A Self-Healing Start-up Method for External Faults in Multi-Source Signal Distribution Networks Based on Self-Supervised Learning

Abstract

This invention relates to the field of power grid fault identification technology, and particularly to a self-healing startup method for external faults in distribution networks based on self-supervised learning. The method involves real-time acquisition of timestamped multi-source fault signals from distribution network terminals; matching and calculation of the multi-source fault signals to obtain initial startup values ​​for bus faults, backup automatic transfer switch failures, and line faults; after correcting and optimizing these initial startup values, they are weighted and fused using dynamic weights to obtain the multi-source self-healing startup value; pre-setting specific startup thresholds adapted to different operating conditions of the distribution network; comparing the multi-source self-healing startup value with the specific startup thresholds for the corresponding operating conditions to determine whether to activate the distribution network self-healing system. By implementing this invention, the self-healing startup of external faults can be made more accurate, adaptive, and intelligent.

Description

Technical Field

[0001] This invention relates to the field of power grid fault identification technology, and in particular to a self-healing startup method for external faults in multi-source signal distribution networks based on self-supervised learning. Background Technology

[0002] Traditional self-healing activation methods for external faults in distribution networks often rely on supervised learning models with pre-defined fixed logical rules or a small amount of labeled data. After collecting some fault signals, a activation value is calculated with fixed weights and then compared with a unified threshold to determine whether activation is necessary. This technology suffers from problems such as a single signal acquisition dimension, lack of time synchronization, poor model adaptability, low judgment accuracy, inability to adapt to complex operating conditions, susceptibility to false activation / missed activation, and difficulty in balancing the benefits and costs of self-healing. Summary of the Invention

[0003] This invention addresses the problems of existing technologies, such as single data acquisition dimension, lack of time synchronization, poor model adaptability, low judgment accuracy, and inability to adapt to complex working conditions. It provides a self-healing startup method for external faults in multi-source signal distribution networks based on self-supervised learning.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] This invention provides a self-healing startup method for external faults in a multi-source signal distribution network based on self-supervised learning. The method includes: real-time acquisition of timestamped multi-source fault signals via a distribution network terminal, wherein the multi-source fault signals include bus fault signals, backup automatic transfer switch failure signals, line fault signals, and long-term voltage loss signals; matching calculations are performed on the multi-source fault signals based on three independently trained self-supervised learning models, correspondingly outputting initial startup values ​​for bus faults, backup automatic transfer switch failures, and line faults; after correcting and optimizing the initial startup values ​​for bus faults, backup automatic transfer switch failures, and line faults, they are weighted and fused using dynamic weights to obtain a multi-source self-healing startup value; a preset startup threshold adapted to different operating conditions of the distribution network is established, and the multi-source self-healing startup value is compared with the startup threshold corresponding to the operating condition to determine whether to start the distribution network self-healing system.

[0006] Optionally, the distribution network terminal can collect multi-source fault signals with timestamps in real time, including: collecting bus fault signals output when a bus fault occurs through the bus protection device of the upstream switch station or substation; acquiring automatic transfer switching failure signals output when the automatic transfer switching fails due to transformer faults or loss of voltage on the external incoming line; collecting line fault signals corresponding to line faults by monitoring the line operating status through the distribution network terminal; continuously monitoring the line voltage through the distribution network terminal, and collecting long-term voltage loss signals when the detected line voltage is less than or equal to a preset voltage threshold and the duration is greater than or equal to a preset duration threshold; and synchronizing the timestamps of the collected bus fault signals, automatic transfer switching failure signals, line fault signals, and long-term voltage loss signals to ensure that the timing of each signal is correlated.

[0007] Optionally, based on three independently trained self-supervised learning models, matching calculations are performed on the multi-source fault signals respectively, outputting the initial start value of bus fault, the initial start value of backup automatic transfer and automatic switching failure, and the initial start value of line fault. This includes: inputting the timestamped bus fault signal and the long-term undervoltage signal into a pre-trained bus fault prediction model, and outputting the initial start value of bus fault; inputting the timestamped backup automatic transfer and automatic switching failure signal and the long-term undervoltage signal into a pre-trained backup automatic transfer and automatic switching failure prediction model, and outputting the initial start value of backup automatic transfer and automatic switching failure; inputting the timestamped line fault signal and the long-term undervoltage signal into a pre-trained line fault prediction model, and outputting the initial start value of line fault.

[0008] The construction process of the bus fault prediction model includes: collecting different types of bus fault signals and associated historical long-term voltage loss signals from the distribution network history, as well as historical outage parameters and switch operation cost parameters corresponding to each signal, and constructing a training sample set. The historical outage parameters include the bus fault outage duration before the self-healing system was activated and the power transfer duration after the self-healing system was activated. The switch operation cost parameters include the hardware loss cost and manual maintenance cost of a single switch operation. The difference between the historical bus fault outage duration before the self-healing system was activated and the power transfer duration after the self-healing system was activated is calculated. The comprehensive power outage benefit parameters are obtained by calculating and normalizing the hardware loss cost and manual maintenance cost in the switch operation cost parameters, and then normalizing them to obtain the comprehensive operation cost parameters. The difference between the comprehensive power outage benefit parameters and the comprehensive operation cost parameters is defined as the self-healing net benefit parameter. With maximizing the self-healing net benefit parameter as the optimization objective, a self-supervised training signal is constructed. A lightweight neural network structure is used to build the basic framework of the bus fault prediction model. The training sample set and the self-supervised training signal are input into the basic framework for iterative training until convergence is verified, resulting in the trained bus fault prediction model.

[0009] Optionally, after correcting and optimizing the initial start value for bus faults, the initial start value for backup automatic transfer / switching failures, and the initial start value for line faults, a weighted fusion is performed using dynamic weights to obtain multiple self-healing start values. This includes: calculating three coupling coefficients based on the distribution network topology and historical fault data, where the three coupling coefficients are the coupling coefficients between bus faults and line faults, backup automatic transfer / switching failures and line faults, and bus faults and backup automatic transfer / switching failures; based on these three coupling coefficients, the initial start values ​​for bus faults, backup automatic transfer / switching failures, and line faults are correlated and corrected to obtain the bus fault coupling correction start value, the backup automatic transfer / switching failure coupling correction start value, and the line fault coupling correction start value. Dynamic values ​​are obtained by collecting real-time operating parameters of the distribution network through the distribution network terminal and calculating correction factors based on the historical prediction deviation rate of the self-supervised learning model. The real-time operating parameters include distributed power source penetration rate, line real-time load rate, and bus real-time voltage deviation. Based on the correction factors, the bus fault coupling correction start-up value, the backup automatic transfer / switching failure coupling correction start-up value, and the line fault coupling correction start-up value are corrected a second time to obtain the bus fault start-up value, the backup automatic transfer / switching failure start-up value, and the line fault start-up value. Based on historical power outage parameters, switch operation cost parameters, three coupling coefficients, and three correction factors, dynamic weights are calculated, and the bus fault start-up value, the backup automatic transfer / switching failure start-up value, and the line fault start-up value are weighted and summed to obtain the multi-source self-healing start-up value.

[0010] The key feature is that, based on the distribution network topology and historical fault data, three coupling coefficients are calculated, including: acquiring the connection methods and electrical distances of buses, lines, and switching equipment to construct the distribution network topology; collecting historical fault data and filtering complete fault records that trigger line faults after bus faults, line faults triggered after backup automatic transfer failures, and backup automatic transfer failures triggered after bus faults to obtain three sets of associated fault samples; and calculating the probability of subsequent faults occurring within a preset time window after the occurrence of a preceding fault for each of the three sets of associated fault samples to obtain the probability of subsequent faults occurring within a preset time window after the occurrence of a preceding fault. The initial correlation between bus faults and line faults, the initial correlation between backup automatic transfer switch failure and line faults, and the initial correlation between bus faults and backup automatic transfer switch failures are determined. Based on the electrical distance in the distribution network topology and combined with the average electrical distance in the distribution network topology, the initial correlation between bus faults and line faults, the initial correlation between backup automatic transfer switch failure and line faults, and the initial correlation between bus faults and backup automatic transfer switch failures are corrected to obtain the coupling coefficients between bus faults and line faults, the coupling coefficients between backup automatic transfer switch failures and line faults, and the coupling coefficients between bus faults and backup automatic transfer switch failures.

[0011] The process involves collecting real-time operating parameters of the distribution network via distribution network terminals and calculating correction factors based on the historical prediction deviation rates of self-supervised learning models. This includes: collecting distributed power source penetration rate, line load rate, and bus voltage deviation in real-time via distribution network terminals; normalizing and weighting the distributed power source penetration rate, line load rate, and bus voltage deviation to obtain a real-time operating condition correction factor; obtaining the deviation data between the historical prediction start-up values ​​of the three self-supervised learning models and the actual self-healing effects of the corresponding scenarios, calculating the historical prediction deviation rates of the three self-supervised learning models, and calculating their arithmetic mean to obtain three average historical prediction deviation rates; and combining the real-time operating condition correction factor with the three average historical prediction deviation rates to calculate three correction factors.

[0012] The calculation of dynamic weights is based on historical outage parameters, switch operation cost parameters, three coupling coefficients, and three correction factors. This includes: calculating the self-healing benefit coefficient for bus fault scenarios based on historical outage parameters (e.g., outage duration before the self-healing system was activated, power transfer duration after activation), and the bus fault scenario correction factor; calculating the self-healing cost coefficient for bus fault scenarios based on switch operation cost parameters (e.g., hardware loss cost and manual maintenance cost per switch operation), and the coupling coefficient between bus faults and line faults; calculating the weight base for bus fault scenarios based on the self-healing benefit coefficient and self-healing cost coefficient; calculating the self-healing benefit coefficient for backup power transfer failure scenarios based on historical outage parameters (e.g., outage duration before the self-healing system was activated, power transfer duration after activation), and the backup power transfer failure scenario correction factor; and calculating the self-healing benefit coefficient for backup power transfer failure scenarios based on switch operation cost parameters (e.g., hardware loss cost and manual maintenance cost per switch operation). The self-healing cost coefficient for the backup automatic transfer switching failure scenario is calculated by combining the coupling coefficient between backup automatic transfer switching failure and line fault. Based on the self-healing benefit coefficient and self-healing cost coefficient of the backup automatic transfer switching failure scenario, the weight base for the backup automatic transfer switching failure scenario is calculated. According to the historical outage parameters (historical outage duration before the self-healing system was activated and the power transfer duration after the self-healing system was activated), and combined with the line fault scenario correction factor, the self-healing benefit coefficient for the line fault scenario is calculated. According to the switch operation cost parameters (hardware loss cost per switch operation and manual maintenance cost), and combined with the coupling coefficient between bus fault and backup automatic transfer switching failure, the self-healing cost coefficient for the line fault scenario is calculated. Based on the self-healing benefit coefficient and self-healing cost coefficient of the line fault scenario, the weight base for the line fault scenario is calculated. The weight bases for the bus fault scenario, the backup automatic transfer switching failure scenario, and the line fault scenario are normalized to obtain three dynamic weights that sum to 1.

[0013] Specifically, the bus fault initiation value, the backup automatic transfer failure initiation value, and the line fault initiation value are weighted and summed to obtain a multi-source self-healing initiation value. This includes: obtaining three dynamic weights for bus fault, backup automatic transfer failure, and line fault scenarios; and weighting and summing the bus fault initiation value, backup automatic transfer failure initiation value, and line fault initiation value according to the three dynamic weights to obtain a multi-source self-healing initiation value.

[0014] Optionally, a specific start-up threshold adapted to different operating conditions of the distribution network is preset. The multi-source self-healing start-up value is compared with the specific start-up threshold of the corresponding operating condition to determine whether to start the distribution network self-healing system. This includes: pre-classifying the distribution network operating conditions based on different distribution network operating condition parameters, wherein the distribution network operating conditions include at least light load conditions, heavy load conditions, high distributed power supply access conditions, and voltage deviation critical conditions; for each type of distribution network operating condition, a specific start-up threshold is preset based on the historical self-healing effect data of the distribution network and the engineering safety operation standards; the distribution network operating conditions are matched according to the real-time operating condition parameters of the distribution network, and the corresponding specific start-up threshold is obtained; the calculated multi-source self-healing start-up value is compared with the corresponding specific start-up threshold. If the multi-source self-healing start-up value is greater than or equal to the specific start-up threshold, it is determined that the self-healing start-up condition is met; if the multi-source self-healing start-up value is less than the specific start-up threshold, it is determined that the start-up condition is not met.

[0015] By implementing this invention, it is possible to collect multi-source fault signals with timestamps in real time through the distribution network terminal. The multi-source fault signals include bus fault signals, automatic transfer switch failure signals, line fault signals, and long-term voltage loss signals, covering the core scenarios of external faults in the distribution network. This ensures the correlation of signal timing and provides complete and accurate raw data support for subsequent fault type identification, avoiding judgment deviations caused by missing signals or disordered timing.

[0016] By implementing this invention, it is possible to perform matching calculations on the multi-source fault signals based on three independently trained self-supervised learning models, and output the initial start value of the bus fault, the initial start value of the backup automatic transfer and automatic disconnection failure, and the initial start value of the line fault respectively. Taking advantage of the self-supervised learning, which does not require manual data labeling, the training cost is reduced. The independent model design can be specifically adapted to the characteristics of different fault types, improve the calculation accuracy of the initial start value, and lay the foundation for subsequent optimization.

[0017] By implementing this invention, it is possible to correct and optimize the initial start value of the bus fault, the initial start value of the backup automatic transfer and automatic disconnection failure, and the initial start value of the line fault, and then perform weighted fusion through dynamic weights to obtain multiple self-healing start values. Coupled correction solves the correlation interference problem of different fault types, secondary correction adapts to real-time operating condition changes and model errors, and dynamic weights reflect the differences in self-healing benefits and costs of different fault scenarios, ultimately improving the reliability and adaptability of multiple self-healing start values.

[0018] By implementing this invention, it is possible to preset exclusive start-up thresholds adapted to different operating conditions of the distribution network, compare the multiple self-healing start-up values ​​with the exclusive start-up thresholds for the corresponding operating conditions, and determine whether to start the distribution network self-healing system. This avoids the inadequacy of a single threshold for adapting to complex operating conditions, and ensures that the self-healing start-up judgment meets engineering safety standards and matches actual self-healing needs in different scenarios such as light load, heavy load, and high distributed power supply access, thereby reducing false start-up or missed start-up situations.

[0019] In summary, by implementing this invention, we can achieve more precise, adaptive, and intelligent self-healing startup for external network faults, significantly improve the accuracy, adaptability, and reliability of self-healing startup judgment, reduce the risk of false startup / missed startup, and optimize the balance between self-healing benefits and operating costs. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a self-healing startup method for external faults in a multi-source signal distribution network based on self-supervised learning, provided by the present invention.

[0021] Figure 2 This invention provides a flowchart illustrating the process of calculating and obtaining the correction factor in a self-healing startup method for external faults in a multi-source signal distribution network based on self-supervised learning. Detailed Implementation

[0022] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0024] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0025] Example 1, as Figure 1 As shown, this embodiment of the invention provides a self-healing startup method for external faults in multi-source signal distribution networks based on self-supervised learning, including:

[0026] S100: Real-time acquisition of multi-source fault signals with timestamps through the distribution network terminal, wherein the multi-source fault signals include bus fault signals, automatic transfer switch failure signals, line fault signals, and long-term voltage loss signals;

[0027] S200: Based on three independently trained self-supervised learning models, the multi-source fault signals are matched and calculated respectively, and the initial start value of bus fault, the initial start value of backup automatic transfer and automatic disconnection failure and the initial start value of line fault are output respectively.

[0028] S300: After correcting and optimizing the initial start value of the bus fault, the initial start value of the backup automatic transfer and automatic disconnection failure and the initial start value of the line fault, the multi-source self-healing start value is obtained by weighted fusion through dynamic weights.

[0029] S400: Preset exclusive start thresholds for different operating conditions of the distribution network, compare the multiple self-healing start values ​​with the exclusive start thresholds for the corresponding operating conditions, and determine whether to start the distribution network self-healing system.

[0030] In step S100 of this application embodiment, the multi-source fault signals with timestamps are collected in real time through the distribution network terminal, including:

[0031] The bus fault signal output when a bus fault occurs is collected through the bus protection device of the upstream switch station or substation.

[0032] The automatic transfer switch (ATS) device in the distribution network is used to obtain the automatic transfer failure signal output when the automatic transfer operation fails due to transformer fault or loss of voltage on the incoming line outside the station.

[0033] The power distribution network terminal monitors the line operation status and collects line fault signals corresponding to the line's own faults.

[0034] The distribution network terminal continuously monitors the line voltage. When the detected line voltage is less than or equal to a preset voltage threshold and the duration is greater than or equal to a preset duration threshold, a long-term voltage loss signal is collected.

[0035] The collected bus fault signals, automatic transfer switch failure signals, line fault signals, and long-term voltage loss signals are time-stamped and synchronized to ensure that the timing of each signal is correlated.

[0036] In this embodiment of the application, the purpose of step S100 is to provide complete, accurate and time-consistent basic data support for the self-healing start-up judgment of the distribution network, so as to ensure the reliability of subsequent self-supervised learning model analysis and start-up decision.

[0037] Specifically, to achieve the above objectives, the first step is to collect the bus fault signal output when a bus fault occurs through the bus protection device of the upstream switching station or substation. The collection of bus fault signals relies on the bus protection device of the upstream switching station or substation. As the core hub of the distribution network, once a fault such as a short circuit occurs, the bus protection device will detect the fault characteristics immediately and output the corresponding bus fault signal. The distribution network terminal can directly capture this signal to complete the collection, ensuring timely acquisition of fault information at the bus level.

[0038] The second step requires acquiring the automatic transfer switch failure signal (ATS) output by the distribution network's backup automatic transfer device (SAST). This signal is generated when a transformer fault or an external incoming line voltage loss causes the automatic transfer to fail. When a transformer fault or an external incoming line voltage loss occurs in the distribution network, the ATS automatically initiates an automatic transfer action, attempting to switch the power supply path to ensure power supply. If the automatic transfer action fails, the ATS outputs a corresponding automatic transfer failure signal, which the distribution network terminal collects to understand the fault scenario information indicating the power supply switching failure.

[0039] The third step requires monitoring the line's operating status through the distribution network terminal and collecting line fault signals corresponding to the line's own faults.

[0040] The acquisition of line fault signals is directly completed by the distribution network terminal. The distribution network terminal monitors the operating status of the line in real time, including parameters such as current, voltage, and phase. When the line experiences its own faults such as overload, short circuit, or grounding, it will exhibit specific changes in operating parameters. After capturing these changes, the distribution network terminal can acquire the corresponding line fault signal.

[0041] The fourth step requires continuous monitoring of the line voltage via the distribution network terminal. When the detected line voltage is less than or equal to a preset voltage threshold and the duration is greater than or equal to a preset duration threshold, a prolonged voltage loss signal is collected. The distribution network terminal continuously monitors the line voltage. The voltage and duration thresholds are preset, for example, a voltage threshold of 0.4kV and a duration threshold of 3 seconds. When the terminal detects that the line voltage is continuously lower than or equal to 0.4kV, and this state is maintained for 3 seconds or more, the collection mechanism is automatically triggered to record the prolonged voltage loss signal, thereby determining whether the line is in a state of continuous power loss.

[0042] The fifth step requires timestamping the collected bus fault signals, automatic transfer switch failure signals, line fault signals, and prolonged power outage signals to ensure the timing correlation of each signal. Since the first four types of signals are collected from different sources, there may be slight differences in the collection time. Therefore, all collected signals need to be timestamped to calibrate the collection time of each type of signal using a unified time standard. This ensures that the bus fault signals, automatic transfer switch failure signals, line fault signals, and prolonged power outage signals are correlated on the timeline, clarifying which signals were caused by the same fault event, and providing a timing basis for subsequent analysis of fault correlation.

[0043] In step S200 of this application embodiment, based on three independently trained self-supervised learning models, matching calculations are performed on the multi-source fault signals respectively, correspondingly outputting the initial start value of bus fault, the initial start value of backup automatic transfer and automatic disconnection failure, and the initial start value of line fault, including:

[0044] The bus fault signal after timestamp synchronization and the long-term undervoltage signal are input into the pre-trained bus fault prediction model, and the initial start value of the bus fault is output.

[0045] The backup self-transfer and self-switching failure signal after the timestamp synchronization and the long-term undervoltage signal are input into the pre-trained backup self-transfer and self-switching failure prediction model, and the initial start value of backup self-transfer and self-switching failure is output.

[0046] The time-stamped line fault signal and the long-term undervoltage signal are input into the pre-trained line fault prediction model, and the initial start value of the line fault is output.

[0047] In this embodiment, the purpose of step S200 is to transform multi-source fault signals into quantified initial start-up values ​​using a targeted self-supervised learning model, providing basic data for subsequent correction, optimization, and weighted fusion. Three independent models are matched to three types of fault scenarios, accurately capturing the characteristic correlations of different faults, avoiding signal interference from different fault types, ensuring that the calculation of start-up values ​​corresponding to various faults is more targeted and accurate, and providing a quantitative basis for determining the self-healing start-up of the distribution network.

[0048] To achieve the above steps, three independent self-supervised learning models need to be constructed first: a bus fault prediction model, a backup automatic transfer and automatic disconnection failure prediction model, and a line fault prediction model. Each model undergoes specialized pre-training. The training process aims to maximize the self-healing net benefit parameter. A training sample set is constructed by combining historical fault signals, power outage parameters, and switch operation cost parameters. The model is iteratively trained using self-supervised training signals until it reaches convergence.

[0049] First, the bus fault signal after timestamp synchronization and the long-term undervoltage signal need to be input into the pre-trained bus fault prediction model to output the initial start value of the bus fault.

[0050] In step S200 of this application embodiment, the process of constructing the bus fault prediction model includes:

[0051] Collect different types of bus fault signals, associated historical long-term voltage loss signals, and historical outage parameters and switch operation cost parameters corresponding to each signal in the history of the distribution network to build a training sample set. The historical outage parameters include the bus fault outage duration when the self-healing system was not activated and the power transfer duration after the self-healing system was activated. The switch operation cost parameters include the hardware loss cost and manual maintenance cost of a single switch operation.

[0052] The difference between the bus fault outage duration when the self-healing system was not activated and the power transfer duration after the self-healing system was activated in the historical outage parameters is calculated and then normalized to obtain the comprehensive outage benefit parameters.

[0053] The hardware loss cost and the manual operation and maintenance cost in the switch operation cost parameters are summed and then normalized to obtain the comprehensive operation cost parameters.

[0054] The difference between the comprehensive power outage revenue parameter and the comprehensive operating cost parameter is defined as the self-healing net revenue parameter. With the maximization of the self-healing net revenue parameter as the optimization objective, a self-supervised training signal is constructed.

[0055] A lightweight neural network structure is used to build the basic framework of the bus fault prediction model;

[0056] The training sample set and the self-supervised training signal are input into the basic framework for iterative training until the verification convergence, thus obtaining the trained bus fault prediction model.

[0057] In this embodiment, training the bus fault prediction model aims to construct a targeted, benefit-oriented bus fault prediction model. This ensures the model can accurately process synchronized bus fault signals and long-term undervoltage signals, outputting an initial bus fault initiation value that reflects the value of self-healing startup. By combining historical fault data, outage impacts, and operational costs, and with maximizing net self-healing benefits as the optimization objective, the model's calculated initial initiation value conforms to fault judgment logic and matches the actual benefit requirements of distribution network operation and maintenance. This provides a scientific and practical quantitative basis for subsequent distribution network self-healing startup decisions.

[0058] To achieve the above objectives, the first step is to collect different types of bus fault signals, associated historical long-term voltage loss signals, and historical outage parameters and switch operation cost parameters corresponding to each signal in the history of the distribution network, and build a training sample set. The historical outage parameters include the bus fault outage duration when the self-healing system was not activated in the past and the power transfer duration after the self-healing system was activated in the past. The switch operation cost parameters include the hardware loss cost of a single switch operation and the manual operation and maintenance cost.

[0059] This step involves comprehensively collecting historical data from the distribution network, covering three core dimensions. First, fault-related signals, including signals from different types of bus faults and historical long-term power outage signals associated with these bus faults, ensuring the model can learn the core signal characteristics at the time of the fault. Second, historical outage parameters, specifically including the duration of bus fault outages before the self-healing system was activated and the duration of power transfer after the self-healing system was activated; these two types of parameters directly reflect the impact of the self-healing system activation on the outage status. Third, switch operation cost parameters, covering the hardware wear and tear cost and manual maintenance cost of a single switch operation, reflecting the actual maintenance investment during the self-healing activation process. These collected signals and parameters are then organized and categorized according to their correspondences to form a complete training sample set, providing sufficient basic data for model training.

[0060] For example, the historical data collected for a certain type of bus fault are as follows: the bus fault signal is a three-phase short circuit characteristic signal, the associated long-term voltage loss signal is a voltage of 0.3kV lasting for 4 seconds, the power outage duration is 2 hours when self-healing is not activated, the power supply duration is 0.5 hours after self-healing is activated, the hardware loss cost of a single switch operation is 500 yuan, and the manual maintenance cost is 300 yuan. These data together constitute a complete training sample.

[0061] The second step is to calculate the difference between the bus fault outage duration when the self-healing system was not activated and the power transfer duration after the self-healing system was activated in the historical outage parameters, and then normalize the result to obtain the comprehensive outage benefit parameters.

[0062] This involves specifically processing historical outage parameters. First, the difference between the outage duration of bus faults before the self-healing system was activated and the power transfer duration after the self-healing system was activated is calculated. This difference directly reflects the improvement in outage time after the self-healing system was activated; the larger the difference, the more significant the effect of the self-healing system in shortening outage duration. Then, this difference is normalized to eliminate the impact of differences in outage duration values ​​across different scenarios, transforming it into a unified quantitative indicator. Finally, a comprehensive outage benefit parameter is obtained, quantifying the benefits of activating the self-healing system in improving outage performance.

[0063] For example, if the power outage duration without self-healing in a certain sample is 2 hours, and the power transfer duration after self-healing is activated is 0.5 hours, the difference is 1.5 hours; assuming that the maximum reduction in power outage time in the historical samples is 3 hours and the minimum is 0 hours, by normalization calculation (1.5-0) / (3-0)=0.5, the comprehensive power outage benefit parameter of this sample is 0.5.

[0064] The third step is to sum the hardware loss cost and the manual operation and maintenance cost in the switch operation cost parameters, and then normalize them to obtain the comprehensive operation cost parameters.

[0065] The comprehensive operating cost parameter is used to quantify the operating costs required to activate the self-healing system. First, the two indicators in the switch operation cost parameter are summed: the hardware wear and tear cost of a single switch operation plus the manual maintenance cost, yielding the total direct cost of a single self-healing operation. Then, this sum is normalized, mapped to a range of 0 to 1, eliminating absolute cost differences caused by different regions and equipment types, resulting in the comprehensive operating cost parameter.

[0066] For example, in a certain sample, the hardware loss cost for a single switch operation is 500 yuan, and the manual maintenance cost is 300 yuan, totaling 800 yuan; in historical samples, the maximum total cost for a single operation is 1600 yuan, and the minimum is 200 yuan. The normalized calculation is (800-200) / (1600-200)≈0.428, so the comprehensive operation cost parameter for this sample is 0.428.

[0067] The fourth step is to define the difference between the comprehensive power outage revenue parameter and the comprehensive operating cost parameter as the self-healing net revenue parameter, and construct a self-supervised training signal with the goal of maximizing the self-healing net revenue parameter.

[0068] Self-supervised training signals are the core optimization direction for model training, directly determining the model's decision-making logic. The difference between the previously calculated comprehensive power outage benefit parameter and the comprehensive operating cost parameter is defined as the self-healing net benefit parameter, which essentially quantifies the net value of activating the self-healing system. The optimization objective of model training is set to maximize the self-healing net benefit parameter, that is, to enable the model to learn under what signal characteristics activating self-healing will yield the greatest net benefit. Based on this, a self-supervised training signal is constructed to provide clear optimization guidance for model iteration.

[0069] Continuing with the example above, with a comprehensive power outage benefit parameter of 0.5 and a comprehensive operating cost parameter of 0.428, the corresponding self-healing net benefit parameter is 0.5 - 0.428 = 0.072. During model training, the goal is to make the self-healing net benefit parameter in this type of scenario approach a larger value.

[0070] The fifth step requires the use of a lightweight neural network structure to build the basic framework of the bus fault prediction model;

[0071] A lightweight neural network structure is used to build the basic framework of the bus fault prediction model. The lightweight structure is chosen to balance computational efficiency and deployment ease, ensuring the model can run quickly on distribution network terminals or related equipment, meeting the real-time requirements of distribution network fault handling. This framework must have an input layer, hidden layers, and an output layer. The input layer receives feature data from the bus fault signal and the long-term undervoltage signal. The hidden layer mines deep correlation features of the signals through operations such as convolution, pooling, or fully connected layers. The output layer outputs the quantized initial start value of the bus fault. For example, a three-layer fully connected neural network can be used as the basic framework. The input layer dimension matches the sum of the feature dimensions of the two types of signals, the hidden layer has two neuron nodes, and the output layer has one node, corresponding to a single quantized initial start value.

[0072] The sixth step requires inputting the training sample set and the self-supervised training signal into the basic framework for iterative training until the verification convergence is achieved, thus obtaining the trained bus fault prediction model.

[0073] The pre-assembled training sample set and the constructed self-supervised training signal are input into the basic framework for iterative model training. During training, the model continuously adjusts its internal parameters, optimizing the learning of the correlation between signal features and self-healing net benefit parameters. After each iteration, the model's output accuracy and stability are verified using a validation set. When the model's prediction error on the validation set is below a preset threshold, and its performance shows no significant improvement after multiple iterations, the model is considered to have reached validation convergence, training is stopped, and the trained bus fault prediction model is finally obtained. For example, if the preset model validation error threshold is 0.05, after 500 iterations, the model's average prediction error on the validation set stabilizes at 0.03, and the error does not significantly decrease in the subsequent 100 iterations, the model training is complete and can be used to calculate the initial initiation value for subsequent bus faults.

[0074] Next, the bus fault signal and the long-term undervoltage signal after timestamp synchronization are input into the pre-trained bus fault prediction model, and the initial start value of the bus fault is output; the backup automatic transfer and automatic switching failure signal and the long-term undervoltage signal after timestamp synchronization are input into the pre-trained backup automatic transfer and automatic switching failure prediction model, and the initial start value of the backup automatic transfer and automatic switching failure is output; the line fault signal and the long-term undervoltage signal after timestamp synchronization are input into the pre-trained line fault prediction model, and the initial start value of the line fault is output.

[0075] The backup automatic transfer and automatic disconnection failure prediction model is constructed by collecting historical backup automatic transfer and automatic disconnection failure signals, associated long-term voltage loss signals, and corresponding historical power outage parameters and switch operation cost parameters to form a training sample set. Subsequently, the training signal is constructed with the goal of maximizing the self-healing net benefit parameter according to the same benefit and cost parameter calculation logic, and is obtained after convergence by training a lightweight neural network.

[0076] When acquiring the line fault prediction model, the training data is replaced with historical line fault signals, associated long-term voltage loss signals and corresponding power outage parameters and switch operation cost parameters. The remaining construction steps are similar to the previous two models.

[0077] Furthermore, the bus fault signal and the long-term undervoltage signal are input into the bus fault prediction model, the automatic transfer switch failure signal and the long-term undervoltage signal are input into the automatic transfer switch failure prediction model, and the line fault signal and the long-term undervoltage signal are input into the line fault prediction model, so that they are matched one by one to avoid confusion.

[0078] Each pre-trained model performs feature matching and quantization calculations on the two sets of input related signals, and outputs the corresponding initial start value for bus fault, initial start value for automatic transfer switch failure, and initial start value for line fault, respectively.

[0079] In step S300 of this application embodiment, after correcting and optimizing the initial start value of the bus fault, the initial start value of the backup automatic transfer and automatic disconnection failure, and the initial start value of the line fault, a weighted fusion is performed using dynamic weights to obtain a multi-source self-healing start value, including:

[0080] Based on the distribution network topology and historical fault data, three coupling coefficients are calculated. These three coupling coefficients are the coupling coefficient between bus fault and line fault, the coupling coefficient between backup automatic transfer failure and line fault, and the coupling coefficient between bus fault and backup automatic transfer failure.

[0081] Based on the three coupling coefficients, the initial start value of bus fault, the initial start value of backup automatic transfer and automatic disconnection failure and the initial start value of line fault are correlated and corrected to obtain the bus fault coupling correction start value, the backup automatic transfer and automatic disconnection failure coupling correction start value and the line fault coupling correction start value.

[0082] The real-time operating parameters of the distribution network are collected by the distribution network terminal, and the correction factor is calculated by combining the historical prediction deviation rate of the self-supervised learning model. The real-time operating parameters include distributed power penetration rate, line real-time load rate, and bus real-time voltage deviation.

[0083] Based on the correction factor, the bus fault coupling correction start value, the backup automatic transfer and automatic disconnection failure coupling correction start value, and the line fault coupling correction start value are corrected a second time to obtain the bus fault start value, the backup automatic transfer and automatic disconnection failure start value, and the line fault start value.

[0084] Based on historical power outage parameters, switch operation cost parameters, three coupling coefficients, and three correction factors, dynamic weights are calculated, and the bus fault initiation value, the backup automatic transfer and automatic disconnection failure initiation value, and the line fault initiation value are weighted and summed to obtain the multi-source self-healing initiation value.

[0085] In this embodiment, the purpose of step S300 is to optimize the initial initiation values ​​of the three types of faults into multiple self-healing initiation values ​​that accurately reflect the actual fault situation and self-healing value of the distribution network through two rounds of correction and dynamic weighted fusion. The coupling coefficient correction solves the problem of correlation interference between different faults, the correction factor adapts to the deviation between real-time operating conditions and model predictions, and the dynamic weights adjust the influence weights of various faults in combination with revenue and cost, ultimately making the multiple self-healing initiation values ​​more in line with the actual operating needs of the distribution network, and providing a more reliable quantitative basis for self-healing initiation judgment.

[0086] To achieve the above objectives, it is first necessary to calculate three coupling coefficients based on the distribution network topology and historical fault data. These three coupling coefficients are the coupling coefficient between bus faults and line faults, the coupling coefficient between automatic transfer switch failure and line faults, and the coupling coefficient between bus faults and automatic transfer switch failures.

[0087] In step S300 of this application embodiment, three coupling coefficients are calculated based on the distribution network topology and historical fault data of the distribution network, including:

[0088] Obtain the connection methods and electrical distances of busbars, lines, and switchgear to construct the distribution network topology.

[0089] Collect historical fault data of distribution network, and filter complete fault records that trigger line faults after bus faults, line faults after backup automatic transfer failures, and backup automatic transfer failures after bus faults to obtain three sets of related fault samples.

[0090] For the three sets of associated fault samples, the probability of subsequent faults occurring within a preset time window after the occurrence of the preceding fault is calculated respectively, so as to obtain the initial correlation degree between bus fault and line fault, the initial correlation degree between backup automatic transfer failure and line fault, and the initial correlation degree between bus fault and backup automatic transfer failure.

[0091] Based on the electrical distance in the distribution network topology and combined with the average electrical distance in the distribution network topology, the initial correlation between bus faults and line faults, the initial correlation between backup automatic transfer switch failure and line faults, and the initial correlation between bus faults and backup automatic transfer switch failures are corrected to obtain the coupling coefficients between bus faults and line faults, backup automatic transfer switch failures and line faults, and bus faults and backup automatic transfer switch failures.

[0092] In this embodiment, calculating the three coupling coefficients quantifies the degree of correlation between the three types of faults, providing a basis for subsequent correlation correction of the initial startup value.

[0093] The first step is to obtain the connection methods and electrical distances of the busbars, lines, and switchgear to construct the distribution network topology.

[0094] First, core information about the busbars, lines, and switchgear in the distribution network is collected, with a focus on clarifying the connection methods of each device, such as direct connection between the busbar and line via a circuit breaker, or series connection between the line and switchgear. Simultaneously, the electrical distances between each device are measured or retrieved, typically in kilometers. Based on this information, a distribution network topology is constructed in graphical or data-driven form, clearly presenting the connection logic and electrical distances between devices.

[0095] For example, a distribution network area has 2 busbars, 3 lines, and 5 switching devices. Busbar 1 and line 1 are directly connected through switch 1, with an electrical distance of 0.8 kilometers; busbar 2 and line 2 are indirectly connected through switch 3, with an electrical distance of 1.2 kilometers. Based on this, a topology network including device connection relationships and electrical distances is constructed as the distribution network topology network.

[0096] The second step is to collect historical fault data of the distribution network, and then filter out complete fault records that trigger line faults after bus faults, line faults after backup automatic transfer failures, and backup automatic transfer failures after bus faults to obtain three sets of related fault samples.

[0097] Specifically, it is necessary to collect past fault records of the distribution network. These records must include information such as fault type, occurrence time, and associated equipment. Then, they are filtered according to three types of associated scenarios: the first type filters complete records of line faults occurring within a preset time window after a bus fault occurs; the second type filters complete records of line faults occurring within a preset time window after a backup automatic transfer switch failure occurs; and the third type filters complete records of backup automatic transfer switch failures within a preset time window after a bus fault occurs, forming three independent sets of associated fault samples.

[0098] For example, with a preset time window of 10 minutes, 5 years of historical fault data were collected. Among them, there were 32 records of line faults triggered within 10 minutes after a bus fault, 18 records of line faults triggered within 10 minutes after a backup automatic transfer switch failure, and 25 records of backup automatic transfer switch failures triggered within 10 minutes after a bus fault. These three types of records constitute three sets of related fault samples.

[0099] The third step is to calculate the probability of subsequent faults occurring within a preset time window after the occurrence of a preceding fault for the three associated fault sample sets, so as to obtain the initial correlation between bus faults and line faults, the initial correlation between automatic transfer switch failure and line faults, and the initial correlation between bus faults and automatic transfer switch failures.

[0100] For each associated fault sample set, the probability of subsequent faults occurring within a preset time window after the occurrence of the preceding fault is obtained by dividing the number of records of associated faults in the associated fault sample set by the total number of records of the corresponding preceding faults. This probability is the initial correlation degree.

[0101] For example, the first type of associated fault sample set has 32 associated records, while the total number of historical bus fault records is 160. Therefore, the initial correlation degree between bus faults and line faults is 32 / 160 = 0.2. The second type of associated fault sample set has 18 associated records, while the total number of automatic transfer switch failure records is 90. Therefore, the initial correlation degree between automatic transfer switch failure and line faults is 18 / 90 = 0.2. The third type of associated fault sample set has 25 associated records, while the total number of bus fault records is 160. Therefore, the initial correlation degree between bus faults and automatic transfer switch failures is 25 / 160 ≈ 0.156.

[0102] The fourth step requires revising the initial correlation between bus faults and line faults, the initial correlation between backup automatic transfer switch failure and line faults, and the initial correlation between bus faults and backup automatic transfer switch failures based on the electrical distance in the distribution network topology and the average electrical distance in the distribution network topology. This will yield the coupling coefficients between bus faults and line faults, backup automatic transfer switch failures and line faults, and bus faults and backup automatic transfer switch failures.

[0103] First, calculate the average electrical distance between all devices in the distribution network topology. Then, for each initial correlation degree, adjust it by the ratio of the actual electrical distance to the average electrical distance of the device involved in the corresponding associated fault. The adjustment formula can be set as: Coupling coefficient = Initial correlation degree × (Average electrical distance / Actual electrical distance), finally obtaining three coupling coefficients.

[0104] For example, the average electrical distance of the distribution network topology is 1.0 km; the actual electrical distance between the equipment involved in the association between bus faults and line faults is 0.8 km, and its coupling coefficient = 0.2 × (1.0 / 0.8) = 0.25; the actual electrical distance between the equipment involved in the association between backup automatic transfer failure and line fault is 1.2 km, and its coupling coefficient = 0.2 × (1.0 / 1.2) ≈ 0.167; the actual electrical distance between the equipment involved in the association between bus faults and backup automatic transfer failure is 1.0 km, and its coupling coefficient = 0.156 × (1.0 / 1.0) = 0.156.

[0105] Furthermore, based on the three coupling coefficients, it is necessary to perform correlation correction on the initial start value of the bus fault, the initial start value of the backup automatic transfer and automatic disconnection failure, and the initial start value of the line fault to obtain the bus fault coupling correction start value, the backup automatic transfer and automatic disconnection failure coupling correction start value, and the line fault coupling correction start value.

[0106] Assume that three coupling coefficients have been obtained: 0.25 for bus fault and line fault, 0.167 for automatic transfer switch failure and line fault, and 0.156 for bus fault and automatic transfer switch failure; the initial start values ​​output by step S200 are: 0.8 for bus fault, 0.6 for automatic transfer switch failure, and 0.7 for line fault.

[0107] For example, the calculation method for the bus fault coupling correction start value can be as follows: the bus fault is less affected by the failure of automatic transfer switch, and only its own initial value and the reverse correction of the line fault are considered, the formula is 0.8-0.7×0.25=0.625; the calculation method for the failure of automatic transfer switch coupling correction start value can be as follows: considering the correlation with the bus fault, the formula is 0.6+0.8×0.156=0.7248; the calculation method for the line fault coupling correction start value can be as follows: simultaneously affected by the correlation of bus fault and failure of automatic transfer switch, the formula is 0.7+0.8×0.25+0.6×0.167≈0.7+0.2+0.1002=0.9002.

[0108] like Figure 2 As shown, in step S300 of this embodiment, real-time operating parameters of the distribution network are collected through the distribution network terminal, and a correction factor is calculated by combining the historical prediction deviation rate of the self-supervised learning model, including:

[0109] Real-time data collection of distributed power penetration rate, line load rate, and bus voltage deviation is achieved through distribution network terminals.

[0110] The distributed power source penetration rate, line real-time load rate, and bus real-time voltage deviation are normalized and then weighted and summed to obtain the real-time operating condition correction factor.

[0111] Obtain the deviation data between the historical prediction start values ​​of the three self-supervised learning models and the actual self-healing effect of the corresponding scenarios, calculate the historical prediction deviation rate of the three self-supervised learning models, and calculate the arithmetic mean to obtain the three average historical prediction deviation rates.

[0112] By combining the real-time operating condition correction factor with the three average historical prediction deviation rates, three correction factors are calculated.

[0113] In this embodiment, the purpose of step S300 is to obtain a correction factor that adapts to the real-time operating status of the distribution network and the model prediction accuracy, providing a basis for the secondary correction of the coupled correction start-up value. Real-time operating condition changes can affect fault characteristics and self-healing effects, and historical prediction deviations of the model may lead to errors in the start-up value. The correction factor can dynamically calibrate the start-up value by combining both factors, further improving its fit with the actual situation of the distribution network.

[0114] To achieve the above objectives, the first step is to collect data on distributed power penetration rate, line load rate, and bus voltage deviation in real time through the distribution network terminal.

[0115] The distribution network terminal captures three key real-time operating parameters: distributed generation penetration rate, line real-time load factor, and bus real-time voltage deviation. These parameters directly reflect the current operating status of the distribution network. Distributed generation penetration rate is the proportion of distributed generation output to the total distribution network load; line real-time load factor is the ratio of the current line load to the rated load; and bus real-time voltage deviation is the proportion of the difference between the actual bus voltage and the rated voltage to the rated voltage. For example, at a certain moment, the distribution network terminal collects data showing a distributed generation penetration rate of 20%, a line real-time load factor of 75%, and a bus real-time voltage deviation of 3%.

[0116] The second step is to normalize the distributed power penetration rate, line real-time load rate, and bus real-time voltage deviation, and then sum them by weight to obtain the real-time operating condition correction factor.

[0117] First, the three types of real-time operating parameters are normalized, mapping the parameter values ​​to the range of 0 to 1, thus eliminating the influence of differences in different parameter dimensions. Assuming the normalization formula is Normalized value = (Actual value - Minimum value) / (Maximum value - Minimum value), the distributed power supply penetration rate is set to a range of 0 to 50%, the line real-time load rate to a range of 0 to 100%, and the bus real-time voltage deviation to a range of 0 to 10%.

[0118] For example, in the above example, the normalized value of distributed power source penetration rate = (20-0) / (50-0) = 0.4; the normalized value of line real-time load rate = (75-0) / (100-0) = 0.75; and the normalized value of bus real-time voltage deviation = (3-0) / (10-0) = 0.3. Then, weights are assigned to the three normalized values. Assuming the weight of distributed power source penetration rate is 0.3, the weight of line real-time load rate is 0.4, and the weight of bus real-time voltage deviation is 0.3, the weighted sum is obtained as the real-time operating condition correction factor = 0.4×0.3 + 0.75×0.4 + 0.3×0.3 = 0.12 + 0.3 + 0.09 = 0.51.

[0119] The third step is to obtain the deviation data between the historical prediction start values ​​of the three self-supervised learning models and the actual self-healing effect of the corresponding scenarios, calculate the historical prediction deviation rate of the three self-supervised learning models, and calculate the arithmetic mean to obtain the three average historical prediction deviation rates.

[0120] This involves acquiring the historical operational data of the three self-supervised learning models, including the historical predicted starting values ​​for each model and the deviation data of the actual self-healing effect in the corresponding scenario. The actual self-healing effect data can be quantified as the actual required starting value. The deviation rate of each actual self-healing effect data point is calculated using the formula: Single Data Point Deviation Rate = |Predicted Starting Value - Actual Starting Value| / Actual Starting Value. Then, the arithmetic mean of all historical deviation rates for each model is calculated to obtain the average historical prediction deviation rate for each of the three models.

[0121] For example, the historical prediction deviation rates of the bus fault prediction model are 5%, 3%, and 4%, respectively, with an average historical prediction deviation rate of (5%+3%+4%) / 3=4%; the historical prediction deviation rates of the automatic transfer switch failure prediction model are 6%, 5%, and 7%, respectively, with an average historical prediction deviation rate of (6%+5%+7%) / 3=6%; and the historical prediction deviation rates of the line fault prediction model are 4%, 6%, and 5%, respectively, with an average historical prediction deviation rate of (4%+6%+5%) / 3=5%.

[0122] The third step is to combine the real-time operating condition correction factor with the three average historical prediction deviation rates to calculate the three correction factors.

[0123] The formula "real-time operating condition correction factor × (1 - average historical prediction deviation rate)" is used, combined with the real-time operating condition correction factor and the average historical prediction deviation rate of the corresponding model, to calculate three correction factors respectively. For example, the correction factor for the bus fault scenario = 0.51 × (1 - 4%) = 0.51 × 0.96 = 0.4896; the correction factor for the automatic transfer switch failure scenario = 0.51 × (1 - 6%) = 0.51 × 0.94 = 0.4794; and the correction factor for the line fault scenario = 0.51 × (1 - 5%) = 0.51 × 0.95 = 0.4845.

[0124] Furthermore, based on the correction factor, the bus fault coupling correction start value, the backup automatic transfer failure coupling correction start value, and the line fault coupling correction start value are corrected a second time to obtain the bus fault start value, the backup automatic transfer failure start value, and the line fault start value.

[0125] Assume that the three coupling correction start values ​​have been obtained: bus fault coupling correction start value 0.625, backup automatic transfer failure coupling correction start value 0.7248, and line fault coupling correction start value 0.9002; and the three correction factors are: bus fault scenario correction factor 0.4896, backup automatic transfer failure scenario correction factor 0.4794, and line fault scenario correction factor 0.4845.

[0126] The secondary correction can be implemented according to the logic of "coupled correction start value × correction factor", as follows: bus fault start value = 0.625 × 0.4896 = 0.306; automatic transfer switch failure start value = 0.7248 × 0.4794 ≈ 0.347; line fault start value = 0.9002 × 0.4845 ≈ 0.436.

[0127] In step S300 of this application embodiment, dynamic weights are calculated based on historical power outage parameters, switch operation cost parameters, three coupling coefficients, and three correction factors, including:

[0128] Based on the historical outage parameters, including the outage duration when the self-healing system was not activated and the power transfer duration after the self-healing system was activated, and combined with the bus fault scenario correction factor, the self-healing benefit coefficient for the bus fault scenario is calculated.

[0129] Based on the hardware loss cost and manual maintenance cost of a single switch operation in the switch operation cost parameters, and combined with the coupling coefficient between bus faults and line faults, the self-healing cost coefficient for bus fault scenarios is calculated.

[0130] The weight base of the bus fault scenario is calculated based on the self-healing benefit coefficient and the self-healing cost coefficient of the bus fault scenario.

[0131] Based on the historical outage parameters, including the duration of outages when the self-healing system was not activated and the duration of power transfer after the self-healing system was activated, and combined with the correction factor for the failure scenario of backup self-transfer and self-disconnection, the self-healing benefit coefficient for the failure scenario of backup self-transfer and self-disconnection is calculated.

[0132] Based on the hardware loss cost and manual maintenance cost of a single switch operation in the switch operation cost parameters, and combined with the coupling coefficient between backup automatic transfer failure and line fault, the self-healing cost coefficient for backup automatic transfer failure scenarios is calculated.

[0133] Based on the self-healing benefit coefficient and the self-healing cost coefficient of the self-healing failure scenario, the weight base of the self-healing failure scenario is calculated.

[0134] Based on the historical outage parameters, including the outage duration when the self-healing system was not activated and the power transfer duration after the self-healing system was activated, and combined with the line fault scenario correction factor, the self-healing benefit coefficient of the line fault scenario is calculated.

[0135] Based on the hardware loss cost and manual maintenance cost of a single switch operation in the switch operation cost parameters, and combined with the coupling coefficient of bus fault and backup automatic transfer failure, the self-healing cost coefficient of the line fault scenario is calculated.

[0136] Based on the self-healing benefit coefficient and the self-healing cost coefficient of the line fault scenario, the weight base of the line fault scenario is calculated.

[0137] The weight bases for bus fault scenarios, backup automatic transfer failure scenarios, and line fault scenarios are normalized to obtain three dynamic weights that sum to 1.

[0138] In step S300 of this application embodiment, the core purpose of the above-mentioned detailed steps is to calculate the weights of three types of scenarios that can dynamically adapt to fault benefits, costs, and related characteristics, so that the impact of various fault initiation values ​​matches the actual self-healing value, cost, and risk during subsequent weighted fusion. Dynamic weights comprehensively consider self-healing benefits, operating costs, fault-related interference, and deviations between operating conditions and models, avoiding the rigidity of fixed weights and improving the rationality of decisions based on multiple self-healing initiation values.

[0139] To achieve the above objectives, it is first necessary to determine various core basic parameters to ensure consistent calculation basis. For example, in historical power outage parameters, the power outage duration when the self-healing system was not activated was 2 hours, and the power transfer duration after the self-healing system was activated was 0.5 hours; in switch operation cost parameters, the hardware loss cost of a single switch operation is 500 yuan, and the manual maintenance cost is 300 yuan; the three coupling coefficients are 0.25 for bus fault and line fault, 0.167 for backup automatic transfer failure and line fault, and 0.156 for bus fault and backup automatic transfer failure; the three correction factors are 0.4896 for bus fault scenario correction, 0.4794 for backup automatic transfer failure scenario correction, and 0.4845 for line fault scenario correction.

[0140] Next, it is necessary to calculate the self-healing benefit coefficient for each scenario.

[0141] The self-healing benefit coefficient is used to quantify the benefits in terms of power outage time after the self-healing system is activated. It needs to be calculated by combining historical power outage parameters and corresponding scenario correction factors. The formula is set as follows: Self-healing benefit coefficient = (Power outage time when the self-healing system was not activated in the past - Power transfer time after the self-healing system was activated in the past) × Scenario correction factor.

[0142] The self-healing benefit coefficient for a bus fault scenario = (2-0.5)×0.4896 = 1.5×0.4896 = 0.7344;

[0143] The self-healing benefit coefficient for the scenario of self-investment and self-switching failure is (2-0.5)×0.4794=1.5×0.4794=0.7191;

[0144] The self-healing benefit coefficient for line fault scenarios = (2-0.5)×0.4845 = 1.5×0.4845 = 0.72675.

[0145] Next, it is necessary to calculate the self-healing cost coefficient for each scenario.

[0146] The self-healing cost coefficient is used to quantify the operational cost of initiating self-healing. It needs to be calculated by combining the switch operation cost parameter and the corresponding coupling coefficient. The formula is set as: self-healing cost coefficient = (hardware loss cost + manual operation and maintenance cost) × corresponding coupling coefficient ÷ 1000. Here, 1000 is used for unit conversion to facilitate the matching of coefficient magnitude.

[0147] For example, the self-healing cost coefficient for a bus fault scenario is (500+300)×0.25÷1000=800×0.25÷1000=0.2; the self-healing cost coefficient for a backup automatic transfer failure scenario is (500+300)×0.167÷1000=800×0.167÷1000=0.1336; and the self-healing cost coefficient for a line fault scenario is (500+300)×0.156÷1000=800×0.156÷1000=0.1248.

[0148] Then, it is necessary to calculate the weight base for each scenario.

[0149] The weight base reflects the overall priority of the scenario and is determined by the difference between the self-healing benefit coefficient and the self-healing cost coefficient. The higher the benefit and the lower the cost, the larger the base and the higher the weight. The formula is set as: Weight base = Self-healing benefit coefficient - Self-healing cost coefficient.

[0150] For example, the weight base for the bus fault scenario is 0.7344 - 0.2 = 0.5344; the weight base for the backup automatic transfer failure scenario is 0.7191 - 0.1336 = 0.5855; and the weight base for the line fault scenario is 0.72675 - 0.1248 = 0.60195.

[0151] Furthermore, the three weight bases need to be normalized so that the sum of the weights is 1. The formula is set as: Dynamic weight = Weight base of a certain scenario ÷ Sum of the weight bases of the three scenarios.

[0152] For example, if the sum of the base weights for the three scenarios is 0.5344 + 0.5855 + 0.60195 = 1.72185, then the dynamic weight for the bus fault scenario is 0.5344 ÷ 1.72185 ≈ 0.310; the dynamic weight for the backup automatic transfer failure scenario is 0.5855 ÷ 1.72185 ≈ 0.340; and the dynamic weight for the line fault scenario is 0.60195 ÷ 1.72185 ≈ 0.350.

[0153] Next, the bus fault start value, the backup automatic transfer failure start value, and the line fault start value need to be weighted and summed to obtain the multi-source self-healing start value.

[0154] In step S300 of this application embodiment, the bus fault initiation value, the standby automatic transfer failure initiation value, and the line fault initiation value are weighted and summed to obtain a multi-source self-healing initiation value, including:

[0155] Obtain three dynamic weights for bus fault, automatic transfer switch failure, and line fault scenarios;

[0156] The bus fault initiation value, the automatic transfer switch failure initiation value, and the line fault initiation value are weighted and summed according to three dynamic weights to obtain the multi-source self-healing initiation value.

[0157] Specifically, it is necessary to obtain the three calculated dynamic weights and the three secondary corrected fault initiation values. For example, the dynamic weight for the bus fault scenario is 0.310, the dynamic weight for the backup automatic transfer failure scenario is 0.340, and the dynamic weight for the line fault scenario is 0.350; the bus fault initiation value is 0.306, the backup automatic transfer failure initiation value is 0.347, and the line fault initiation value is 0.436.

[0158] Then, the multi-source recovery start value is calculated by multiplying each fault start value by its corresponding dynamic weight and then summing the results. For example, in the example above, the multi-source recovery start value = 0.306 × 0.310 + 0.347 × 0.340 + 0.436 × 0.350 ≈ 0.0949 + 0.1180 + 0.1526 ≈ 0.3655.

[0159] In step S400 of this application embodiment, a specific start-up threshold adapted to different operating conditions of the distribution network is preset, and the multi-self-healing start-up value is compared with the specific start-up threshold of the corresponding operating condition to determine whether to start the distribution network self-healing system, including:

[0160] Based on different distribution network operating parameters, distribution network operating conditions are pre-defined, wherein the distribution network operating conditions include at least light load conditions, heavy load conditions, high distributed power source access conditions, and voltage deviation critical conditions.

[0161] For each type of distribution network operation condition, a specific start threshold is preset based on the historical self-healing effect data of the distribution network and the engineering safety operation standards;

[0162] Based on the real-time operating parameters of the distribution network, match the operating conditions of the distribution network and obtain the corresponding exclusive start threshold;

[0163] The calculated multi-self self-healing start value is compared with the corresponding exclusive start threshold. If the multi-self self-healing start value is greater than or equal to the exclusive start threshold, it is determined that the self-healing start condition is met. If the multi-self self-healing start value is less than the exclusive start threshold, it is determined that the start condition is not met.

[0164] In this embodiment, the core purpose of step S400 is to match a corresponding dedicated activation threshold based on the real-time operating conditions of the distribution network, and to accurately determine whether to activate the distribution network self-healing system by comparing multiple self-healing activation values ​​with this threshold. Distribution network operating characteristics and self-healing requirements differ under different operating conditions. Presetting a dedicated threshold avoids the problem of insufficient adaptability of a single threshold, ensuring that the self-healing system is activated promptly when needed, while preventing unnecessary activation that could lead to cost waste or security risks, thus ensuring the scientific and targeted nature of the decision-making.

[0165] To achieve the above objectives, it is first necessary to pre-define the distribution network operating conditions based on different distribution network operating parameters.

[0166] This means using core operating parameters of the distribution network as the basis for classification, and clearly defining the standards for various operating conditions. These parameters include line load factor, distributed generation penetration rate, and bus voltage deviation, etc., classifying distribution network operating conditions into at least four categories: light load, heavy load, high distributed generation access, and critical voltage deviation. For example, a line load factor ≤ 40% is defined as a light load condition, a line load factor ≥ 80% as a heavy load condition, a distributed generation penetration rate ≥ 30% as a high distributed generation access condition, and a bus voltage deviation ≥ 8% and < 10% as a critical voltage deviation condition.

[0167] Next, for each type of distribution network operation condition, it is necessary to preset a specific start threshold based on the historical self-healing effect data of the distribution network and the engineering safety operation standards.

[0168] For each categorized operating condition, historical self-healing performance data of the distribution network under that condition is collected, including self-healing success rate, reduction in outage time, and incidence of fault escalation risk. Simultaneously, combined with engineering safety operation standards, a specific activation threshold is determined for each type of operating condition. Threshold settings must balance self-healing benefits with safety risks, and thresholds differ across different operating conditions. For example: under light load conditions, the distribution network operates stably with a small fault impact range, so a preset activation threshold of 0.3 is used; under heavy load conditions, faults are prone to escalation, requiring strict control of activation conditions, so a preset activation threshold of 0.5 is used; under conditions with high distributed power source access, there is a risk of power fluctuation, so a preset activation threshold of 0.45 is used; under conditions with critical voltage deviation, prioritizing voltage stability is required, so a preset activation threshold of 0.4 is used.

[0169] Then, based on the real-time operating parameters of the distribution network, the operating conditions of the distribution network are matched, and the corresponding exclusive start-up thresholds are obtained.

[0170] For each categorized operating condition, historical self-healing performance data of the distribution network under that condition is collected, including self-healing success rate, reduction in outage time, and incidence of fault escalation risk. Simultaneously, combined with engineering safety operation standards, a dedicated activation threshold is comprehensively determined for each type of operating condition. The setting of dedicated activation thresholds must balance self-healing benefits with safety risks, and the dedicated activation thresholds differ for different operating conditions. For example, under light load conditions, the distribution network operates stably with a small fault impact range, and the preset dedicated activation threshold is 0.3; under heavy load conditions, faults are prone to escalation, requiring strict control of activation conditions, and the preset dedicated activation threshold is 0.5; under conditions with high distributed power source access, there is a risk of power fluctuation, and the preset dedicated activation threshold is 0.45; under conditions with critical voltage deviation, prioritizing voltage stability is crucial, and the preset dedicated activation threshold is 0.4.

[0171] Furthermore, it is necessary to match the operating conditions of the distribution network with the real-time operating parameters of the distribution network and obtain the corresponding exclusive start-up threshold.

[0172] This involves collecting real-time operating parameters from the distribution network terminal, matching the current operating condition of the distribution network according to preset operating condition definition standards, and then retrieving the corresponding exclusive start-up threshold. For example, if the real-time operating parameters of the distribution network are a line load rate of 35%, a distributed power supply penetration rate of 20%, and a bus voltage deviation of 2%, it is matched as a light load condition according to the definition standards, and the corresponding exclusive start-up threshold is 0.3.

[0173] Finally, the calculated multi-self self-healing start value is compared with the corresponding exclusive start threshold. If the multi-self self-healing start value is greater than or equal to the exclusive start threshold, it is determined that the self-healing start condition is met. If the multi-self self-healing start value is less than the exclusive start threshold, it is determined that the start condition is not met.

[0174] The self-healing start value calculated in step S300 is compared with the matched dedicated start threshold. A judgment rule is set: if the multi-source self-healing start value is greater than or equal to the dedicated start threshold, the self-healing start condition is met; if the multi-source self-healing start value is less than the dedicated start threshold, the start condition is not met. For example, if the current multi-source self-healing start value is 0.3655, and the matched light-load condition dedicated start threshold is 0.3, since 0.3655 ≥ 0.3, the self-healing start condition is met, and the distribution network self-healing system is started; if the multi-source self-healing start value is 0.28, which is less than 0.3, the start condition is not met, and the self-healing system is not started.

[0175] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0176] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0177] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0178] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0179] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0180] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the invention and its equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A method for starting self-healing of external faults of multi-source signal distribution network based on self-supervised learning, characterized in that, The method comprises: Real-time acquisition of multi-source fault signals with time stamps through a network terminal, wherein the multi-source fault signals include bus fault signals, backup automatic switching failure signals, line fault signals and long-time voltage loss signals; Based on three independently trained self-supervised learning models, matching calculation is performed on the multi-source fault signals respectively, and bus fault initial starting values, backup automatic switching failure initial starting values and line fault initial starting values are correspondingly output; After correction and optimization of the bus fault initial starting values, backup automatic switching failure initial starting values and line fault initial starting values, weighted fusion is performed through dynamic weights to obtain multi-source self-healing starting values; A preset exclusive starting threshold value suitable for different network conditions is compared with the multi-source self-healing starting values to determine whether to start the network self-healing system.

2. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 1, characterized in that, Real-time acquisition of multi-source fault signals with time stamps through a network terminal, comprising: Acquisition of bus fault signals output by bus protection devices of a superior switch station or a transformer substation when a bus fault occurs; Acquisition of backup automatic switching failure signals output by a network backup automatic switching device when a transformer fault or a line voltage loss outside a station causes a self-cutting action failure; Acquisition of line fault signals corresponding to line faults through network terminal monitoring of line operation states; Continuous monitoring of line voltages through a network terminal, and acquisition of long-time voltage loss signals when the line voltage is detected to be less than or equal to a preset voltage threshold value and the duration is greater than or equal to a preset duration threshold value; Time stamp synchronization of the acquired bus fault signals, backup automatic switching failure signals, line fault signals and long-time voltage loss signals to ensure the time sequence correlation of each signal.

3. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 1, characterized in that, Based on three independently trained self-supervised learning models, matching calculation is performed on the multi-source fault signals respectively, and bus fault initial starting values, backup automatic switching failure initial starting values and line fault initial starting values are correspondingly output, comprising: Input of the time-stamped bus fault signals and long-time voltage loss signals into a pre-trained bus fault prediction model to output bus fault initial starting values; Input of the time-stamped backup automatic switching failure signals and long-time voltage loss signals into a pre-trained backup automatic switching failure prediction model to output backup automatic switching failure initial starting values; Input of the time-stamped line fault signals and long-time voltage loss signals into a pre-trained line fault prediction model to output line fault initial starting values.

4. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 3, characterized in that, The construction process of the bus fault prediction model comprises: Collection of different types of bus fault signals, associated historical long-time voltage loss signals, and historical power outage parameters and switch operation cost parameters corresponding to each signal in network history to form a training sample set, wherein the historical power outage parameters include bus fault outage duration when the self-healing system is not started and transfer power duration after the self-healing system is started, and the switch operation cost parameters include hardware loss cost and manual operation and maintenance cost of a single switch operation; Difference calculation of the historical bus fault outage duration when the self-healing system is not started and the transfer power duration after the self-healing system is started in the historical power outage parameters, and then normalization processing is performed to obtain a comprehensive power outage benefit parameter; Summing and calculating hardware loss cost and manual operation cost in the switch operation cost parameter, and then performing normalization processing to obtain a comprehensive operation cost parameter; Defining a difference between the comprehensive outage benefit parameter and the comprehensive operation cost parameter as a self-healing net benefit parameter, and constructing a self-supervised training signal with a maximization of the self-healing net benefit parameter as an optimization target; A lightweight neural network structure is adopted to build a basic framework of the busbar fault prediction model; The training sample set and the self-supervised training signal are input into the basic framework for iterative training until verification convergence is achieved, and a trained busbar fault prediction model is obtained.

5. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 1, characterized in that, After the busbar fault initial starting value, the backup automatic switching self-cutting failure initial starting value and the line fault initial starting value are corrected and optimized, a multi-source self-healing starting value is obtained through dynamic weight weighted fusion, including: Based on the distribution network topology relationship network and the distribution network historical fault data, three coupling coefficients are calculated, wherein the three coupling coefficients are the coupling coefficient of busbar fault and line fault, the coupling coefficient of backup automatic switching self-cutting failure and line fault, and the coupling coefficient of busbar fault and backup automatic switching self-cutting failure; Based on the three coupling coefficients, the busbar fault initial starting value, the backup automatic switching self-cutting failure initial starting value and the line fault initial starting value are associated and corrected to obtain a busbar fault coupling correction starting value, a backup automatic switching self-cutting failure coupling correction starting value and a line fault coupling correction starting value; Real-time working condition parameters of the distribution network are collected through a distribution network terminal, and a correction factor is calculated based on the historical prediction deviation rate of the self-supervised learning model, wherein the real-time working condition parameters include a distributed power penetration rate, a line real-time load rate and a busbar real-time voltage deviation; Based on the correction factor, the busbar fault coupling correction starting value, the backup automatic switching self-cutting failure coupling correction starting value and the line fault coupling correction starting value are secondarily corrected to obtain a busbar fault starting value, a backup automatic switching self-cutting failure starting value and a line fault starting value; Based on the historical outage parameters, the switch operation cost parameters, the three coupling coefficients and the three correction factors, a dynamic weight is calculated, and the busbar fault starting value, the backup automatic switching self-cutting failure starting value and the line fault starting value are weighted and summed to obtain a multi-source self-healing starting value.

6. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 5, characterized in that, Based on the distribution network topology relationship network and the distribution network historical fault data, three coupling coefficients are calculated, including: The connection mode and electrical distance of the busbar, line and switch device are obtained to construct a distribution network topology relationship network; The distribution network historical fault data are collected, and complete fault records of busbar fault triggering line fault, backup automatic switching self-cutting failure triggering line fault and busbar fault triggering backup automatic switching self-cutting failure are selected respectively to obtain three associated fault sample sets; For the three associated fault sample sets, the occurrence probability of the subsequent fault within a preset time window after the occurrence of the previous fault is calculated respectively to obtain an initial association degree of busbar fault and line fault, an initial association degree of backup automatic switching self-cutting failure and line fault, and an initial association degree of busbar fault and backup automatic switching self-cutting failure. Based on the electrical distance in the network topology relationship network, the initial correlation degree of bus fault and line fault, the initial correlation degree of backup power automatic switching failure and line fault, and the initial correlation degree of bus fault and backup power automatic switching failure are corrected in combination with the average electrical distance of the network topology relationship network, to obtain the coupling coefficients of bus fault and line fault, backup power automatic switching failure and line fault, and bus fault and backup power automatic switching failure.

7. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 5, characterized in that, The real-time working condition parameters of the distribution network are collected by the distribution network terminal, and the historical prediction deviation rate of the self-supervised learning model is combined to calculate the correction factor, including: The distributed power penetration rate, line real-time load rate, and bus real-time voltage deviation are collected in real time by the distribution network terminal; The distributed power penetration rate, line real-time load rate, and bus real-time voltage deviation are normalized and weighted summed to obtain the real-time working condition correction factor; The historical prediction start value of the three self-supervised learning models and the deviation data of the actual self-healing effect of the corresponding scene are obtained, the historical prediction deviation rate of the three self-supervised learning models is calculated, and the arithmetic mean is calculated to obtain the three average historical prediction deviation rates; The three correction factors are calculated in combination with the real-time working condition correction factor and the three average historical prediction deviation rates.

8. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 5, characterized in that, Based on the historical outage parameters, switch operation cost parameters, three coupling coefficients, and three correction factors, the dynamic weight is calculated, including: According to the fault outage time when the self-healing system is not started and the transferred power time after the self-healing system is started in the historical outage parameters, in combination with the bus fault scene correction factor, the self-healing benefit coefficient of the bus fault scene is calculated; According to the switch single operation hardware loss cost and artificial operation and maintenance cost in the switch operation cost parameters, in combination with the coupling coefficient of bus fault and line fault, the self-healing cost coefficient of the bus fault scene is calculated; Based on the self-healing benefit coefficient of the bus fault scene and the self-healing cost coefficient of the bus fault scene, the bus fault scene weight base is calculated; According to the fault outage time when the self-healing system is not started and the transferred power time after the self-healing system is started in the historical outage parameters, in combination with the backup power automatic switching failure scene correction factor, the self-healing benefit coefficient of the backup power automatic switching failure scene is calculated; According to the switch single operation hardware loss cost and artificial operation and maintenance cost in the switch operation cost parameters, in combination with the coupling coefficient of backup power automatic switching failure and line fault, the self-healing cost coefficient of the backup power automatic switching failure scene is calculated; Based on the self-healing benefit coefficient of the backup power automatic switching failure scene and the self-healing cost coefficient of the backup power automatic switching failure scene, the backup power automatic switching failure scene weight base is calculated; According to the fault outage time when the self-healing system is not started and the transferred power time after the self-healing system is started in the historical outage parameters, in combination with the line fault scene correction factor, the self-healing benefit coefficient of the line fault scene is calculated; According to the switch single operation hardware loss cost and artificial operation and maintenance cost in the switch operation cost parameters, in combination with the coupling coefficient of bus fault and backup power automatic switching failure, the self-healing cost coefficient of the line fault scene is calculated; A line fault scenario weight base number is calculated based on a self-recovery benefit coefficient of the line fault scenario and a self-recovery cost coefficient of the line fault scenario; The busbar fault scenario weight base number, the backup automatic switching failure scenario weight base number and the line fault scenario weight base number are normalized to obtain three dynamic weights whose sum is 1.

9. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 5, characterized in that, The busbar fault starting value, the backup automatic switching failure starting value and the line fault starting value are weighted and summed to obtain a multi-source self-recovery starting value, including: Three dynamic weights of busbar fault, backup automatic switching failure and line fault scenarios are obtained; The busbar fault starting value, the backup automatic switching failure starting value and the line fault starting value are weighted and summed according to the three dynamic weights to obtain a multi-source self-recovery starting value.

10. The self-supervised learning based multi-source signal distribution network external fault self-healing starting method according to claim 1, characterized in that, A preset exclusive starting threshold value is adapted to different working conditions of the distribution network, and the multi-source self-recovery starting value is compared with the exclusive starting threshold value corresponding to the working condition to determine whether to start the distribution network self-recovery system, including: Different distribution network working condition parameters are used to pre-classify the distribution network working conditions, wherein the distribution network working conditions at least include a light load working condition, a heavy load working condition, a high distributed power access working condition and a voltage deviation critical working condition; For each type of distribution network working condition, an exclusive starting threshold value is preset in combination with historical self-recovery effect data of the distribution network and engineering safety operation standards; According to real-time working condition parameters of the distribution network, the distribution network working condition is matched, and the corresponding exclusive starting threshold value is obtained; The calculated multi-source self-recovery starting value is compared with the corresponding exclusive starting threshold value, if the multi-source self-recovery starting value is greater than or equal to the exclusive starting threshold value, it is determined that the self-recovery starting condition is met, and if the multi-source self-recovery starting value is less than the exclusive starting threshold value, it is determined that the starting condition is not met.

Images