Power grid fault location inference method and system

Abstract

The application discloses a power grid fault position inference method and system, comprising factorization of conditional probability distribution functions of network topology Y of given multi-source feedback data to obtain conditionally independent fault factors; parameterization of the fault factors according to available historical statistical interruption information, establishment of a Bayesian network for each distribution feeder; and inference of fault positions by performing an inference task on the Bayesian network by using a sampling algorithm. The application adopts a probabilistic graphical modeling method for data fusion, and a high-dimensional joint probability distribution function of the system is decomposed into a group of more easily managed probability factors, which are obtained by conditional independence, so as to reduce the calculation complexity of the high-dimensional joint probability distribution function representing the system and improve the robustness and accuracy of a fault research and judgment framework.

Description

Technical Field

[0001] This invention relates to the field of power grid big data mining technology, specifically to a method and system for inferring the location of power grid faults. Background Technology

[0002] Electrical networks include power generation, transmission, and distribution networks. The distribution system is the last step in transmitting electricity to individual consumers. Due to the rapid growth in human needs over the past few decades, the demand for electricity and the number of electricity consumers have been increasing. Therefore, the distribution network has had to expand its distribution lines and feeders to meet consumer demand. Moreover, as society becomes more and more dependent on energy, the requirements for power reliability are also increasing.

[0003] A fundamental task for power system operators is fault diagnosis and location. Faults can lead to network equipment damage, service interruptions, and network instability, thereby reducing network reliability and causing economic losses to customers and power companies. Traditional methods for fault diagnosis and location in distribution network feeders are inefficient, especially when the network has a wide geographical distribution. The required manpower and equipment are high, and the time consumption is significant. Therefore, automatic and rapid fault prediction and location are crucial in distribution networks. An automated fault prediction and location framework offers advantages such as saving time and manpower, enhancing system maintenance preparedness, revising future plans, and improving economic factors. These factors increase customer satisfaction and improve system reliability indicators.

[0004] The primary cause of power system energy failures is power outages, especially those in the distribution system (line-to-ground faults account for approximately 70%). Power outages are caused by a variety of factors, such as facility failures, lighting, storms, severe weather, rain, isolation faults, trees, and birds. The considerable uncertainty of these data sources can lead to incorrect fault locations and additional costs for utilities; for example, due to hardware and software issues, only a portion of the fault signals can be transmitted to the utility's data center. Given the limitations and uncertainties of single data sources, a key issue is how to quickly and accurately locate distribution system faults from these multi-source factors and multi-source fault feedback data. A reliable, low-latency, and robust power grid fault assessment framework for scenarios with massive amounts of multi-source data can improve system reliability and power supply continuity, accelerate power restoration, and thus shorten service downtime.

[0005] The research on the fault assessment framework of traditional power grids in the context of multi-source massive data scenarios faces two key issues: (1) A fundamental challenge in power grid fault assessment under multi-source massive data scenarios is the computational complexity of the problem. The computational complexity of the power grid fault assessment framework greatly affects the fault response delay of the power grid. Because fault location inference is a process of calculating the probability of topology candidates after an interruption event using available information received by the utility sector, estimating these probability values ​​requires obtaining the joint probability distribution function of unknown state variables and evidence, which is a high-dimensional mathematical computational object. For actual distribution systems, directly quantifying this joint distribution is computationally infeasible, which requires listing the probabilities of all possible combinations of variables. (2) The accuracy of fault assessment needs to be improved because the fault information feedback data sources have heterogeneous characteristics, such as accuracy and reporting rate; moreover, multi-source data may provide inconsistent or even contradictory information. How to integrate these data sources and improve the accuracy of fault assessment under multi-source data feedback is a challenge.

[0006] In related technologies, Chinese invention patent application CN113791307A discloses a method for locating fault sections in a hybrid line distribution network based on discrete Bayesian networks. This method constructs a Bayesian probabilistic network using fault information from line measurement devices, and then trains the parameters of the discrete Bayesian network using the expectation-maximization algorithm based on historical fault information. Finally, it uses a confidence propagation algorithm to infer the discrete Bayesian network and obtain the fault status of each line segment under the current observation information. However, this scheme uses data from a single line measurement device to construct a single-distribution Bayesian probabilistic network, which leads to increased errors due to relying solely on a single fault data source. Furthermore, the confidence propagation algorithm used to infer the discrete Bayesian network is time-consuming and may encounter high latency risks in large-scale power grid applications with abundant data.

[0007] Chinese invention patent application CN113725862A discloses a method for identifying the topology of a distribution network based on Bayesian networks. This method constructs a Bayesian probabilistic network using voltage fluctuation and voltage power information, and finally uses a confidence propagation algorithm to infer the discrete Bayesian network, obtaining the fault status of each line segment under the current observation information. This scheme uses single voltage fluctuation and voltage power information to construct a single-distribution Bayesian probabilistic network, but this reliance on a single fault data source leads to increased errors. Furthermore, using a confidence propagation algorithm to infer the discrete Bayesian network, while seemingly simple, can be time-consuming and potentially encounter high latency risks in large-scale power grid applications with abundant data. Summary of the Invention

[0008] The technical problem to be solved by this invention is how to provide a power grid fault location inference method with low computational complexity and low latency.

[0009] The present invention solves the above-mentioned technical problems through the following technical means:

[0010] In a first aspect, the present invention proposes a method for inferring the location of a power grid fault, the method comprising the following steps:

[0011] For a given network topology with multi-source feedback data Y Factorize the conditional probability distribution function to obtain conditionally independent fault factors;

[0012] The fault factors are parameterized based on available historical statistical interruption information, and a Bayesian network is established for each distributed feeder.

[0013] A sampling algorithm is used to perform inference tasks on the Bayesian network to infer the location of the fault.

[0014] Furthermore, the multi-source data feedback set includes network flow fault data at the user load end, fault signals from electricity meters, power grid system fault feedback information, power grid system physical parameters, and environmental parameters.

[0015] Furthermore, the network topology for given multi-source feedback data Y Factoring the conditional probability distribution function yields conditionally independent fault factors, including:

[0016] Network topology based on given multi-source feedback data E Y Calculation of conditional probability distribution function Y The joint distribution term of E ,in, Y is a polynomial variable, represented by the connection states D of the network branches and the connections of the customer switch C;

[0017] Based on random variables The conditional independence between them decomposes the joint distribution term into a set of smaller fault factors, where each fault factor is a conditional probability distribution function composed of sub-variables and parent variables, and the sub-variables include the first fault factor in the feeder. i The connection status of the branch, the first i The first branch j The connection status of each customer switch, network flow fault data at the user load end, and fault signals of smart meters.

[0018] Furthermore, the decomposed form of the joint distribution term is as follows:

[0019]

[0020] In the formula: , Indicates the first in the feeder i The connection status of each branch Indicates the first i The first branch j Connection status of each customer switch This indicates network flow fault data at the user load end. This indicates a fault signal in the electricity meter.

[0021] Furthermore, the parameterization of the fault factor based on available historical statistical interruption information, and the establishment of a Bayesian network for each distributed feeder, includes:

[0022] The random variable As vertices of the Bayesian network, directed edges are drawn from the vertices, starting from the parent variable and ending at the child variable, to construct the structure of the Bayesian network.

[0023] Based on available historical statistical interruption information, the conditional probability distribution function of each fault factor in the structure of the Bayesian network is parameterized, and a Bayesian network is established for each distribution feeder.

[0024] Furthermore, the parameterization of the conditional probability distribution function of each fault factor in the structure of the Bayesian network based on available historical statistical interruption information, and the establishment of a Bayesian network for each distribution feeder, includes:

[0025] For fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the branch state variable is... ,in, It is the connection state of adjacent upstream branches. It is the first i Fault feedback from each branch, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch;

[0026] For fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the user state variable is... ;

[0027] For fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution.

[0028] For fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal The conditional probability distribution function of the failure factor is parameterized, becoming conditionally independent of the other variables.

[0029] Furthermore, the aforementioned fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, including:

[0030] When the parent branch loses power, based on variables Given the binary nature of the condition, the conditional probability distribution function of the fault factor can be parameterized as follows:

[0031] ;

[0032] When the adjacent upstream branch is energized, the conditional probability distribution function of this fault factor can be expressed as a Bernoulli distribution:

[0033]

[0034] In the formula: Indicates the first i The probability of failure for each branch.

[0035] Furthermore, the first i Failure probability of each branch A vulnerability model was developed for this purpose. , and The function is expressed by the formula:

[0036]

[0037] In the formula: L It is used to support the first i The number of distribution poles on the branch line, K It is the first i The number of conductors between two adjacent poles in a branch circuit, Φ is the standard normal probability integral, χ is the median of the vulnerability function, and ξ is the logarithmic standard deviation of the strength measurement. Indicates the first i The probability of failure of the conductor in the branch circuit.

[0038] Furthermore, the aforementioned fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, including:

[0039] When the main branch loses power, the conditional probability distribution function of this fault factor is parameterized as follows:

[0040] ;

[0041] When the main branch is energized, the conditional probability distribution function of this fault factor is parameterized as follows:

[0042]

[0043] In the formula: The value is randomly selected.

[0044] Furthermore, the aforementioned fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution, and its formula is expressed as:

[0045]

[0046] In the formula: Define values ​​for users. It is a variable value.

[0047] Furthermore, the aforementioned fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal The conditional probability distribution function of the failure factor, conditionally independent of other variables, is parameterized as follows:

[0048]

[0049] In the formula: and These represent the reliability of network flow communication on the load side and the probability of meter failure, respectively.

[0050] Furthermore, the step of using a sampling algorithm to perform an inference task on the Bayesian network to infer the fault location includes:

[0051] During the period following the occurrence of the fault Within this framework, all power outage evidence from client devices, along with branch-level evidence, is collected to construct an evidence sample set. The power outage evidence includes branch-level evidence. i The j The branch-level evidence includes fault data of the load-side network flow of individual customers and fault signals of electricity meters, including power grid system fault feedback information, power grid system physical parameters and environmental parameters;

[0052] Randomly assign any sample from the evidence sample set to all unknown state variables. Choose any state variable as the sampling starting point, where D represents the state of the network branch and C represents the customer switch.

[0053] In the (τ+1)th iteration of the Gibbs sampling, according to the structure of the Bayesian network, the data assigned to... Evidence samples of the parent and child variables are inserted into the local Bayesian estimator to approximate the latest samples. The conditional probability distribution function, where, Indicates the first in the feeder i The connection status of each branch;

[0054] A new sample is drawn using the inverse transform method, and the local sampling process for the next non-evidence variable is executed until all unknown variables in the Bayesian network have been sampled, thus completing one iteration of Gibbs sampling.

[0055] Based on the sample sequence generated by Gibbs sampling, the connection status of all branches and customers is inferred, and the power outage branch closest to the substation is selected as the location of the power outage event.

[0056] Furthermore, in the (τ+1)th iteration of the Gibbs sampling, according to the structure of the Bayesian network, the data allocated to... Evidence samples of the parent and child variables are inserted into the local Bayesian estimator to approximate the latest samples. The conditional probability distribution function, where, Indicates the first in the feeder i The connection status of each branch includes:

[0057]

[0058]

[0059]

[0060] In the formula: Except All other recent samples, including the values ​​of the evidence variables, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch, Indicates the first i-1 The state variables of the network branches in the τth iteration Indicates the first i+1 The state variables of the network branches in the τth iteration Indicates the first i+1 In the network branch, the first j The state variables of a user network-side switch in the τth iteration.

[0061] Furthermore, the connection status of all branches and customers is inferred from the sample sequence generated based on Gibbs sampling, and the branch closest to the substation that experienced a power outage is selected as the location of the power outage event, including:

[0062] The connection status of all branches and customers is inferred from the sample sequences generated by Gibbs sampling, as expressed by the formula:

[0063]

[0064] In the formula: M Indicates the number of iterations. Indicates the first i The sample sequences of each branch, where E represents multi-source feedback data;

[0065] Will By comparing with the set threshold, the first... i The connection status of each branch and customer is determined, and the branch closest to the substation that experienced a power outage is selected as the location of the power outage event.

[0066] Furthermore, after performing an inference task on the Bayesian network using a sampling algorithm to infer the fault location, the method further includes:

[0067] For each iteration of the sampling algorithm, the inter-sequence variation and intra-sequence variation of the sample sequence are calculated.

[0068] Based on the inter-sequence variation and the intra-sequence variation, a scaling factor is determined;

[0069] Based on the aforementioned scaling factor, the convergence of the diagnostic sampling algorithm under different iteration numbers is determined, and the maximum number of iterations is determined.

[0070] Further, for each iteration of the sampling algorithm, calculating the inter-sequence variation and intra-sequence variation of the sample sequence includes:

[0071] For each iteration, the sampling algorithm generates data for each unknown variable in the Bayesian network. n Starting with a sample sequence, each sample sequence is divided into two equal halves to supplement the original sample sequence. All sample sequences are then concatenated into a single sequence of size [number missing]. matrix ;

[0072] Based on the matrix Calculate the inter-sequence variation of the sample sequences. and intra-sequence variation The formula is expressed as:

[0073]

[0074]

[0075] In the formula: Represents the average value within the sequence. This represents the population mean. Indicates the first j Variance of individual sample sequences.

[0076] Furthermore, the scaling factor is determined based on the inter-sequence variation and the intra-sequence variation, expressed by the following formula:

[0077]

[0078] In the formula: Indicates variation between sequences. Indicates intra-sequence variation. n Indicates the number of samples.

[0079] Secondly, this invention proposes a power grid fault location inference system, the system comprising:

[0080] The factorization module is used to factor the network topology of given multi-source feedback data. Y Factorize the conditional probability distribution function to obtain conditionally independent fault factors;

[0081] The parameterization module is used to parameterize the fault factors based on available historical statistical interruption information and to establish a Bayesian network for each distributed feeder.

[0082] The inference module is used to perform inference tasks on the Bayesian network using a sampling algorithm to infer the location of the fault.

[0083] The advantages of this invention are:

[0084] (1) This invention identifies and locates transverse fault events in partially observable power distribution systems based on multi-source data fusion. A probabilistic graphical modeling method is used for data fusion. The high-dimensional joint probability distribution function of the system is decomposed into a set of more manageable probability factors. These factors are obtained by conditional independence to reduce the computational complexity of representing the high-dimensional joint probability distribution function of the system and improve the robustness and accuracy of the fault judgment framework. By establishing a Bayesian Network (BN) for each distribution feeder, the BN uses a graph-based representation method as the basis for analyzing the statistical relationship between random variables. The system topology from the single-line graph and the data flow information from the user-end network equipment are used. The graphical parameters are learned from historical power outage data based on experience. The sampling algorithm is used to perform inference tasks on the Bayesian Network. The power outage location process based on data fusion is effectively transformed into online inference on the BN, which can quickly infer the location of power grid faults.

[0085] (2) The method proposed in this invention can seamlessly integrate heterogeneous data sources. Different data sources can complement each other, increasing the amount of power outage information, thereby solving the problem of low coverage of smart fault feedback devices or low network flow reporting rate of user-end devices in actual power grids. At the same time, by utilizing the inherent conditional independence between evidence and state variables in the distribution system, the exponential computational complexity of the power outage location task is reduced to the linear complexity of the number of variables.

[0086] (3) This invention incorporates prior expert knowledge of public utilities into the fault reasoning model. The fault location process based on data fusion is effectively transformed into online reasoning for BN; the reasoning task is solved by using the Gibbs Sampling (GS) algorithm, which, as an algorithm based on Markov Chain Monte Carlo (MCMC), can provide complete features of the distribution of unknown variables by generating a series of samples.

[0087] (4) Since the uncertainty of each data source is explicitly modeled with probabilistic graphical parameters, the proposed method is robust to false alarms and inconsistencies in power outage evidence and has a high accuracy in fault diagnosis.

[0088] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0089] Figure 1 This is a flowchart illustrating the power grid fault location inference method proposed in this invention;

[0090] Figure 2 This is a framework diagram for inferring the location of power grid faults proposed in this invention;

[0091] Figure 3 This is a schematic diagram of the BN structure construction and parameterization in this invention;

[0092] Figure 4 This is a block diagram illustrating the principle of constructing a typical radiation distribution system BN structure based on various fault factors according to the present invention.

[0093] Figure 5 This is a schematic diagram of the power grid fault location inference system proposed in this invention. Detailed Implementation

[0094] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0095] like Figure 1 As shown, this invention proposes a method for inferring the location of a power grid fault, the method comprising the following steps:

[0096] S10. Network topology for given multi-source feedback data Y Factorize the conditional probability distribution function to obtain conditionally independent fault factors;

[0097] S20. Parameterize the fault factors based on available historical statistical interruption information, and establish a Bayesian network for each distributed feeder.

[0098] S30. Use a sampling algorithm to perform inference tasks on the Bayesian network and infer the location of the fault.

[0099] This embodiment uses multi-source data fusion to identify and locate transverse fault events in partially observable distribution systems. A probabilistic graphical modeling method is employed for data fusion, decomposing the system's high-dimensional joint probability distribution function into a set of more manageable probability factors. These factors are obtained through conditional independence to reduce the computational complexity representing the system's high-dimensional joint probability distribution function, thereby improving the robustness and accuracy of the fault assessment framework. A Bayesian network (BN) is established for each distribution feeder. The BN utilizes a graph-based representation as the basis for analyzing the statistical relationships between random variables, drawing on the system topology from a single-line graph and data flow information from user-end network devices. The graphical parameters are learned empirically from historical power outage data. A sampling algorithm is used to perform inference tasks on the Bayesian network. The power outage location process based on data fusion is effectively transformed into online inference on the BN, enabling rapid inference of power grid fault locations.

[0100] In one embodiment, the multi-source data feedback set includes network flow fault data from the user load end, fault signals from the electricity meter, power grid system fault feedback information, power grid system physical parameters, and environmental parameters.

[0101] In one embodiment, step S10: For the network topology of the given multi-source feedback data... Y Factoring the conditional probability distribution function to obtain conditionally independent fault factors involves the following steps:

[0102] S11. Network topology based on given multi-source feedback data E Y Calculation of conditional probability distribution function Y The joint distribution term of E ,in, Y is a polynomial variable, represented by the connection states D of the network branches and the connections of the customer switch C;

[0103] S12, Based on random variables The conditional independence between them decomposes the joint distribution term into a set of smaller fault factors, where each fault factor is a conditional probability distribution function composed of sub-variables and parent variables, and the sub-variables include the first fault factor in the feeder. i The connection status of the branch, the first i The first branch j The connection status of each customer switch, network flow fault data at the user load end, and fault signals of smart meters.

[0104] Furthermore, given multi-source feedback data The power grid fault assessment and reasoning process is mathematically formulated using a Bayesian estimator, where the network topology is given by the data feedback set. The conditional probability distribution function is expressed as: And calculated as and The joint distribution term, using The most likely candidate topology, which also determines the location of the fault event, is obtained by maximizing this conditional probability distribution function, as shown in the following formula:

[0105]

[0106] in, It is the most likely network topology after a failure.

[0107] It is a polynomial variable, represented by the connection states D of the main network branches and the connections of the customer switches C, such as... .here, ,in It is the number of branches in the feeder. It is a binary variable representing the first [number] in the feeder. Connection status of each branch: This indicates that the branch is energized. In other words, there is an uninterrupted path between the branch and the substation. This indicates that the branch circuit is not powered. Similarly, ,in Indicates the first Each branch office provides a set of connection states for all customers. Therefore, ,in It is connected to the first Total number of customers in each branch It is the first Customer status: This indicates that the customer has power on. This indicates a power outage for the customer. The topology before the outage was determined by assigning 0 to all state variables (i.e., all branches were energized and the customer was energized). Therefore, Based on newly defined variables The joint probability distribution function can be rewritten as follows:

[0108]

[0109] Thus, maximizing the topological candidates can be conveniently transformed into using their conditional probability distribution functions. and , for belonging to The optimal values ​​are sought for each branch and customer state. These conditional probability distribution functions s are obtained using the marginalization process of the joint probability distribution function. The formula is shown below:

[0110]

[0111]

[0112] The joint probability distribution function needs to be quantized when solving. Given the complexity of distribution networks, obtaining an explicit representation of this joint probability distribution function is difficult to manage. To address the computational complexity and overfitting issues in fault location inference, this embodiment utilizes random variables. The conditional independence between them decomposes the joint probability distribution function into a set of factors with significantly smaller dimensions. Using this computationally efficient method, the conditional probability distribution function of the state of each major branch and customer switch can be inferred from interruption-related data feedback from various data sources to quickly identify the location of lateral interruption events.

[0113] The main idea behind BN-based representations is to use conditional independence encoded in graph structures to compactly decompose a high-dimensional joint probability distribution function with a set of factors. Here, a fault factor refers to a low-dimensional and more manageable conditional probability distribution function, which is determined by two parts: a sub-variable such as and some by The parent variable, for example, Parents represent the direct causal sources of influence for child variables. In other words, each child is a random function of its parents. Therefore, if the values ​​of the parents are known, the child variable will be conditionally independent of random variables that do not directly influence it causally. It can be shown that by applying the chain rule to these conditional independences defined by the parent-child relationship, the joint probability distribution function of a set of random variables can be simplified to the product of the identified factors. In the fault location problem, this decomposition leads to the following data fusion representation of the joint probability distribution function:

[0114]

[0115] In the formula: The factors are: , , and For any . Indicates the first in the feeder i The connection status of each branch Indicates the first i The first branch j Connection status of each customer switch This indicates network flow reports from the user's load side, including load-side device failure signals and social media messages; This indicates meter-based evidence from the client, such as a smart meter fault signal.

[0116] It should be noted that, compared to the original model, it requires... Compared to individual parameters, the formula after decomposing the joint distribution term only requires... There are several parameters. It can be observed that the number of parameters in the new formula is a function of the size of the parent class of each variable. Considering that the number of parent classes of variables is typically small, the new formula achieves a fundamental reduction in the complexity of fault location inference.

[0117] In one embodiment, step S20: parameterizing the fault factor based on available historical statistical interruption information and establishing a Bayesian network for each distributed feeder, specifically includes the following steps:

[0118] S21, the random variable As vertices of the Bayesian network, directed edges are drawn from the vertices, starting from the parent variable and ending at the child variable, to construct the structure of the Bayesian network.

[0119] S22. Parameterize the conditional probability distribution function of each fault factor in the structure of the Bayesian network based on the available historical statistical interruption information, and establish a Bayesian network for each distribution feeder.

[0120] It should be noted that, as a directed acyclic graph, BN provides a convenient way to represent the formulas after factorization. Therefore, random variables Represented as vertices in a Batch Normalization (BN). Using failure factors as vertices in a BN is achieved by drawing directed edges from a parent vertex to its child vertex. BN provides a graphical way to encode conditional independence defined by the following factors: any vertex Conditionally independent of non-descendant vertices in the graph If the value of its parent is known. This is symbolically represented as , It is the vertex set of Batch Normalization (BN), excluding... The parent node does not point to the origin of the path. X , mean A and B It is marginally independent.

[0121] In one embodiment, constructing a BN requires discovering the structure of the graph and the parameters of the conditional probability distribution function. The constructed BN probability graph is, for example... Figures 3 to 4 As shown. This embodiment utilizes grid topology information and causal relationships to reveal the conditional independence between variables, and parameterizes the conditional probability distribution function (i.e., factors) based on available statistical outage information. Accordingly, step S22: parameterizing the conditional probability distribution function of each fault factor in the structure of the Bayesian network based on available historical statistical outage information, and establishing a Bayesian network for each distribution feeder, specifically includes the following steps:

[0122] S221, Regarding fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the branch state variable is... ,in, It is the connection state of adjacent upstream branches. It is the first i Fault feedback from each branch, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch;

[0123] S222, Regarding fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the user state variable is... ;

[0124] S223, Regarding fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution.

[0125] S224, Regarding fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal The conditional probability distribution function of the failure factor is parameterized, becoming conditionally independent of the other variables.

[0126] In one embodiment, S221: for fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, specifically including:

[0127] factor Representative at The independence factor under the condition is chosen as the parent of the branch state variable. ,like Figure 3 As shown. Here, It is the connection state of adjacent upstream branches. It is the first Fault feedback from each branch, among which It can be represented as power grid weather geographic system data, which includes objective environmental factors such as weather and geography; It can provide fault feedback information for the power grid system, including fault signals from distribution boxes, etc. Indicates the first The physical parameters of each branch include conductor length and number of poles. Based on this parent selection scheme for branch state variables, Including the first in the feeder All variables outside the downstream branches. To show the effect of these four variables on... The direct causal effect is described in two cases: and .

[0128] (1) In the first case, when the parent branch is de-energized, then The probability is 1. Therefore, from the substation to... All variables on the path, using It means that, in In the case of, with Conditional independence. Because in a radial network, there can only be one unique path between the substation and each branch; if this path is... Any interruption at any arbitrary point can be automatically calculated. The conclusion is reached regardless of where the path is interrupted. Therefore, considering the variables... The binary properties of the conditional probability distribution function. It can be expressed as:

[0129]

[0130] (2) In the second case, if the adjacent upstream branch is energized, then the first All upstream branches of the branch are also energized, with a probability of 1, indicating they were not affected by the power outage. .in this case, This will only occur when that branch is damaged. Therefore, there are three background variables. , and As the first Causal fault feedback of the branch state to estimate the th branch state The probability of power outage in each branch. Conditional probability distribution function. It can be represented as a Bernoulli distribution, as shown below:

[0131]

[0132] Among them, branches The failure probability is expressed as ,yes , and A function.

[0133] Furthermore, this embodiment utilizes a vulnerability model to formulate this function. The vulnerability model is a series model that can perform vulnerability analysis on each rod and conductor within a branch. In the case of branch estimation Given context variables , and Failure probability:

[0134]

[0135] Where L is used to support the first The number of distribution poles in a branch circuit, K is the number of poles in the branch circuit. The number of conductors between two adjacent poles in a branch circuit, Φ is the standard normal probability integral, χ is the median of the vulnerability function, and ξ is the logarithmic standard deviation of the strength measurement. Indicates the first The probability of failure of the conductor in the branch circuit.

[0136] In one embodiment, step S222: for fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, specifically including:

[0137] factor User representing a given parent variable The conditional probability distribution function of the state. The parent variable of the user state variable is chosen as... .here, It is the supply of the first The status of each customer's direct upstream branch. To display and The accidental relationship between them can be considered in two cases: and .

[0138] (1) In the first case, if the main branch is de-energized, due to the radial structure of the feeder, The probability is 1. Using this deterministic relationship, It can be written as the following formula:

[0139]

[0140] (2) In the second case, if the main branch is energized, then the substation and the first The paths between the branches are valid. In this case, the user experiences a power outage. It can only be caused by overload / fault on the user side. The probability of this occurring is given by the Bernoulli distribution used in statistical power outage information:

[0141]

[0142] In order to explain the parameters Uncertainty, using user-defined hyperparameters and Define a beta distribution:

[0143]

[0144] in, It is a normalization constant, defined as , .

[0145] In one embodiment, step S223: for fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution, specifically including:

[0146] factor Representative at It is independent under certain conditions. It is based on user-generated fault feedback. The parent was chosen as , This refers to the time elapsed after the fault occurred.

[0147] More accurately, T represents the time period that load-side user network equipment needs to wait before sending a power outage report via the public network flow. This is to avoid false alarms from clients due to temporary events. Clearly, there is a trade-off between the number of fault reports based on human intervention and the waiting time inferred from the outage location. For example, when the observability of the actual feeder is extremely low, the utility company might increase... This is to receive more human-based fault feedback for power outage location inference. During the period, in time After the power outage occurred, the time it took to receive user-generated fault feedback was recorded. It follows an exponential distribution:

[0148]

[0149] Therefore, considering , The probability can be calculated as follows:

[0150]

[0151] Therefore, factor The following results can be obtained:

[0152]

[0153] in, This represents a small, user-defined value to account for the possibility of false alarms, such as illegal faulty calls and errors in social media data processing.

[0154] This represents a value that varies depending on the actual situation. Because the time t for receiving fault feedback from the user's device follows an exponential distribution, we have: In various real-world scenarios, the fault feedback time of user-end devices can be simulated as an exponential distribution function, therefore here... The corresponding value is selected from the simulated exponential distribution function.

[0155] In one embodiment, step S224: for fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal To become conditionally independent of other variables, the conditional probability distribution function of the failure factor is parameterized, specifically including:

[0156] factor Is Independence factor under [the specified conditions]. Compared with user-based signals. In contrast, notification mechanisms based on network flows from load-side devices can be transmitted to utility companies almost instantly.

[0157] Therefore, the parent term for fault feedback based on the electricity meter was selected as... When the state of the customer's switch is known, It becomes conditionally independent of the remaining variables, encoded by the following factors:

[0158]

[0159] in, and These represent the reliability of network flow communication on the load side and the probability of meter failure, respectively; specifically... It is the probability that the last breath can be correctly transmitted to the utility for power outage notification. It is the probability that the electricity meter will lose power and send its last signal due to its own malfunction.

[0160] Furthermore, the values ​​of these two parameters are determined based on historical interruption reports. Considering the limited size of historical data, the uncertainty of these two parameters is modeled using a beta distribution as follows:

[0161]

[0162] In the formula: , , , All of these are user-defined hyperparameters, and by default they are all constants of 1; It is a normalization constant, defined as , It is a normalization constant, defined as ,in .

[0163] In one embodiment, step S30: using a sampling algorithm to perform an inference task on the Bayesian network to infer the fault location, specifically includes the following steps:

[0164] S31, the period elapsed after the fault occurred Within this framework, all power outage evidence from client devices, along with branch-level evidence, is collected to construct an evidence sample set. The power outage evidence includes branch-level evidence. i The j The branch-level evidence includes fault data of the load-side network flow of individual customers and fault signals of electricity meters, including power grid system fault feedback information, power grid system physical parameters and environmental parameters;

[0165] S32. Randomly assign any sample from the evidence sample set to all unknown state variables. Choose any state variable as the sampling starting point, where D represents the state of the network branch and C represents the customer switch.

[0166] S33. In the (τ+1)th iteration of Gibbs sampling, according to the structure of the Bayesian network, the data allocated to... Evidence samples of the parent and child variables are inserted into the local Bayesian estimator to approximate the latest samples. The conditional probability distribution function, where, Indicates the first in the feeder i The connection status of each branch;

[0167] S34. Use the inverse transform method to extract a new sample, and perform the local sampling process for the next non-evidence variable until all unknown variables in the Bayesian network have been sampled, thus completing one iteration of Gibbs sampling.

[0168] S35. Based on the sample sequence generated by Gibbs sampling, infer the connection status of all branches and customers, and select the power outage branch closest to the substation as the location of the power outage event.

[0169] It should be noted that after constructing and parameterizing BN, the multi-source data fusion power outage localization process is effectively transformed into probabilistic inference on the graphical model. However, even Simplified, solved and It is still necessary to compute the expensive summation operation simultaneously on all nodes of the graph. This is not scalable for large-scale distribution networks. To address this issue, the GS algorithm can be used to perform inference tasks on BN.

[0170] GS is an approximate inference method based on MCMC (Monovariate Conditional Distribution) that provides a good representation of the probability distribution function by utilizing random variable instantiation without needing to know the mathematical properties of all distributions. A key advantage of this method is its univariate conditional distribution sampling, reducing its dependence on the spatial dimension of the random variables. Therefore, compared to commonly used exact inference methods such as variable elimination and cliff trees, GS is insensitive to the size of the BN (Browser Notation), suggesting that the GS method is particularly advantageous for complex practical applications.

[0171] When a power outage occurs, the probability of a branch / customer power outage can be inferred using the GS algorithm and BN structure. To do this, firstly, After T, collect all evidence of power outages from client devices. If the utility department receives from the branch The Evidence of a customer's load-side equipment network power outage signal or the last gasp signal from a smart meter. or Set to 1. Branch-level evidence. The data was obtained from the data center of the power grid equipment department and the power grid meteorological and geographic information collection system. After collecting all the evidence, arbitrary initial samples were randomly assigned to all unknown state variables. .

[0172] Then, choose an arbitrary state variable as the sampling starting point, for example... In the (τ+1)th iteration of GS, according to the structure of BN, the allocation is... The samples of parents and children are inserted into the local Bayesian estimator, as shown in the following equation, to approximate the sum of the latest samples. Conditional probability distribution function:

[0173]

[0174] in, Except All other recent samples, including the values ​​of the evidence variables, and:

[0175]

[0176]

[0177] therefore, It can be directly calculated using the aforementioned formula representing the fault factor, since... It is a probability distribution function of a single random variable given all other samples, and this calculation can be performed efficiently.

[0178] use Use the inverse transform method to extract a new sample To replace Then, the algorithm moves to the next non-evidence variable in BN and performs a local sampling process.

[0179] Once all unknown variables in Batch Normalization (BN) have been sampled once, one iteration of Gas GS is complete. This process propagates information throughout BN and, combined with data from different sources, effectively infers the location of the power outage. The sampling process is repeated until all unknown variables are sampled. Generate a sufficient number of random samples.

[0180] In one embodiment, step S35: inferring the connection status of all branches / customers based on the sample sequence generated by Gibbs sampling, and selecting the branch closest to the substation as the location of the power outage event, specifically includes:

[0181] The connection status of all branches and customers is inferred from the sample sequences generated by Gibbs sampling, as expressed by the formula:

[0182]

[0183] In the formula: M Indicates the number of iterations. Indicates the first i The sample sequences of each branch, where E represents multi-source feedback data;

[0184] Will By comparing with the set threshold, the first... iThe connection status of each branch and customer is determined, and the branch closest to the substation that experienced a power outage is selected as the location of the power outage event.

[0185] It should be noted that after the GS process, the most probable value of each branch and client state is determined based on the obtained approximate conditional probability distribution function. To achieve this, a threshold of 0.5 is used due to the binary nature of the state variables; for example, Indicates branch It is powered on. After determining the connection status of all branches and customers, the location of the power outage event is determined by selecting the branch closest to the substation that was de-energized.

[0186] Generally, if the iteration time is insufficient, sampling can severely mislead the target distribution, thus reducing the accuracy of inference. Conversely, if the value of M is large enough, the theory of MCMC can guarantee a static distribution of samples generated using the GS algorithm. However, such a strategy leads to high computation time, thereby increasing outage time and cost. Therefore, by using GS, there is a trade-off between the accuracy of outage location and computation time. Figure 2 As shown, in order to find a reasonable maximum number of iterations for a specific BN, a potential size reduction factor R is used to diagnose the convergence of GS at different number of iterations.

[0187] In one embodiment, after step S30: performing an inference task on the Bayesian network using a sampling algorithm to infer the fault location, the method further includes the following steps:

[0188] S40. For the sample sequence generated in each iteration of the sampling algorithm, calculate the inter-sequence variation and intra-sequence variation of the sample sequence;

[0189] S41. Determine the scaling factor based on the inter-sequence variation and the intra-sequence variation;

[0190] S42. Based on the scale reduction coefficient, diagnose the convergence of the sampling algorithm under different iteration numbers and determine the maximum number of iterations.

[0191] In one embodiment, step S40: For the sample sequence generated in each iteration of the sampling algorithm, calculating the inter-sequence variation and intra-sequence variation of the sample sequence specifically includes the following steps:

[0192] S41. For each iteration, the sampling algorithm generates data for each unknown variable in the Bayesian network. n Starting with a sample sequence, each sample sequence is divided into two equal halves to supplement the original sample sequence. All sample sequences are then concatenated into a single sequence of size [number missing]. matrix ;

[0193] Specifically, for each M, we start with n sample sequences generated by GS for each unknown variable in BN. After discarding the samples generated during the warm-up period, each sequence is split into two equal halves, m, and used to supplement the original sequence. All sample sequences are concatenated into a single sequence of size m. matrix .

[0194] S42, Based on the matrix Calculate the inter-sequence variation of the sample sequences. and intra-sequence variation The formula is expressed as:

[0195]

[0196]

[0197] In the formula: Represents the average value within the sequence. ; This represents the population mean. ; Indicates the first j Variance of each sample sequence .

[0198] In one embodiment, the scaling factor is determined based on the inter-sequence variation and the intra-sequence variation, expressed by the following formula:

[0199]

[0200] In the formula: Indicates variation between sequences. Indicates intra-sequence variation. n Indicates the number of samples.

[0201] In theory, when hour, The value is equal to 1. This means that the variance of any estimate can be further reduced through more iterations. In other words, the generated sequence has not yet completed a full examination of the target probability distribution function; or, if The sequence then approximates the target probability distribution function. Here, following previous work, a threshold is used. To choose the value of M. Therefore, Set to satisfy The number of iterations, in the BN structure, for any Both are applicable.

[0202] The power grid fault location inference method proposed in this invention has the following advantages compared with related technologies:

[0203] (1) The complexity of Bayesian networks is different: This invention introduces multi-source heterogeneous fault information to construct a high-dimensional joint probability distribution function, which can greatly improve the accuracy and robustness of the judgment. To solve the analysis of the high-dimensional joint probability distribution function, a graph-based representation method is used to analyze the statistical relationship between random variables. The high-dimensional joint probability distribution function of the system is decomposed into a set of more manageable probability factors.

[0204] (2) Advantages of low latency and real-time performance: This invention uses the Gibbs sampling (GS) algorithm to solve the latency problem. As an algorithm based on Markov chain Monte Carlo (MCMC), GS can provide complete features of the distribution of unknown variables by quickly generating a series of samples.

[0205] In addition, such as Figure 5 As shown, the second embodiment of the present invention proposes a power grid fault location inference system, the system comprising:

[0206] Factorization module 10 is used to factor the network topology of given multi-source feedback data. Y Factorize the conditional probability distribution function to obtain conditionally independent fault factors;

[0207] Parameterization module 20 is used to parameterize the fault factors based on available historical statistical interruption information and establish a Bayesian network for each distributed feeder;

[0208] The inference module 30 is used to perform inference tasks on the Bayesian network using a sampling algorithm to infer the location of the fault.

[0209] This embodiment uses multi-source data fusion to identify and locate transverse fault events in partially observable distribution systems. A probabilistic graphical modeling method is employed for data fusion, decomposing the system's high-dimensional joint probability distribution function into a set of more manageable probability factors. These factors are obtained through conditional independence to reduce the computational complexity representing the system's high-dimensional joint probability distribution function, thereby improving the robustness and accuracy of the fault assessment framework. By establishing a Bayesian network (BN) for each distribution feeder, BN utilizes a graph-based representation as the basis for analyzing the statistical relationships between random variables. The system topology is derived from a single-line graph, and data flow information from user-end network equipment is used. The graphical parameters are learned empirically from historical power outage data. A sampling algorithm is used to perform inference tasks on the Bayesian network. The power outage location process based on data fusion is effectively transformed into online inference on the BN, enabling rapid inference of power grid fault locations.

[0210] In one embodiment, the multi-source data feedback set includes network flow fault data from the user load end, fault signals from the electricity meter, power grid system fault feedback information, power grid system physical parameters, and environmental parameters.

[0211] In one embodiment, the factorization module 10 includes:

[0212] Joint distributed unit, used for network topology based on given multi-source feedback data E Y Calculation of conditional probability distribution function Y The joint distribution term of E ,in, Y is a polynomial variable, represented by the connection states D of the network branches and the connections of the customer switch C;

[0213] Decomposition unit, used for random variables The conditional independence between them decomposes the joint distribution term into a set of smaller fault factors, where each fault factor is a conditional probability distribution function composed of sub-variables and parent variables, and the sub-variables include the first fault factor in the feeder. i The connection status of the branch, the first i The first branch j The connection status of each customer switch, network flow fault data at the user load end, and fault signals of smart meters.

[0214] Furthermore, the decomposed form of the joint distribution term is as follows:

[0215]

[0216] In the formula: , Indicates the first in the feeder i The connection status of each branch Indicates the first i The first branch j Connection status of each customer switch This indicates network flow fault data at the user load end. This indicates a fault signal in the electricity meter.

[0217] In one embodiment, the parameterization module 20 includes:

[0218] Structural building blocks, used to incorporate the random variables As vertices of the Bayesian network, directed edges are drawn from the vertices, starting from the parent variable and ending at the child variable, to construct the structure of the Bayesian network.

[0219] The parameterization unit is used to parameterize the conditional probability distribution function of each fault factor in the structure of the Bayesian network based on the available historical statistical interruption information, and to establish a Bayesian network for each distribution feeder.

[0220] In one embodiment, the parameterization unit is specifically used to perform the following steps:

[0221] For fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the branch state variable is... ,in, It is the connection state of adjacent upstream branches. It is the first i Fault feedback from each branch, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch;

[0222] For fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the user state variable is... ;

[0223] For fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution.

[0224] For fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal The conditional probability distribution function of the failure factor is parameterized, becoming conditionally independent of the other variables.

[0225] Furthermore, the aforementioned fault factors Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, including:

[0226] When the parent branch loses power, based on variables Given the binary nature of the condition, the conditional probability distribution function of the fault factor can be parameterized as follows:

[0227] ;

[0228] When the adjacent upstream branch is energized, the conditional probability distribution function of this fault factor can be expressed as a Bernoulli distribution:

[0229]

[0230] In the formula: Indicates the first i The probability of failure for each branch.

[0231] Furthermore, the first i Failure probability of each branch A vulnerability model was developed for this purpose. , and The function is expressed by the formula:

[0232]

[0233] In the formula: L It is used to support the first i The number of distribution poles on the branch line, K It is the first i The number of conductors between two adjacent poles in a branch circuit, Φ is the standard normal probability integral, χ is the median of the vulnerability function, and ξ is the logarithmic standard deviation of the strength measurement. Indicates the first i The probability of failure of the conductor in the branch circuit.

[0234] Furthermore, the aforementioned fault factors Based on the fact that the parent variable of the user state variable is 1 when the main branch is powered off and when the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized, including:

[0235] When the main branch loses power, the conditional probability distribution function of this fault factor is parameterized as follows:

[0236] ;

[0237] When the main branch is energized, the conditional probability distribution function of this fault factor is parameterized as follows:

[0238]

[0239] In the formula: This indicates a random value.

[0240] Furthermore, the aforementioned fault factors Its parent variable, based on user-generated fault feedback, is During the period after the failure occurred within, within After a power outage occurs, the time based on the user's fault feedback received will be used to determine the cause of the outage. t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution, and its formula is expressed as:

[0241]

[0242] In the formula: Define values ​​for users. This represents a change in value.

[0243] Furthermore, the aforementioned fault factors Its parent variable based on the fault feedback of the electricity meter is Based on the fact that the state of the customer's switch is known, the meter's fault signal The conditional probability distribution function of the failure factor, conditionally independent of other variables, is parameterized as follows:

[0244]

[0245] In the formula: and These represent the reliability of network flow communication on the load side and the probability of meter failure, respectively.

[0246] In one embodiment, the inference module 30 specifically includes:

[0247] An evidence collection unit is used to collect evidence during the period following the occurrence of a failure. Within this framework, all power outage evidence from client devices, along with branch-level evidence, is collected to construct an evidence sample set. The power outage evidence includes branch-level evidence. i The j The branch-level evidence includes fault data of the load-side network flow of individual customers and fault signals of electricity meters, including power grid system fault feedback information, power grid system physical parameters and environmental parameters;

[0248] The sampling start point selection unit is used to randomly assign any sample from the evidence sample set to all unknown state variables. Choose any state variable as the sampling starting point, where D represents the state of the network branch and C represents the customer switch;

[0249] The conditional probability distribution function determination unit is used, in the (τ+1)th iteration of Gibbs sampling, to allocate the distribution function according to the structure of the Bayesian network. Evidence samples of the parent and child variables are inserted into the local Bayesian estimator to approximate the latest samples. The conditional probability distribution function, where, Indicates the first in the feeder i The connection status of each branch;

[0250] The sample extraction unit is used to extract a new sample using the inverse transform method, and to perform the local sampling process of the next non-evidence variable until all unknown variables in the Bayesian network have been sampled, thus completing one iteration of Gibbs sampling.

[0251] The location selection unit is used to infer the connection status of all branches and customers based on the sample sequence generated by Gibbs sampling, and select the power outage branch closest to the substation as the location of the power outage event.

[0252] In one embodiment, the conditional probability distribution function determination unit is approximately given the latest sample. The conditional probability distribution function, where, Indicates the first in the feeder i The connection status of each branch includes:

[0253]

[0254]

[0255]

[0256] In the formula: Except All other recent samples, including the values ​​of the evidence variables, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch, Indicates the first i-1 The state variables of the network branches in the τth iteration Indicates the first i+1 The state variables of the network branches in the τth iteration Indicates the first i+1 In the network branch, the first j The state variables of a user network-side switch in the τth iteration.

[0257] In one embodiment, the position selection unit is specifically used for:

[0258] The connection status of all branches / clients is inferred from the sample sequences generated by Gibbs sampling, as expressed by the formula:

[0259]

[0260] In the formula: M Indicates the number of iterations. Indicates the first i The sample sequences of each branch, where E represents multi-source feedback data;

[0261] Will By comparing with the set threshold, the first... i The connection status of each branch and customer is determined, and the branch closest to the substation that experienced a power outage is selected as the location of the power outage event.

[0262] In one embodiment, the system further includes a position calibration module, wherein the position calibration module specifically includes:

[0263] The mutation calculation unit is used to calculate the inter-sequence mutation and intra-sequence mutation of the sample sequence generated in each iteration of the sampling algorithm.

[0264] A coefficient determination unit is used to determine a scaling factor based on the inter-sequence variation and the intra-sequence variation;

[0265] The diagnostic unit is used to diagnose the convergence of the sampling algorithm under different iteration numbers based on the scale reduction coefficient, and to determine the maximum number of iterations.

[0266] In one embodiment, the mutation calculation unit is specifically used for:

[0267] For each iteration, the sampling algorithm generates data for each unknown variable in the Bayesian network. n Starting with a sample sequence, each sample sequence is divided into two equal halves to supplement the original sample sequence. All sample sequences are then concatenated into a single sequence of size [number missing]. matrix ;

[0268] Based on the matrix Calculate the inter-sequence variation of the sample sequences. and intra-sequence variation The formula is expressed as:

[0269]

[0270]

[0271] In the formula: Represents the average value within the sequence. This represents the population mean. Indicates the first j Variance of individual sample sequences.

[0272] In one embodiment, the coefficient determining unit determines the scaling factor, expressed by the formula:

[0273]

[0274] In the formula: Indicates variation between sequences. Indicates intra-sequence variation. n Indicates the number of samples.

[0275] This embodiment employs probabilistic graphical modeling to achieve data fusion, which can seamlessly integrate heterogeneous data sources. Different data sources can complement each other, increasing the amount of power outage information and thus solving the problems of low coverage of smart fault feedback devices or low network flow reporting rates of user-end devices in actual power grids. Simultaneously, by utilizing the inherent conditional independence between evidence and state variables in the distribution system, the exponential computational complexity of the power outage location task is reduced to a linear complexity based on the number of variables.

[0276] It should be noted that other embodiments or implementation methods of the power grid fault location inference method system described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0277] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0278] Furthermore, 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0279] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method of inferring a fault location in an electric power network, characterized by, The method comprises: Factorizing conditional probability distribution function of network topology of given multi-source feedback data Y to obtain conditionally independent fault factors parameterizing the failure factors from available historical statistical outage information, establishing a Bayesian network for each distribution feeder, including a random variable constructing a structure of the Bayesian network from vertices, drawing directed edges from a parent variable to a child variable, starting from a vertex, E is multi-source feedback data, D is a connection state of a network branch in a feeder, and C is a customer switch. Based on available historical statistical interruption information, the conditional probability distribution function of each fault factor in the Bayesian network structure is parameterized, and a Bayesian network is established for each distribution feeder. Specifically, this includes parameterizing the conditional probability distribution function of each fault factor. Based on the fact that the parent variable of the branch state variable is 1 when the parent branch is de-energized and the adjacent upstream branch is energized, the conditional probability distribution function of the fault factor is parameterized, and the parent variable of the branch state variable is... ,in, For the first feeder i The connection status of each branch It refers to the connection state of adjacent upstream branches. It is the first i Fault feedback from each branch, Indicates environmental parameters of the power grid system; This indicates fault feedback information in the power grid system. Indicates the first i Physical parameters of the power grid system in each branch; For the fault factor , the conditional probability distribution function of the fault factor is parameterized based on the parent variables of the user state variables when the main branch is powered off and the main branch is powered on, the parent variables of the user state variables are , , and the connection state of the first customer switch of the first branch is i . j ​ For the fault factor The parent variable based on the user's artificial fault feedback is The period elapsed after the fault occurs In After the power outage occurs at the moment, according to the time of receiving the user's artificial fault feedback t The conditional probability distribution function of the fault factor is parameterized according to the exponential distribution, Indicates the network flow fault data at the user load end; For the fault factor its parent variable based on the fault feedback of the meter is based on the fault signal of the meter when the state of the customer switch is known the conditional probability distribution function for this fault factor is parameterized conditionally independent from the rest of the variables; The inference task is performed on the Bayesian network by using a sampling algorithm to infer the fault location.

2. The power system fault location method of claim 1, wherein, The multi-source feedback data comprises network flow fault data at a user load end, fault signals of an electric meter, power grid system fault feedback information, power grid system physical parameters and environmental parameters.

3. The power grid fault location method of claim 1, wherein, The network topology of the given multi-source feedback data Y is factorized to obtain conditionally independent fault factors, including: Network topology based on given multi-source feedback data E Y Conditional probability distribution function computation Y Joint distribution term of E and E wherein, Y is a polynomial variable represented by the connection state D of the network branches in the feeder and the connection of the customer switch C; Based on random variables The conditional independence between them decomposes the joint distribution term into a set of smaller fault factors, where each fault factor is a conditional probability distribution function composed of sub-variables and parent variables, and the sub-variables include the first fault factor in the feeder. i The connection status of the branch, the first i The first branch j The connection status of each customer switch, network flow fault data at the user load end, and fault signals of smart meters.

4. The power grid fault location method of claim 3, wherein, The expression of the joint distribution item after decomposition is: In the formula: , represents the connection state of the i th branch in the feeder, represents the connection state of the i th branch and the j th customer switch, represents the network flow failure data of the user load end, represents the failure signal of the electric meter.

5. The power grid fault location method of claim 1, wherein, The conditional probability distribution function for the fault factor is parameterized based on the parent variable of the branch state variable being 1 when the parent branch is de-energized and the adjacent upstream branch is energized, including: When a parent branch is de-energized, the conditional probability distribution function of the fault factor is parameterized based on the binary nature of the connection status of the branches in the feeder i as:​ ； When the adjacent upstream branch is powered on, the conditional probability distribution function of the fault factor is expressed as a Bernoulli distribution: wherein: represents the failure probability of the i branch.

6. The power grid fault location method of claim 5, wherein, The failure probability of the first branch i is a function formulated using a vulnerability model, expressed by the equation: , and ​​ where: L is the number of distribution poles for the i branch, K is the number of conductors between two adjacent poles of the i branch, Φ is the standard normal probability integral, χ is the median of the vulnerability function, and ξ is the logarithmic standard deviation of the strength measure, is the failure probability of the conductor of the i branch.

7. The power grid fault location method of claim 1, wherein, The condition probability distribution function for the fault factor is parameterized based on the parent variable of the user state variable being 1 at the time of the main branch outage and the main branch energization. When the main branch is powered off, the conditional probability distribution function of the fault factor is parameterized as: ； When the main branch is powered on, the conditional probability distribution function of the fault factor is parameterized as: In the formulae: denotes a random value.

8. The power grid fault location method of claim 1, wherein, The failure factor The parent variable based on the user's human failure feedback is The period elapsed after the failure The period elapsed after the failure After the power outage at time t, according to the time of receiving the user's human failure feedback t The conditional probability distribution function of the failure factor is parameterized according to the exponential distribution, which is represented as: In the formulae: is a user-defined value, denotes a change value.

9. The power grid fault location method of claim 1, wherein, The fault factor whose parent variables are based on the fault feedback of the electric meter whose fault signal becomes conditionally independent from the rest of the variables, parameterizing the conditional probability distribution function of this fault factor, expressed as: In the formula: and respectively represent the load side device network flow communication reliability and the probability of meter failure values.

10. The power grid fault location method of claim 1, wherein, The inference task is performed on the Bayesian network by using a sampling algorithm to infer the fault location, comprising: a period of time elapsed after the fault occurs collects all outage evidences from the client device and branch level evidences to build an evidence sample set, wherein the outage evidences include branch i load end network flow failure data of the first j customer and failure signals of the electric meter, and the branch level evidences include power grid system failure feedback information, power grid system physical parameters and environmental parameters; randomly assigning any sample in the set of evidence samples to all unknown state variables selecting any one state variable as a starting point for sampling, wherein D represents the connection status of network branches in the feeder line, and C represents a customer switch In the (t+1)th iteration of Gibbs sampling, the evidence samples assigned to the parent and child variables of are inserted into the local Bayesian estimators to approximate the conditional probability distribution functions given the latest where represents the connection state of the i th branch in the feeder; A new sample is extracted by using an inverse transformation method, a local sampling process of a next non-evidence variable is performed, and after all unknown variables in the Bayesian network are sampled, one iteration of the Gibbs sampling is completed; The connection states of all branches and customers are inferred based on the sample sequence generated by the Gibbs sampling, and the power-off branch closest to the transformer substation is selected as the location of the power-off event.

11. The power grid fault location method of claim 10, wherein, said τ+1th iteration of Gibbs sampling, the evidence samples assigned to the parent and child variables of are inserted into the local Bayesian estimator to approximate the conditional probability distribution function given the most recent wherein represents the connection state of the i th branch in the feeder, comprising: wherein: is the value of the evidence variable for all recent samples except represents the grid system environment parameters; represents the grid system fault feedback information, represents the grid system physical parameters of the i th branch, represents the state variable of the i-1 th network branch in the τth iteration, represents the state variable of the i+1 th network branch in the τth iteration, represents the state variable of the i+1 th user network end switch in the j th network branch in the τth iteration.​ 12. The power grid fault location method of claim 10, wherein, The connection states of all branches and customers are inferred based on the sample sequence generated by the Gibbs sampling, and the power-off branch closest to the transformer substation is selected as the location of the power-off event, comprising: The connection states of all branches and customers are inferred based on the sample sequence generated by the Gibbs sampling, and the connection states are expressed by a formula as: In the formula: M represents the number of iterations, represents the sample sequence of the i branch, and E represents the multi-source feedback data. The comparison with the set threshold infers the connection status of the i branch and the customer and selects the branch closest to the substation as the location of the outage event.

13. The electric power network fault location method according to any one of claims 1 to 12, characterized by, After the inference task is performed on the Bayesian network by using the sampling algorithm to infer the fault location, the method further comprises: For the sample sequence generated by each iteration of the sampling algorithm, sequence inter-variation and sequence intra-variation of the sample sequence are calculated; Based on the sequence inter-variation and the sequence intra-variation, a scale reduction coefficient is determined; Based on the scale reduction coefficient, the convergence of the sampling algorithm at different iteration numbers is diagnosed to determine a maximum iteration number.

14. The power grid fault location method of claim 13, wherein, For the sample sequence generated by each iteration of the sampling algorithm, sequence inter-variation and sequence intra-variation of the sample sequence are calculated, comprising: For each iteration, the sampling algorithm generates data for each unknown variable in the Bayesian network. n Starting with a sample sequence, each sample sequence is divided into two equal halves to supplement the original sample sequence. All sample sequences are then concatenated into a single sequence of size [number missing]. matrix ; based on the matrix , calculate inter-sequence variation of the sample sequence and intra-sequence variation , expressed as: where: denotes the average value within a sequence, denotes the overall mean value, denotes the variance of the sequence of the j first n samples.

15. The power grid fault location method of claim 13, wherein, Based on the sequence inter-variation and the sequence intra-variation, a scale reduction coefficient is determined, and the scale reduction coefficient is expressed by a formula as: wherein: represents inter-sequence variation, represents intra-sequence variation, n represents the number of samples.

16. A power grid fault location inference system characterized by, The system is used to implement the power grid fault location inference method according to any one of claims 1-15, and the system comprises: The factorization module is used to factor the network topology of given multi-source feedback data. Y Factorize the conditional probability distribution function to obtain conditionally independent fault factors; A parameterization module is configured to parameterize the fault factor according to available historical statistical outage information, and to establish a Bayesian network for each distribution feeder; An inference module is configured to perform an inference task on the Bayesian network by using a sampling algorithm to infer the fault location.

Images