has naturally subsided, the reclosing will be executed; otherwise, if the fire is still continuing or spreading, the reclosing will be blocked.

[0115] For wind deflection faults, depending on whether the real-time wind speed in the fault section is lower than the line design wind speed, if the wind speed drops below the corresponding line design wind speed value, the reclosing operation is performed; otherwise, the reclosing is blocked.

[0116] like Figure 5 As shown, this is a comparison of the resynthetic power improvement effect of the test sample set in a specific embodiment of the present invention, as detailed below:

[0117] The resynthesis power was highest for icing-insulator skirt bridging faults, reaching 94.73%, which is 23.72% higher than the historical resynthesis power of 71.01%.

[0118] It also performed exceptionally well in wind-induced deflection faults, with a resynthesis power of 90.47%, an increase of 32.14% compared to the historical resynthesis power of 58.33%.

[0119] For the icing-de-icing jump fault, the resynthesis power was 66.67%, which is 29.78% higher than the historical resynthesis power of 36.89%.

[0120] For foreign object faults, the resynthesis success rate was 52.63%, an increase of 13.57% compared to the historical resynthesis success rate of 39.06%.

[0121] Regarding the wildfire failure, the resynthesis power was 50%, an increase of 20.27% compared to the historical resynthesis power of 29.73%.

[0122] The present invention also discloses a system for implementing the method disclosed in the first aspect of the present invention, comprising: a data acquisition module for acquiring transient waveform data and historical reclosing operation data when a fault occurs in a transmission line; a fault identification module for extracting time-frequency features from the transient waveform data based on a pre-constructed time-frequency network classification model, and probabilistically identifying the cause of the fault based on the time-frequency features, and outputting a fault cause probability distribution, which represents the probability that the current fault belongs to each of the preset fault cause categories; a priori modeling module for constructing a priori probability distribution of reclosing power for each preset fault cause category based on the historical reclosing operation data; a probability fusion module for calculating the reclosing success probability under the current fault based on the fault cause probability distribution of the current fault belonging to each preset fault cause category and the priori probability distribution of reclosing power for each preset fault cause category through a probability fusion model; and a decision execution module for comparing the calculated reclosing success probability with a preset threshold, generating and outputting a graded reclosing decision instruction based on the comparison result.

[0123] The present invention also discloses an electronic device comprising: at least one processor; and at least one memory storing a computer program that, when executed by the at least one processor, causes the electronic device to perform the method disclosed in the first aspect of the present invention.

[0124] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method disclosed in the first aspect of the present invention.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hierarchical reclosing decision-making method for transmission lines applicable to winter faults, characterized in that, Including the following steps: Acquire transient waveform data when a power transmission line fault occurs; Based on the time-frequency network classification model, time-frequency features are extracted from transient waveform data, and the causes of faults are probabilistically identified based on the time-frequency features. The probability distribution of fault causes is output, which represents the probability that the current fault belongs to each of the preset fault cause categories. Acquire historical reclosing operation data, and based on the historical reclosing operation data, construct the prior probability distribution of reclosing power for each preset fault cause category through beta-binomial distribution; Based on the probability distribution of the fault cause and the prior probability distribution of the re-synthesis power, the probability of successful re-synthesis under the current fault is calculated by a probability fusion model. The calculated reclosing success probability is compared with a preset threshold, and a graded reclosing decision instruction is generated based on the comparison result.

2. The method according to claim 1, characterized in that, The time-frequency network classification model includes: The time-frequency transformation layer is constructed to perform time-frequency transformation on the input one-dimensional multi-channel transient waveform data and generate a two-dimensional time-frequency distribution map. The backbone feature extraction network, which is connected to the time-frequency transformation layer, is used to extract and classify features from the two-dimensional time-frequency distribution map and output the probability distribution of the cause of the fault.

3. The method according to claim 2, characterized in that, The time-frequency transformation layer is a time-frequency convolutional layer including a real part convolution kernel and an imaginary part convolution kernel. It convolves the transient waveform data through the real part convolution kernel and the imaginary part convolution kernel to generate real part feature maps and imaginary part feature maps respectively, and calculates the modulus of the real part feature maps and the imaginary part feature maps to synthesize the two-dimensional time-frequency distribution map.

4. The method according to claim 1, characterized in that, Based on historical reclosing operation data, a prior probability distribution of reclosing power for each preset fault cause category is constructed using a beta-binomial distribution, including: For each type of preset fault cause, the total number of historical reclosing actions and the number of historical successful reclosings are counted from its historical fault event records. Treating the recombinant power as a random variable, we use the beta-binomial distribution as the prior probability distribution model for this random variable; Based on the historical number of successful overlaps and the total number of historical overlap actions, two shape parameters of the beta-binomial distribution are determined. One shape parameter is positively correlated with the historical number of successful overlaps, and the other shape parameter is positively correlated with the historical number of failed overlaps. The expected value and confidence interval of the beta-binomial distribution are calculated based on the shape parameters. The expected value is used as a point estimate of the recombinant power, and the confidence interval is used to characterize the uncertainty of the estimate.

5. The method according to claim 1, characterized in that, The calculation of the success probability of rendezvous under the current fault using a probabilistic fusion model includes: For each type of preset fault cause, multiple recombining power samples are independently extracted from the corresponding prior probability distribution of recombining power using the Monte Carlo method, and used as a recombining success sample matrix. The probability distribution of the current fault belonging to each preset fault cause category is combined with the extracted recombined power sample matrix and weighted based on the full probability formula to obtain multiple recombined power prediction values. The arithmetic mean of the multiple predicted re-synthesis power values ​​is calculated as the probability of successful re-synthesis under the current fault.

6. The method according to claim 1, characterized in that, The transient waveform data includes timing sampled values ​​of at least the following electrical quantity channels obtained from the fault recording device: phase A voltage, phase B voltage, phase C voltage, zero-sequence voltage, phase A current, phase B current, phase C current, and zero-sequence current. In the step of acquiring transient waveform data when a fault occurs in a transmission line, transient waveform data within a time window containing at least one cycle before the fault and at least two cycles after the fault are extracted, based on the time of the fault occurrence.

7. The method according to claim 1, characterized in that, The preset fault cause categories include at least two of the following: foreign object short circuit, wildfire, wind deflection, icing-de-icing jump, and icing-insulator skirt bridging.

8. The method according to claim 7, characterized in that, The preset threshold includes a high threshold and a low threshold; wherein the high threshold is determined based on the average recombining power of all preset fault cause categories, and the low threshold is determined based on the lowest recombining power among all preset fault cause categories. The step of generating a graded reclosing decision instruction based on the comparison results includes: If the probability of successful reclosing is greater than the high threshold, it is determined to be a Level I decision, and an immediate reclosing command is generated. If the probability of successful overlap is between the low threshold and the high threshold, it is determined to be a Level II decision, and a fault confirmation and differentiated handling instruction is generated. If the probability of successful reclosing is less than or equal to the low threshold, it is determined to be a Level III decision, and a reclosing blocking command is generated.

9. The method according to claim 8, characterized in that, The fault confirmation and differentiated handling instructions include: The dominant fault cause is determined based on the probability distribution of fault causes belonging to each preset fault cause category. Based on a preset fault cause-handling strategy mapping relationship, the reclosing control strategy corresponding to the dominant fault cause is executed, including: If the primary fault cause is icing-de-icing jump, then the reclosing control strategy is delayed reclosing, and the delay time is set according to the empirical value of conductor stability. If the primary fault cause is icing-insulator skirt bridging, then the reclosing control strategy is fast reclosing; If the primary fault cause is a short circuit caused by a foreign object, the reclosing control strategy is as follows: based on the assessment result of the possibility of the fault arc clearing the foreign object, decide whether to perform reclosing or block reclosing. If the primary cause of the fault is a wildfire, the reclosing control strategy is: to decide whether to perform reclosing or block reclosing based on whether the fire situation is under control. If the primary cause of the fault is wind deflection, the reclosing control strategy is as follows: depending on whether the real-time wind speed in the fault section is lower than the line design wind speed, decide whether to perform reclosing or block reclosing.

10. A system for implementing the method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire transient waveform data when a fault occurs in the transmission line and historical reclosing operation data; The fault identification module extracts time-frequency features from transient waveform data based on a pre-built time-frequency network classification model, and performs probabilistic identification of fault causes based on the time-frequency features, outputting a fault cause probability distribution, which represents the probability that the current fault belongs to each of the preset fault cause categories. The prior modeling module is used to construct the prior probability distribution of reclosing power for each preset fault cause category based on the historical reclosing operation data; The probability fusion module is used to calculate the success probability of re-combining under the current fault based on the probability distribution of the fault cause belonging to each preset fault cause category and the prior probability distribution of the re-combining power of each preset fault cause category through the probability fusion model. The decision execution module is used to compare the calculated reclosing success probability with a preset threshold, generate and output a graded reclosing decision instruction based on the comparison result.

Images