CN116680635A

CN116680635A - Power grid fault position inference method and system

Info

Publication number: CN116680635A
Application number: CN202310677656.6A
Authority: CN
Inventors: 骆晨; 阮翔勇; 汪柏松; 冯玉; 吴凯; 吴少雷; 周建军; 陈振宁; 胡钰杰; 左宇翔; 李博; 邵珺伟; 卞真旭; 张晨晨
Original assignee: State Grid Anhui Electric Power Co Ltd Anqing Power Supply Co; Electric Power Research Institute of State Grid Anhui Electric Power Co Ltd; Hefei Power Supply Co of State Grid Anhui Electric Power Co Ltd
Current assignee: State Grid Anhui Electric Power Co Ltd Anqing Power Supply Co; Electric Power Research Institute of State Grid Anhui Electric Power Co Ltd; Hefei Power Supply Co of State Grid Anhui Electric Power Co Ltd
Priority date: 2023-06-07
Filing date: 2023-06-07
Publication date: 2023-09-01

Abstract

The invention discloses a power grid fault position inference method and a system, comprising the steps of factoring a conditional probability distribution function of a network topology Y given multi-source feedback data to obtain a fault factor with independent conditions; parameterizing the fault factors according to available historical statistical interruption information, and establishing a Bayesian network for each distributed feeder line; and (3) performing an inference task on the Bayesian network by adopting a sampling algorithm, and deducing the fault position. The invention adopts a probability graph modeling method for data fusion, and the high-dimensional joint probability distribution function of the system is decomposed into a group of probability factors which are easier to manage, and the factors are obtained by conditional independence, so that the calculation complexity of the high-dimensional joint probability distribution function of the representative system is reduced, and the robustness and the accuracy of a fault studying and judging frame are improved.

Description

Power grid fault position inference method and system

Technical Field

The invention relates to the technical field of big data mining of power distribution networks, in particular to a power grid fault position inference method and a power grid fault position inference system.

Background

The electrical network includes a power generation, transmission and distribution network, and the distribution system is the last step of transmitting power to individual consumers, and as the demand for power and the number of power consumers are increasing rapidly in the past decades, the distribution network has to expand distribution lines and feeder lines to meet the demands of consumers, and as the society has higher and higher dependence on energy, the demand for power reliability has higher and higher.

One of the basic tasks of power system operators is fault determination and localization, and faults may cause problems such as network equipment damage, service interruption, unstable network, etc., thereby reducing network reliability and causing economic loss to customers and power companies. The traditional method for judging and positioning the feeder line faults of the power distribution network is low in efficiency, particularly when the geographical distribution range of the network is wide, the cost for judging and positioning the faults is high and the time consumption is long in terms of required manpower and equipment, so that the method for automatically and quickly predicting and positioning the faults has important significance in the power distribution network. An automatic fault prediction and location research framework has the advantages of saving time, saving human resources, enhancing system maintenance power preparation, modifying future plans, improving economic factors, and the like, which improve customer satisfaction and improve system reliability indexes.

The main cause of power system energy faults is power faults, especially in power distribution systems (line to ground faults account for about 70%), which are caused by a number of factors, such as utility faults, lighting, storms, bad weather, rain, insulation faults, trees, birds, etc. The considerable uncertainty of these data sources may lead to erroneous fault locations and additional costs for the utility, e.g., due to hardware and software issues, only a portion of the fault signals may be transmitted to the utility's data center. Aiming at the limitation and uncertainty of a single data source, how to quickly and accurately locate the power distribution system faults from the multi-source factors and multi-source fault feedback data is a key problem, and a reliable power grid fault research and judgment framework facing low time delay and high robustness under a multi-source massive data scene can improve the system reliability and power supply continuity and accelerate the power supply recovery speed, so that the service downtime is shortened.

Aiming at the research of a power grid fault research and judgment framework of a traditional power grid in a multi-source mass data scene, two key problems are faced: (1) One fundamental challenge in the scenario of multi-source mass data for power grid fault determination is the computational complexity of the problem, which greatly affects the fault reflection delay of the power grid. Because fault location inference is a process that calculates the probability of topology candidates after an outage event using available information received by the utility, estimating these probability values requires obtaining a joint probability distribution function of unknown state variables and evidence, which is a high-dimensional mathematical computation object. For practical power distribution systems, it is computationally infeasible to directly quantify such joint distributions, which requires enumerating the probabilities of all possible variable combinations. (2) The fault determination accuracy is to be improved because the fault information feedback data source has the characteristic of heterogeneity, such as accuracy and reporting rate; and multi-source data may provide inconsistent or even opposite information, it is a challenge to integrate these data sources and improve the accuracy of fault determination under feedback of the multi-source data.

In the related technology, the Chinese patent application document with the application publication number of CN113791307A discloses a method for positioning a fault section of a hybrid line distribution network based on a discrete Bayesian network, wherein a Bayesian probability network is constructed through fault information of a line measuring device, and finally, parameters of the discrete Bayesian network are trained by using an expected maximum algorithm according to historical fault information; and reasoning the discrete Bayesian network by using a confidence propagation algorithm to obtain the fault state of each line segment under the current observation information. However, the scheme uses single line measurement device data to construct a single distributed Bayesian probability network, but only a single fault data source is used for studying and judging, so that the error is increased; in addition, the scheme infers the discrete Bayesian network through a confidence propagation algorithm, which is time-consuming, and can suffer from high delay risks in the application of large power grid multi-data.

The Chinese patent application publication No. CN113725862A discloses a same-parent power distribution network topology identification method based on a Bayesian network, wherein a Bayesian probability network is constructed through voltage fluctuation and voltage power information, and finally a confidence propagation algorithm is also used for reasoning a discrete Bayesian network, so that fault states of all line segments under current observation information are obtained. The scheme uses single voltage fluctuation and voltage power information to construct a single distributed Bayesian probability network, but in this way, only a single fault data source is used for studying and judging, and the error is increased; whereas discrete bayesian networks are inferred by belief propagation algorithms, which are time consuming, the risk of high delays may be encountered in large grid multi-data applications.

Disclosure of Invention

The technical problem to be solved by the invention is how to provide a power grid fault position deducing method with small calculation complexity and low time delay.

The invention solves the technical problems by the following technical means:

in a first aspect, the present invention proposes a method of power grid fault location inference, the method comprising the steps of:

factorization is carried out on the conditional probability distribution function of the network topology Y given the multi-source feedback data, so as to obtain a fault factor with independent conditions;

parameterizing the fault factors according to available historical statistical interruption information, and establishing a Bayesian network for each distributed feeder line;

and (3) performing an inference task on the Bayesian network by adopting a sampling algorithm, and deducing the fault position.

Further, the multi-source data feedback set comprises network flow fault data of a user load end, fault signals of an ammeter, power grid system fault feedback information, power grid system physical parameters and environment parameters.

Further, the factoring the conditional probability distribution function of the network topology Y given the multi-source feedback data to obtain a fault factor with independent conditions includes:

Calculating a joint distribution term P (Y, E) of Y and E based on a conditional probability distribution function of a network topology Y given multisource feedback data E, wherein Y is a polynomial variable represented by the state of the network branch D and the connection of the customer switch C;

based on the conditional independence among random variables { D, C, E }, decomposing the joint distribution item into a group of fault factors with smaller sizes, wherein each fault factor is a conditional probability distribution function formed by a child variable and a father variable, and the child variable comprises the connection state of an ith branch, the connection state of a jth customer switch of the ith branch, network flow fault data of a user load end and fault signals of a smart meter.

Further, the decomposed representation of the joint distribution item is as follows:

wherein: u= |e|, D _i Representing the connection status of the i-th branch in the feeder,representing the connection status of the j-th customer exchange of the i-th branch,/and%>Network flow failure data representing user load side, < >>A fault signal indicative of the electricity meter.

Further, the parameterizing the fault factor according to available historical statistical interruption information, and establishing a bayesian network for each distributed feeder line, including:

Taking the random variables { D, C, E } as vertexes of the Bayesian network, drawing directed edges from parent variables to child variables from the vertexes, and constructing a structure of the Bayesian network;

and parameterizing a conditional probability distribution function of each fault factor in the structure of the Bayesian network according to the available historical statistical interruption information, and establishing a Bayesian network for each distribution feeder line.

Further, the parameterizing the conditional probability distribution function of each fault factor in the structure of the bayesian network according to the available historical statistical interruption information, and establishing a bayesian network for each distribution feeder line, including:

for the fault factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, parameterizing the conditional probability distribution function of the fault factor, wherein the parent variable of the branch state variable isWherein D is _i-1 Is the state flow branch of the adjacent upper layer, +.>Is the fault feedback of the ith branch, +.>Representing environmental parameters of a power grid system; />Representing power grid system fault feedback information->Representing the physical parameters of the power grid system of the ith branch;

for fault factors Based on the parent variable of the user state variable being 1 at main branch power-off and main branch power-on, the fault factor is calculatedParameterizing a conditional probability distribution function, the parent variable of the user state variable being +.>

For fault factorsIts parent variable based on fault feedback by user isIn the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to exponential distribution according to the received artificial fault feedback time t of the user;

for fault factorsIts parent variable based on fault feedback of ammeter isBased on the fault signal of the electricity meter when the state of the customer switch is known +.>The conditional probability distribution function of the fault factor is parameterized to be conditionally independent of the remaining variables.

Further, the fault factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, parameterizing the conditional probability distribution function of the fault factor, comprising:

based on variable D when parent branch is powered down _i Parameterizing the conditional probability distribution function of the fault factor is expressed as:

When the adjacent upstream branch is powered on, the conditional probability distribution function of the fault factor is expressed as Bernoulli distribution:

wherein: p (P) _l ⁱ Indicating the probability of failure of the i-th branch.

Further, the failure probability P of the ith branch _l ⁱ One concern for modeling vulnerabilityAnd->Is expressed as:

wherein: l is the number of distribution bars used to support the ith branch, K is the number of wires between two adjacent poles of the ith branch, Φ is the standard normal probability integral, χ is the median of the vulnerability function, ζ is the logarithmic standard deviation of the intensity measurements,indicating the probability of failure of the conductor of the i-th branch.

Further, the fault factorBased on user state variables at main branch power-down and main branch power-upThe parent variable is 1, and parameterizing the conditional probability distribution function of the fault factor comprises:

when the main branch is powered off, the conditional probability distribution function of the fault factor is parameterized as follows:

the conditional probability distribution function of the fault factor is parameterized at main branch power-on as:

wherein: pi ₂ Is a random value.

Further, the fault factorIts parent variable based on user-artificial fault feedback is +. >In the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to the received artificial fault feedback time t of the user and the exponential distribution, and the conditional probability distribution function is shown as follows:

wherein: pi ₃ For user-defined values lambda ₁ Is a variable value.

Further, the fault factorIts parent variable based on fault feedback of the electricity meter is +.>Based on the fault signal of the electricity meter when the state of the customer switch is known +.>The conditional probability distribution function of the fault factor is parameterized, conditionally independent of the remaining variables, expressed as:

wherein: pi ₄ And pi ₅ Representing the network flow communication reliability of the load side equipment and the fault probability value of the ammeter respectively.

Further, the performing an inference task on the bayesian network by adopting a sampling algorithm to infer a fault location comprises:

collecting all outage evidences from the client device and branch-level evidences to construct an evidence sample set in a period delta T after the fault occurs, wherein the outage evidences comprise load-end network flow fault data and fault signals of an ammeter of a jth client of a branch i, and the branch-level evidences comprise power grid system fault feedback information, power grid system physical parameters and environment parameters;

Randomly assigning any sample in the evidence sample set to all unknown state variables { D, C }, selecting any one state variable as a sampling start point, wherein D represents the state of a network branch, C represents a client switch,

in the (t+1) th iteration of Gibbs sampling, D will be assigned according to the structure of the Bayesian network _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the ith in the feederThe connection state of the branch;

extracting a new sample by using an inverse transformation method, and executing the local sampling process of the next non-evidence variable until all unknown variables in the Bayesian network are sampled, so as to complete one iteration of Gibbs sampling;

and deducing the connection states of all branches and clients based on a sample sequence generated by Gibbs sampling, and selecting a power-off branch closest to a transformer substation as the position of a power failure event.

Further, in the τ+1th iteration of Gibbs sampling, the structure of the Bayesian network is allocated to D _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line, comprising:

wherein: d, d _-i ^(τ) Is to remove d _i ^(τ) All but the most recent samples include the value of the evidence variable,representing environmental parameters of a power grid system; />Representing power grid system fault feedback information->Representing the physical parameters of the power grid system of the ith branch, d _i-1 ^(τ) Representing the state variable of the i-1 th network branch in the τ -th iteration, d _i+1 ^(τ) Representing the state variable of the (i+1) th network branch in the (t) th iteration, +.>Representing the state variable of the jth user network side switch in the ith+1th network branch in the τ iteration.

Further, the sample sequence generated based on Gibbs sampling deduces the connection states of all branches and clients, and selects the power-off branch closest to the transformer substation as the position of the power failure event, including:

the connection state of all branches and clients is deduced based on a sample sequence generated by Gibbs sampling, and the formula is as follows:

wherein: m represents the number of iterations and,a sample sequence representing an ith branch, E representing multi-source feedback data;

p (D) _i =1|e) is compared with a set threshold value, the connection state of the ith branch and the customer is deduced, and the power-off branch closest to the substation is selected as the position of the power failure event.

Further, after performing an inference task on the bayesian network using a sampling algorithm to infer a fault location, the method further comprises:

for a sample sequence generated by each iteration of a sampling algorithm, calculating the inter-sequence variation and the intra-sequence variation of the sample sequence;

determining a scaling factor based on the inter-sequence variation and the intra-sequence variation;

and based on the scale-down coefficient, diagnosing the convergence of the sampling algorithm under different iteration numbers, and determining the maximum iteration times.

Further, the calculating, for the sample sequence generated by each iteration of the sampling algorithm, the inter-sequence variation and the intra-sequence variation of the sample sequence includes:

for each iterative process, starting from n sample sequences generated by a sampling algorithm for each unknown variable in the Bayesian network, dividing each sample sequence into two halves with the same size for supplementing an original sample sequence, and connecting all sample sequences in series into a matrix theta with the size of 2n multiplied by m;

calculating the inter-sequence variation B of the sample sequence based on the matrix θ _i And intra-sequence variation V _i The formula is:

wherein:represents the mean value within the sequence,/- >Represents the overall mean value->Representing the j-th sample sequence variance.

Further, the determining a scaling factor based on the inter-sequence variation and the intra-sequence variation, the equation being:

wherein: b (B) _i Representing between sequencesVariation, V _i Intra-sequence variation is represented, and n represents the number of samples.

In a second aspect, the present invention proposes a power grid fault location inference system, the system comprising:

the factorization module is used for factorizing a conditional probability distribution function of the network topology Y given the multi-source feedback data to obtain a fault factor with independent conditions;

the parameterization module is used for parameterizing the fault factors according to available historical statistics interruption information and establishing a Bayesian network for each distributed feeder line;

and the reasoning module is used for executing the reasoning task on the Bayesian network by adopting a sampling algorithm and deducing the fault position.

The invention has the advantages that:

(1) The invention identifies and locates the transverse fault event in the partly observable distribution system based on multi-source data fusion, adopts a probability graph modeling method for data fusion, and the high-dimensional joint probability distribution function of the system is decomposed into a group of probability factors which are easier to manage, and the factors 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 the accuracy of a fault research and judgment frame; by establishing a Bayesian network (BayesianNetwork, BN) for each distributed feeder line, BN uses a graph-based representation method as a basis for analyzing statistical relationships between random variables, system topology from a single line graph and data flow information from user side network equipment, and graph parameters are learned from historical power outage data according to experience; the sampling algorithm is adopted to execute the reasoning task on the Bayesian network, the power failure positioning process based on data fusion is effectively converted into online reasoning of BN, and the power network fault position can be rapidly deduced.

(2) The method provided by the invention can seamlessly integrate heterogeneous data sources. Different data sources can be mutually supplemented, and the power outage information quantity is increased, so that the problem that the coverage rate of intelligent power grid fault feedback equipment or the network flow reporting rate of user side equipment in an actual power grid is low is solved. At the same time, by taking advantage of the conditional independence inherent between evidence and state variables in the power distribution system, the exponential computational complexity of the outage localization task is reduced to the linear complexity of the number of variables.

(3) The present invention incorporates prior expert knowledge of the utility into a fault inference model. The fault locating process based on data fusion is effectively converted into online reasoning of BN; the reasoning task is solved by using Gibbs Sampling (GS) algorithm, which is a markov chain monte carlo (Markov Chain Monte Carlo, MCMC) based algorithm that can provide a complete characterization of the unknown variable distribution by generating a series of samples.

(4) Because the uncertainty of each data source is definitely modeled by using the probability map parameters, the proposed method is robust in terms of false alarm and inconsistency of outage evidence, and the fault judgment accuracy is high.

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.

Drawings

FIG. 1 is a flow chart of a power grid fault location inference method according to the present invention;

FIG. 2 is a diagram of a power grid fault location inference framework in accordance with the present invention;

FIG. 3 is a schematic diagram of BN structure construction and parameterization in the present invention;

FIG. 4 is a schematic block diagram of a BN structure of a typical radiation distribution system constructed based on various fault factors in accordance with the present invention;

fig. 5 is a schematic structural diagram of a power grid fault location inference system according to the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As shown in fig. 1, the invention provides a power grid fault location inference method, which comprises the following steps:

s10, factorizing a conditional probability distribution function of a network topology Y given multi-source feedback data to obtain a fault factor with independent conditions;

S20, parameterizing the fault factors according to available historical statistics interruption information, and establishing a Bayesian network for each distributed feeder line;

s30, performing an inference task on the Bayesian network by adopting a sampling algorithm, and deducing the fault position.

The embodiment identifies and locates the transverse fault event in the partially observable distribution system based on multi-source data fusion, adopts a probability graph modeling method for data fusion, and the high-dimensional joint probability distribution function of the system is decomposed into a group of probability factors which are easier to manage, and the factors 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 the accuracy of a fault research and judgment framework; establishing a Bayesian network BN for each distributed feeder line, wherein the BN uses a graph-based representation method as a basis for analyzing statistical relationships among random variables, a system topology structure from a single line diagram and data flow information from user side network equipment, and graph parameters are learned from historical power failure data according to experience; the sampling algorithm is adopted to execute the reasoning task on the Bayesian network, the power failure positioning process based on data fusion is effectively converted into online reasoning of BN, and the power network fault position can be rapidly deduced.

In an embodiment, the multi-source data feedback set includes network flow fault data of a user load end, fault signals of an ammeter, power grid system fault feedback information, power grid system physical parameters and environment parameters.

In one embodiment, the step S10: factorization is carried out on a conditional probability distribution function of the network topology Y of given multi-source feedback data to obtain a fault factor with independent conditions, and the method specifically comprises the following steps:

s11, calculating joint distribution items P (Y, E) of Y and E based on a conditional probability distribution function of a network topology Y of given multi-source feedback data E, wherein Y is a polynomial variable and is represented by the state of a network branch D and the connection of a client switch C;

s12, decomposing the joint distribution item into a group of fault factors with smaller sizes based on conditional independence among random variables { D, C and E }, wherein each fault factor is a conditional probability distribution function formed by a child variable and a father variable, and the child variable comprises a connection state of an ith branch, a connection state of a jth customer switch of the ith branch, network flow fault data of a user load end and fault signals of a smart meter in a feeder line.

Further, given the multisource feedback data E, the grid fault finding reasoning process is mathematically formulated using a bayesian estimator, where the conditional probability distribution function of the network topology Y for a given data feedback set is denoted as P (y|e) and calculated as a joint distribution term of Y and E, denoted as P (Y, E). The most likely candidate topology also determines the location of the fault event, and is derived by maximizing this conditional probability distribution function, as follows:

Wherein Y is ^* Is the most likely network topology after failure.

Y is a polynomial variable represented by the state of the primary network branch D and the connection of the customer switch C, as y= { D, C }. Here, d= [ D ₁ ,…,D _k ]Where k is the number of branches in the feed line, D _i Is a binary variable representing the connection state of the ith branch in the feeder: d (D) _i =0 indicates that the branch is powered on. In other words, there is an uninterrupted path between the branch and the substation, D _i =1 indicates that the branch is not energized. Similarly, c= [ C ₁ ,…,C _k ]Wherein C _i Representing the provision of the ith branch officeConnection state set of all clients. Thus, the first and second substrates are bonded together,wherein z is _i Is the total number of clients connected to the ith branch, is->Is the state of the jth client: />Indicating that the client is powered on +.>Indicating that the customer is powered down. The pre-outage topology is determined by assigning 0's to all state variables (i.e., all branches are powered on and clients are powered on). Thus, P (y|e) can be rewritten according to the joint probability distribution function of the newly defined variable P (D, C, E) as follows:

in this way, maximization of topology candidates can be conveniently translated into a conditional probability distribution function using them, P (D _i |e) andfind the best value for each branch and client state belonging to { D, C }. These conditional probability distribution functions s are obtained using the marginalization process of the joint probability distribution function +. >The formula is as follows:

the joint probability distribution function P (D, C, E) needs to be quantized when solving. Taking into account the complexity of the distribution network, it is difficult to obtain an explicit representation of this joint probability distribution function, which in order to cope with the computational complexity and the overfitting in fault location inference, the present embodiment exploits the conditional independence between the random variables { D, C, E } to decompose the joint probability distribution function into a set of significantly smaller-sized factors. Using this computationally efficient method, a conditional probability distribution function of the state of each primary branch and client switch can be inferred from interrupt-related data feedback from various data sources to quickly identify the location of a lateral interrupt event.

The main idea of BN-based representation is to compactly decompose a high-dimensional joint probability distribution function with a set of factors using the conditional independence encoded in the graph structure. A fault factor is referred to herein as a low-dimensional and more manageable conditional probability distribution function, which is determined in two parts: one sub-variable, e.g. D _i And some parent variables represented by Pa (,), e.g., pa (D) _i ) Parents represent a direct causal impact source for child variables. In other words, each child is a random function of its parents. Thus, if the value of the parent is known, the child variable will be conditionally independent of random variables that do not directly influence it in a causal manner. It can be demonstrated that by using the chain law for these conditional independence defined by the parent-child relationships, the joint probability distribution function of a set of random variables can be reduced to the product of the identified factors. In the fault localization problem, this decomposition results in the following data fusion representation of the joint probability distribution function:

Wherein: u= |e|, the factor is: p (D) _i ∣Pa(D _i ))，Andfor arbitrary->D _i Indicating the connection status of the i-th branch in the feeder, etc.>Representing the connection status of the j-th customer exchange of the i-th branch,/and%>Representing a network flow report from a user load side, including a load side device failure signal and social media messages; />Representing meter-based evidence, such as a smart meter fault signal, from the client.

It should be noted that the model is 2 to the original model ^r Compared with 1 independent parameter, the formula after the decomposition of the joint distribution term only needs to beAnd parameters. It can be observed that the number of parameters in the new formula is a function of the parent size of each variable. Considering that the number of parent classes of variables is generally small, the new formula enables a radical reduction in the complexity of fault location inference.

In one embodiment, the step S20: parameterizing the fault factor according to available historical statistical interruption information, and establishing a Bayesian network for each distributed feeder line, wherein the method specifically comprises the following steps of:

s21, taking the random variables { D, C, E } as vertexes of the Bayesian network, drawing directed edges from parent variables to child variables from the vertexes, and constructing a structure of the Bayesian network;

S22, parameterizing a conditional probability distribution function of each fault factor in the structure of the Bayesian network according to available historical statistical interrupt information, and establishing a Bayesian network for each distribution feeder line.

It should be noted that BN provides a convenient way to represent the factorized formula as a directed acyclic graph. Thus, the random variable { D, C, E } is represented as the vertex of BN. Vertices using the fault factor as BN are connected by drawing directed edges starting from the parent vertex and ending at the child vertex. BN provides a graphical way to encode conditional independence defined by: any vertex X is conditionally independent of the non-descendant vertex Nd (X) in the graph if the value of its parent is known. This is symbolically denoted as (X ζ Nd (X) |pa (X)), nd (X) being the vertex set of BN, parent node excluding X, the absence of a directed path originating from X, a ζ B meaning that a and B are marginally independent.

In an embodiment, constructing BN entails finding the structure of the graph and parameters of the conditional probability distribution function, the constructed BN probability graph is shown in fig. 3-4, for example. The present embodiment utilizes grid topology information and causal relationships to reveal conditional independence between variables, parameterizing a conditional probability distribution function (i.e., factor) based on available statistical break information. Accordingly, the step S22: parameterizing a conditional probability distribution function of each fault factor in the structure of the Bayesian network according to available historical statistical interruption information, and establishing a Bayesian network for each distribution feeder line, wherein the method specifically comprises the following steps of:

S221, for fault factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, parameterizing the conditional probability distribution function of the fault factor, wherein the parent variable of the branch state variable isWherein D is _i-1 Is the state flow branch of the adjacent upper layer, +.>Is the fault feedback of the ith branch, +.>Representing environmental parameters of a power grid system; />Representing power grid system fault feedback information->Representing the physical parameters of the power grid system of the ith branch;

s222, for fault factorsParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and when the main branch is powered on, wherein the parent variable of the user state variable is +.>

S223, for fault factorsIts parent variable based on fault feedback by user isIn the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to exponential distribution according to the received artificial fault feedback time t of the user;

s224, for fault factorsIts parent variable based on fault feedback of ammeter isBased on the fault signal of the electricity meter when the state of the customer switch is known +. >The conditional probability distribution function of the fault factor is parameterized to be conditionally independent of the remaining variables.

In one embodiment, the step S221: for the fault factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, parameterizing the conditional probability distribution function of the fault factor, specifically comprising:

factor P (D) _i ∣Pa(D _i ) Represented at D) _i ⊥Nd(D _i )∣Pa(D _i ) The independence factor under the condition, parent of the branch state variable is selected asAs shown in fig. 3. Here, D _i-1 Is the state flow branch of the adjacent upper layer, +.>Is the fault feedback of the ith branch, wherein +.>The system data can be expressed as grid weather geographic system data, and comprise objective environmental factors such as weather geography and the like; />The fault feedback information of the power grid system can be included, and the fault information comprises a distribution box fault signal and the like; />The physical parameters representing the ith branch include conductor length and pole count, etc. Based on this parent selection scheme of branch state variables, nd (D _i ) Including all variables in the feeder line except downstream of the i-th branch. To display the four variable pairs D _i Two cases are described herein: d (D) _i-1 =1 and D _i-1 ＝0。

(1) In the first case, when the parent branch is powered down, then D _i =1, probability 1. Thus, from substation to D _i-1 All variables on the path, with { D ₁ ,…,D _i-2 Indicated at D _i-1 In the case of=1, and { D _i Conditionally independent. Because in a radial network there is only one unique path between the substation and each branch; if this path is in { D ₁ ,…,D _i-2 Any one of the arbitrary points in the sequence is interrupted, and D can be automatically obtained _i-1 The conclusion of =1, regardless of the location of the break in the path. Thus, consider the variable D _i The binary nature of (c) a conditional probability distribution function,can be expressed as:

(2) In the second case, if the adjacent upstream branch is energized, then all upstream branches of the ith branch are also energized with a probability of 1, unaffected by the power outage, { D ₁ ＝0,…,D _i-2 =0 }. In this case, D _i =1 only occurs when the branch is damaged. Thus, three background variablesAnd->Is used as causal fault feedback of the state of the ith branch to estimate the outage probability of the ith branch. Conditional probability distribution function->May be expressed as a bernoulli distribution, as follows:

wherein the failure probability of branch i is expressed asIs->And->Is a function of (a).

Further, the present embodiment uses a vulnerability model to formulate this function, the vulnerability model being a series model that can be used to make vulnerability analysis for each rod and conductor in the branch, for D _i-1 In the case of =0, branch i is estimated to give the context variableAnd->Is a failure probability of (1):

where L is the number of distribution bars used to support the ith branch, K is the number of wires between two adjacent poles of the ith branch, Φ is the standard normal probability integral, χ is the median of the vulnerability function, ζ is the logarithmic standard deviation of the intensity measurements,indicating the probability of failure of the conductor of the i-th branch.

In one embodiment, the step S222: for fault factorsParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and the main branch is powered on, specifically comprising:

factors ofA conditional probability distribution function representing the state of user j for a given parent variable. The parent of the user state variable is selected as +.>Here, D _i Is the state of the direct upstream branch of the j-th client. To display C _i ^j And D _i The contingency between them, here considered are two cases: d (D) _i =1 and D _i ＝0。

(1) In the first case, if the main branch is de-energized, the main branch, thanks to the radial structure of the feed line,the probability of (1). Using this deterministic relationship, ++>The following formula can be written:

(2) In the second case, if the primary branch is powered on, the path between the substation and the i-th branch is valid. In this case, the user fails, Pi can only be caused by overload/failure on the user side ₂ The probability of occurrence is expressed by the bernoulli distribution employed in the statistical outage information:

wherein, for the purpose of illustration of the parameter pi ₂ With uncertainty of user-defined hyper-parameter alpha ₂ And beta ₂ Defining a beta distribution:

wherein, gamma ₂ Is a normalization constant defined as gamma ₂ ＝Γ(α ₂ +β ₂ )，

In an embodiment, the step S223: for fault factorsIts parent variable based on user-artificial fault feedback is +.>In the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to the index distribution according to the received artificial fault feedback time t of a user, and specifically comprises the following steps:

factors ofRepresented by->Is independent under the condition. User-based artificial faultFeedback->Is selected as +.>Δt refers to the time elapsed after the occurrence of the failure.

More precisely, Δt represents the period of time that the load-side customer network end device needs to wait before sending a blackout report over the public network stream. This is to avoid that the client gives a false alarm due to a temporary event. It is apparent that there is a tradeoff between the number of artificial fault feedback and the latency of power outage position estimation. For example, when the observability of the actual feeder is very low, the utility company may increase ΔT to receive more artificial fault feedback for outage position estimation. During the period DeltaT, at time T ₀ After the power failure occurs, the time t for receiving the artificial fault feedback of the user is distributed according to an index:

therefore, in consideration of DeltaT,the probability of (2) can be calculated as: />

Thus, the factorThe following results were obtained:

wherein pi ₃ A small value defined by the user is represented to account for the possibility of false positives, such as illegal faulty calls and social media data processing errors.

λ ₁ The value representing the change according to the actual situation, because the fault feedback time t of the received user equipment is distributed according to the index, there are:in each practical scenario, the fault feedback time of the ue can be modeled as an exponential distribution function, thus λ here ₁ The corresponding value is selected from the simulated exponential distribution function.

In one embodiment, the step S224: for fault factorsIts parent variable based on fault feedback of the electricity meter is +.>Based on the fault signal of the electricity meter when the state of the customer switch is known +.>Parameterizing the conditional probability distribution function of the fault factor, to be conditionally independent of the remaining variables, specifically comprises:

factors ofIs at->The following independence factor. And user-based signals->In contrast, notification mechanisms based on load side device network flows are almost instantaneously delivered to utility departments.

Thus, the parent item of ammeter-based fault feedback is selected asWhen the state of the client switch is known, +.>Becomes conditionally independent of the remaining variables, encoded by the following factors:

wherein pi ₄ And pi ₅ Respectively representing the network flow communication reliability of the load side equipment and the fault probability value of the ammeter; specifically, pi ₄ Is the probability that the last wheeze can be correctly delivered to the utility for outage notification, pi ₅ It is the probability that the meter will power down due to its own failure and send the last breath signal.

Further, the values of the two parameters are determined from the historical interrupt report, and given the limited size of the historical data, the uncertainty of the two parameters is modeled using the beta distribution as follows:

wherein: alpha ₄ ，β ₄ ，α ₅ ，β ₅ Super parameters which are customized by users are set as constant 1 by default; gamma ray ₄ Is a normalization constant defined as gamma ₄ ＝Γ(α ₄ +β ₄ )，γ ₅ Is a normalization constant defined as gamma ₅ ＝Γ(α ₅ +β ₅ ) Wherein

In one embodiment, the step S30: the sampling algorithm is adopted to execute an inference task on the Bayesian network, and the fault position is inferred, and the method specifically comprises the following steps:

s31, collecting all outage evidences from the client device and branch-level evidences to construct an evidence sample set in a period delta T passing after the occurrence of the fault, wherein the outage evidences comprise load-end network flow fault data of a j-th client of a branch i and fault signals of an ammeter, and the branch-level evidences comprise power grid system fault feedback information, power grid system physical parameters and environment parameters;

S32, randomly distributing any sample in the evidence sample set to all unknown state variables { D, C }, selecting any one state variable as a sampling starting point, wherein D represents the state of a network branch, C represents a client switch,

s33, distributing the data to D according to the structure of the Bayesian network in the (tau+1) th iteration of Gibbs sampling _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line;

s34, extracting a new sample by using an inverse transformation method, and executing a local sampling process of the next non-evidence variable until all unknown variables in the Bayesian network are sampled, so as to complete one iteration of Gibbs sampling;

and S35, deducing connection states of all branches and clients based on a sample sequence generated by Gibbs sampling, and selecting a power-off branch closest to a transformer substation as a position of a power failure event.

It should be noted that after construction and parameterization of BN, the multi-source data fusion outage localization process is effectively transformed into probabilistic reasoning on the graphical model. However, even if P (D, C, E) is simplified, P (D) _i |E) and the likeThere is still a need to calculate expensive solutions simultaneously on all nodes of the graphSum operation->This is not scalable for large-scale distribution networks. To solve this problem, the reasoning task can be performed on BN using GS algorithm.

GS is an approximate reasoning method based on MCMC that can provide a good representation of the probability distribution function by using random variable instantiations without knowing the mathematical properties of all distributions. The main advantage of this approach is that it uses univariate conditional distributions for sampling, dependency on random variable spatial dimensions. Therefore, the GS is insensitive to BN size compared to common accurate reasoning methods, such as variable elimination and cliff, which suggests that the GS method is particularly advantageous for complex practical applications.

When a blackout fault occurs, the probability of outage of the branch/client can be deduced using the GS algorithm and BN structure, and to do so, first, after the DeltaT has passed, all blackout evidence from the client device is collectedIf the utility receives a network flow outage signal or a last wheeze signal from the j-th customer of branch i, corresponding evidence +.>Or->Is set to 1. Evidence of branch level Obtained from a data center of the power grid equipment department and a power grid meteorological geographic collection information system. After all evidence is collected, any initial sample is randomly assigned to all unknown state variables

Then, an arbitrary state variable is selected as the sampling start point, e.g. D _i The method comprises the steps of carrying out a first treatment on the surface of the In the (tau+1) th iteration of GS, D is allocated according to the structure of BN _i Is inserted into a local Bayesian estimator as shown below to approximate the D given the latest sample _i Is a conditional probability distribution function of (1):

wherein d _-i (τ) is the division of d _i All up-to-date samples except, including the value of the evidence variable, and:

thus, P _Φ (D _i ∣d _-i′ ^(τ) ) Can be directly calculated by the fault factor expression formula determined by the prior method, due to P _Φ (D _i ∣d _-i′ ^(τ) ) Is a probability distribution function given a single random variable for all other samples, this calculation can be done efficiently.

By P _Φ (D _i ∣d _-i′ ^(τ) ) Extracting a new sample using an inverse transform methodReplace->The algorithm then moves to the next non-evidence variable of the BN, performing the local sampling process.

When all unknown variables of the BN are sampled once, one iteration of the GS is completed, this process can propagate information throughout the BN and effectively infer the location of the outage in combination with data from different sources, and the sampling process is applied repeatedly until a sufficient number of random samples are generated for the unknown variables D, C.

In one embodiment, the step S35: deducing the connection state of all branches/clients based on a sample sequence generated by Gibbs sampling, and selecting a power-off branch closest to a transformer substation as the position of a power failure event, wherein the method specifically comprises the following steps:

It is noted that after the GS process, the most probable values for each branch and client state are solved according to the obtained approximate conditional probability distribution function. To achieve this, a threshold of 0.5 is used, e.g., P (D) _i =1|e) +.0.5 indicates branch i is powered on. After deducing the connection status of all branches and customers, the location of the outage event is obtained by selecting the outage branch closest to the substation

In general, if the iteration time is not long enough, sampling may severely mislead the target distribution, thereby reducing the accuracy of the 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 strategies can result in high computation time, thereby increasing outage time and cost. Thus, by using GS, there is a tradeoff between accuracy and computation time for the outage location. As shown in fig. 2, to find a reasonable maximum iteration number for a particular BN, a potential scaling factor R is used to diagnose the convergence of GS at different iteration numbers.

In one embodiment, in the step S30: performing an inference task on the bayesian network by adopting a sampling algorithm, and after deducing the fault location, the method further comprises the following steps:

s40, for a sample sequence generated by each iteration of a sampling algorithm, calculating the inter-sequence variation and the intra-sequence variation of the sample sequence;

s41, determining a scale-down coefficient based on the inter-sequence variation and the intra-sequence variation;

s42, based on the scale-down coefficient, diagnosing the convergence of the sampling algorithm under different iteration numbers, and determining the maximum iteration times.

In one embodiment, the step S40: for a sample sequence generated by each iteration of a sampling algorithm, calculating the inter-sequence variation and the intra-sequence variation of the sample sequence, wherein the method specifically comprises the following steps of:

s41, for each iteration process, starting from n sample sequences generated by a sampling algorithm for each unknown variable in the Bayesian network, dividing each sample sequence into two halves with the same size for supplementing an original sample sequence, and connecting all sample sequences in series to form a matrix theta with the size of 2n multiplied by m;

specifically, for each M, starting from n sample sequences generated by GS for each unknown variable in the BN. After discarding the samples generated during the warm-up period, each sequence is split into two halves of the same size, i.e. m, and used to supplement the original sequence. All sample sequences are concatenated into a matrix θ of size 2n×m.

S42, calculating the inter-sequence variation B of the sample sequence based on the matrix theta _i And intra-sequence variation V _i The formula is:

wherein:represents the mean value within the sequence,/-> Represents the overall mean value-> Representing the j-th sample sequence variance, +.>

In one embodiment, the determining the scaling factor based on the inter-sequence variation and the intra-sequence variation is formulated as:

wherein: b (B) _i Representing inter-sequence variation, V _i Intra-sequence variation is represented, and n represents the number of samples.

In theory, when 2m → infinity, R _i The value of (2) is equal to 1.R is R _i The expression > 1 indicates that the variance of any one estimate can be further reduced by more iterations. In other words, the generated sequence has not yet completed a complete investigation of the target probability distribution function; alternatively, if R _i Sequence of 1The target probability distribution function is approached. Here, according to the previous work, a threshold R is used _ψ =1.1 to select the value of M. Therefore, m+.2m is set to satisfy R _i ≤R _ψ Is applicable to any i in the BN structure.

Compared with the related technology, the power grid fault position deducing method provided by the invention has the following advantages:

(1) Bayesian networks vary in complexity: according to the invention, by introducing multi-source heterogeneous fault information, a high-dimensional joint probability distribution function is constructed, so that the accuracy and the robustness of research and judgment can be greatly improved, and in order to solve the analysis of the high-dimensional joint probability distribution function, a graph-based representation method is adopted as a statistical relationship between analysis random variables, and the high-dimensional joint probability distribution function of the system is decomposed into a group of probability factors which are easier to manage.

(2) Low-delay and real-time performance: the present invention utilizes the Gibbs Sampling (GS) algorithm to solve the latency problem, as a Markov Chain Monte Carlo (MCMC) based algorithm, where the GS can provide a complete characterization of the unknown variable distribution by rapidly generating a series of samples.

Further, as shown in fig. 5, a second embodiment of the present invention proposes a power grid fault location inference system, the system comprising:

the factorization module 10 is configured to factorize a conditional probability distribution function of the network topology Y given the multi-source feedback data to obtain a fault factor with independent conditions;

a parameterization module 20, configured to parameterize the fault factor according to available historical statistical interruption information, and establish a bayesian network for each distributed feeder line;

an inference module 30, configured to perform an inference task on the bayesian network using a sampling algorithm, and infer a fault location.

The embodiment identifies and locates the transverse fault event in the partially observable distribution system based on multi-source data fusion, adopts a probability graph modeling method for data fusion, and the high-dimensional joint probability distribution function of the system is decomposed into a group of probability factors which are easier to manage, and the factors 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 the accuracy of a fault research and judgment framework; by establishing a Bayesian network for each distributed feeder line, BN uses a graph-based representation method as a basis for analyzing statistical relationships among random variables, system topology structures from a single line diagram and data flow information from user side network equipment, and graph parameters are learned from historical power failure data according to experience; the sampling algorithm is adopted to execute the reasoning task on the Bayesian network, the power failure positioning process based on data fusion is effectively converted into online reasoning of BN, and the power network fault position can be rapidly deduced.

In one embodiment, the factorization module 10 includes:

a joint distribution unit for calculating joint distribution terms P (Y, E) of Y and E based on a conditional probability distribution function of a network topology Y given the multi-source feedback data E, wherein Y is a polynomial variable represented by the state of the network branch D and the connection of the client switch C;

the decomposition unit is used for decomposing the joint distribution item into a group of fault factors with smaller sizes based on the conditional independence among the random variables { D, C and E }, wherein each fault factor is a conditional probability distribution function formed by a child variable and a father variable, and the child variable comprises the connection state of the ith branch, the connection state of the jth customer switch of the ith branch, network flow fault data of a user load end and fault signals of the intelligent ammeter.

wherein: u= |e|, D _i Representing the connection status of the i-th branch in the feeder, Representing the connection status of the j-th customer exchange of the i-th branch,/and%>Network flow failure data representing user load side, < >>A fault signal indicative of the electricity meter.

In one embodiment, the parameterization module 20 includes:

a structure construction unit, configured to construct a structure of the bayesian network by using the random variables { D, C, E } as vertices of the bayesian network, and drawing directed edges from parent variables to child variables from the vertices;

and the parameterization unit is used for parameterizing the conditional probability distribution function of each fault factor in the structure of the Bayesian network according to the available historical statistical interrupt information, and establishing a Bayesian network for each distribution feeder line.

In an embodiment, the parameterization unit is specifically configured to perform the following steps:

for the fault factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, parameterizing the conditional probability distribution function of the fault factor, wherein the parent variable of the branch state variable isWherein D is _i-1 Is the state flow branch of the adjacent upper layer, +.>Is the fault feedback of the ith branch, +.>Representing environmental parameters of a power grid system; / >Representing power grid system fault feedback information->Representing the physical parameters of the power grid system of the ith branch;

for fault factorsParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and when the main branch is powered on, wherein the parent variable of the user state variable is +.>

wherein: p (P) _l ⁱ Indicating the probability of failure of the i-th branch.

Further, the fault factorParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and when the main branch is powered on, comprising:

Wherein: pi ₂ Representing a random value.

Further, the fault factorIts parent variable based on user-artificial fault feedback is +.>In the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to the received artificial fault feedback time t of the user and the exponential distribution, and the conditional probability distribution function is shown as follows:

wherein: pi ₃ For user-defined values lambda ₁ Representing a variation value.

In one embodiment, the reasoning module 30 specifically includes:

the evidence collection unit is used for collecting all outage evidences from the client device and evidence of a branch level to construct an evidence sample set in a period delta T after the occurrence of the fault, wherein the outage evidences comprise load end network flow fault data of a j-th client of the branch i and fault signals of an ammeter, and the evidence of the branch level comprises power grid system fault feedback information, power grid system physical parameters and environment parameters;

A sampling start point selection unit, configured to randomly allocate any sample in the evidence sample set to all unknown state variables { D, C }, and select any one state variable as a sampling start point, where D represents a state of a network branch, and C represents a client switch;

a conditional probability distribution function determining unit for assigning D in the (t+1) th iteration of Gibbs sampling according to the structure of the Bayesian network _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line;

the sample extraction unit is used for extracting a new sample by using an inverse transformation method, executing the local sampling process of the next non-evidence variable, and completing one iteration of Gibbs sampling after all unknown variables in the Bayesian network are sampled;

the position selection unit is used for deducing the connection states of all branches and clients based on a sample sequence generated by Gibbs sampling, and selecting the power-off branch closest to the transformer substation as the position of the power failure event.

In one embodiment, the conditional probability distribution function determination unit approximates the given latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line, comprising:

In an embodiment, the location selection unit is specifically configured to:

the connection state of all branches/clients is deduced based on the sample sequence generated by Gibbs sampling, and the formula is:

p (D) _i =1|e) is compared with a set threshold value to infer the connection state of the ith branch and the customer, and then selectThe power down branch closest to the substation is taken as the location of the outage event.

In one embodiment, the system further comprises a position calibration module, and the position calibration module specifically comprises:

A variation calculation unit, configured to calculate, for a sample sequence generated by each iteration of a sampling algorithm, an inter-sequence variation and an intra-sequence variation of the sample sequence;

a coefficient determination unit configured to determine a scale-down coefficient based on the inter-sequence variation and the intra-sequence variation;

and the diagnosis unit is used for diagnosing the convergence of the sampling algorithm under different iteration numbers based on the scale-down coefficient and determining the maximum iteration times.

In an embodiment, the variance calculating unit is specifically configured to:

wherein:represents the mean value within the sequence,/->Represents the overall mean value->Representing the j-th sample sequence variance.

In an embodiment, the coefficient determining unit determines a scale-down coefficient, expressed as:

According to the embodiment, the probability map modeling is adopted to realize data fusion, heterogeneous data sources can be integrated seamlessly, different data sources can be mutually supplemented, and the power outage information quantity is increased, so that the problem that the coverage rate of intelligent power grid fault feedback equipment or the network flow reporting rate of user side equipment in an actual power grid is low is solved. At the same time, by taking advantage of the conditional independence inherent between evidence and state variables in the power distribution system, the exponential computational complexity of the outage localization task is reduced to the linear complexity of the number of variables.

It should be noted that, other embodiments of the power grid fault location inference method system or the implementation method thereof according to the present invention may refer to the above-mentioned method embodiments, and are not repeated herein.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims

1. A method of power grid fault location inference, the method comprising:

2. The power grid fault location inference method as defined in claim 1, wherein the multisource data feedback set includes network flow fault data at the customer load side, fault signals of the electricity meter, grid system fault feedback information, grid system physical parameters and environmental parameters.

3. The method of power grid fault location inference as claimed in claim 1, wherein factoring the conditional probability distribution function of the network topology Y given the multi-source feedback data to obtain a conditional independent fault factor comprises:

4. A power grid fault location inference method as claimed in claim 3, wherein the decomposed representation of the joint distribution term is:

5. The method of power grid fault location inference as claimed in claim 4, wherein said parameterizing the fault factor based on available historical statistical outage information comprises, for each distributed feeder, establishing a bayesian network comprising:

6. The method of power grid fault location inference as claimed in claim 5, wherein parameterizing the conditional probability distribution function of each fault factor in the structure of the bayesian network based on available historical statistical outage information comprises establishing a bayesian network for each distribution feeder line:

for fault factorsParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and when the main branch is powered on, wherein the parent variable of the user state variable is

for fault factorsIts parent variable based on fault feedback of the electricity meter is +.>Based on the fault signal of the electricity meter when the state of the customer switch is known +. >The conditional probability distribution function of the fault factor is parameterized to be conditionally independent of the remaining variables.

7. The power grid fault location inference method as claimed in claim 6, wherein said fault-related factor P (D _i ∣Pa(D _i ) Based on the parent variable of the branch state variable being 1 when the parent branch is powered off and the adjacent upstream branch is powered on, the fault factor is conditionalThe rate distribution function is parameterized, including:

wherein: p (P) _l ⁱ Indicating the probability of failure of the i-th branch.

8. The power grid fault location inference method as claimed in claim 6, wherein the fault probability P of the i-th branch _l ⁱ One concern for modeling vulnerabilityAnd->Is expressed as:

wherein: l is the number of distribution bars for supporting the ith branch, K is the number of wires between two adjacent poles of the ith branch, and phi is the standardQuasi-normal probability integration, χ is the median of the vulnerability function, ζ is the logarithmic standard deviation of the intensity measurements, Indicating the probability of failure of the conductor of the i-th branch.

9. The power grid fault location inference method as recited in claim 6, wherein the fault factor pairParameterizing a conditional probability distribution function of the fault factor based on a parent variable of a user state variable of 1 when the main branch is powered off and when the main branch is powered on, comprising:

wherein: pi ₂ Representing a random value.

10. The power grid fault location inference method as recited in claim 6, wherein the fault factor pairIts parent variable based on user-artificial fault feedback is +.>In the period delta T passing after the fault occurs, at T ₀ After power failure occurs at moment, parameterizing a conditional probability distribution function of the fault factor according to the received artificial fault feedback time t of the user and the exponential distribution, and the conditional probability distribution function is shown as follows:

11. The power grid fault location inference method as recited in claim 6, wherein the fault factor pair Its parent variable based on fault feedback of the electricity meter is +.>Based on the fault signal of the electricity meter when the state of the customer switch is known +.>The conditional probability distribution function of the fault factor is parameterized, conditionally independent of the remaining variables, expressed as:

12. The power grid fault location inference method as claimed in claim 1, wherein said employing a sampling algorithm to perform inference tasks on said bayesian network to infer a fault location comprises:

In the (t+1) th iteration of Gibbs sampling, D will be assigned according to the structure of the Bayesian network _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line;

13. The power grid fault location inference method as claimed in claim 12, wherein said assigning to D is performed in accordance with the structure of said bayesian network in a τ+1th iteration of gibbs sampling _i Evidence samples of parent-child variables of (a) are inserted into a local Bayesian estimator, approximating given the latest sample D _i Wherein D is a conditional probability distribution function of _i Representing the connection state of the ith branch in the feeder line, comprising:

14. The power grid fault location inference method as claimed in claim 12, wherein the inferring connection status of all branches/customers based on the sample sequence generated by gibbs sampling, selecting the power down branch closest to the substation as the location of the power outage event, comprises:

15. A power grid fault location inference method as claimed in any one of claims 1 to 14, wherein after said employing a sampling algorithm to perform an inference task on said bayesian network to infer a fault location, said method further comprises:

16. The power grid fault location inference method as claimed in claim 15, wherein the calculating the inter-sequence variation and the intra-sequence variation of the sample sequence for the sample sequence generated by each iteration of the sampling algorithm comprises:

17. The power grid fault location inference method as claimed in claim 15, wherein the determining a scaling factor based on the inter-sequence variation and the intra-sequence variation is formulated as:

wherein: b (B) _i Representing inter-sequence variation, V _i Representation sequenceIntra-column variation, n, represents the number of samples.

18. A power grid fault location inference system, the system comprising: