A method and system for analyzing faults of power distribution terminal modules
By building a DBN neural network and fault identification model, analyzing the fault samples of the power distribution terminal module, the problem of inaccurate fault analysis in the existing technology is solved, more efficient fault identification and fault rate calculation is achieved, and the power distribution terminal is improved Operation and maintenance management level.
Patent Information
- Application Number
- CN202210227201.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-08
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-03-08
AI Technical Summary
The prior art lacks accuracy in the fault analysis of distribution terminals, and fails to effectively consider the operating status and fault laws, resulting in insufficient accuracy in the fault analysis.
By obtaining the fault samples of the power distribution terminal module, calculating its feature vectors, building a training data set, and using DBN neural network for training, generating a fault recognition model, randomly generating an expanded feature vector for linear fit, and determining the fault distribution function type and fault Density value, calculate the failure rate.
It greatly improves the accuracy of fault analysis of power distribution terminal modules, can more accurately identify and calculate the fault rate, and improves the reliability of operation and maintenance management.
Smart Images

Figure CN114624549B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power distribution fault analysis, and in particular to a power distribution terminal module fault analysis method and system. Background Art
[0002] As the end of the power supply system, the distribution network is a bridge connecting the power generation side and the power consumption side. Once the distribution network is out of power or the power supply is disturbed, it will not only endanger the public security and personal safety of the society, but also cause significant losses or impacts in politics or economy.
[0003] Therefore, it is the social responsibility, political responsibility and legal responsibility of power supply enterprises to ensure the safe and reliable operation of distribution networks. Improving the technical level of distribution network operation and maintenance management is an important means to improve the reliability of distribution network power supply.
[0004] At present, due to various reasons such as the bad on-site operating environment, product quality problems, construction technology, equipment aging, and primary equipment renovation and construction, the distribution terminal will inevitably produce various faults or enter an abnormal working state. In the traditional solution to the distribution terminal failure, more attention is paid to the operating status and equipment failure law of the primary equipment. For the secondary equipment with poor contact such as the distribution terminal, the method of direct replacement after the failure is adopted, and its operating status and failure law are not considered, which greatly reduces the accuracy of the distribution terminal module failure analysis. Summary of the invention
[0005] The present invention provides a method and system for analyzing faults of a power distribution terminal module, which solves the technical problem that the accuracy of fault analysis of a power distribution terminal module is greatly reduced.
[0006] In view of this, a first aspect of the present invention provides a method for analyzing a power distribution terminal module failure, comprising the following steps:
[0007] Obtain several fault samples of the power distribution terminal module, calculate the feature vector corresponding to each fault sample according to a preset formula, obtain several fault samples and their corresponding feature vectors, and construct a training data set;
[0008] Taking the feature vector as input and the corresponding fault sample as output, the training data set is substituted into the DBN neural network for training to obtain the fault recognition network model;
[0009] Randomly generate a number of extended feature vectors and substitute them into the fault identification network model, output corresponding extended fault samples, and construct an extended data set according to the extended feature vectors and their corresponding extended fault samples;
[0010] Performing linear fitting on each of the expanded feature vectors in the expanded data set and its corresponding expanded fault sample to obtain a plurality of linear regression functions and their regression coefficients;
[0011] Calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type that the power distribution terminal module obeys;
[0012] The fault density value is calculated according to the fault density function of the fault distribution function type to be obeyed, and the failure rate of the power distribution terminal module is calculated according to the fault density value through the failure rate calculation formula.
[0013] Preferably, a plurality of fault samples of the power distribution terminal module are obtained, and a feature vector corresponding to each fault sample is calculated according to a preset formula to obtain a plurality of fault samples and their corresponding feature vectors. The step of constructing a training data set specifically includes:
[0014] Obtain fault samples of the power distribution terminal module at multiple fault time points, sort the fault samples in ascending order according to the time sequence, substitute each fault time point and its fault sample into the median rank formula in order to calculate the feature vector corresponding to each fault sample, and obtain several fault samples and their corresponding feature vectors, where the median rank formula is,
[0015]
[0016] In formula 1, F(t k ) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples;
[0017] A training data set is constructed based on the mapping relationship between each fault sample and its corresponding feature vector.
[0018] Preferably, the steps of taking the feature vector as input and the corresponding fault sample as output, substituting the training data set into the DBN neural network for training, and obtaining the fault identification network model specifically include:
[0019] Constructing a DBN neural network, wherein the DBN neural network uses an RBM neural network as a backbone neural network structure, each layer of the RBM neural network consists of a visible layer and a hidden layer, and the hidden layer of the previous layer of the RBM neural network serves as the visible layer of the adjacent next layer of the RBM neural network;
[0020] The feature vector is substituted as the input into the visible layer of the DBN neural network to obtain the visible layer state vector. The corresponding weight is added to the visible layer state vector and transmitted to the hidden layer of the DBN neural network to generate the hidden layer state vector H. Each layer of the RBM neural network is trained separately and unsupervised until convergence, and the fault recognition network model is obtained through training.
[0021] Preferably, the step of performing linear fitting on each of the expanded feature vectors in the expanded data set and its corresponding expanded fault sample to obtain a plurality of linear regression functions and their regression coefficients specifically includes:
[0022] Construct the initial linear regression function as,
[0023] Y=A+BX Formula 2
[0024] In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients;
[0025] The initial linear regression function is assigned a value by a preset assignment rule to obtain a plurality of linear regression functions and their regression coefficients, wherein the preset assignment rule is:
[0026] X=ln(t), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0027] X=ln(t-μ), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0028] X=t,Y=φ -1 [F(t k )], A = -μ / σ, B = 1 / σ;
[0029] X=lgt,Y=φ -1 [F(lgt)],A=-μ 1gt / σ lgt , B=1 / σ lgt ;
[0030] Where x = F(t k ).
[0031] Preferably, the step of calculating the observed value of each linear regression function, obtaining the linear regression function with the smallest observed value, matching the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determining the fault distribution function type obeyed by the power distribution terminal module specifically includes:
[0032] The observation value of each linear regression function is calculated by the observation equation as follows:
[0033]
[0034] In Formula 3, L represents the observed value, γ i represents the straight-line distance between the i-th observation point and the corresponding linear regression function, X i , Y i represents the observation point, d represents the total number of observation points, i=1,...,d;
[0035] A linear regression function with the smallest observed value is determined according to the observed values of all linear regression functions, and the linear regression function with the smallest observed value is matched based on a preset fault distribution function type library to determine the fault distribution function type obeyed by the power distribution terminal module. The preset fault distribution function type library includes a mapping relationship between fault distribution function types and regression coefficients, and the fault distribution function types include two-parameter Weibull distribution, three-parameter Weibull distribution, normal distribution and lognormal distribution.
[0036] Preferably, the step of calculating the fault density value according to the fault density function of the fault distribution function type to be obeyed, and calculating the failure rate of the power distribution terminal module according to the fault density value by using the failure rate calculation formula specifically includes:
[0037] The fault density value is calculated according to the fault density function of the fault distribution function type. If the fault distribution function type is a two-parameter Weibull distribution, the fault density function is:
[0038]
[0039] If the fault distribution function type obeyed is the three-parameter Weibull distribution, then the fault density function is,
[0040]
[0041] If the fault distribution function type is normal distribution, then the fault density function is:
[0042]
[0043] If the fault distribution function type obeyed is lognormal distribution, then the fault density function is,
[0044]
[0045] The failure rate of the power distribution terminal module is calculated according to the failure density value by using a failure rate calculation formula, and the failure rate calculation formula is:
[0046]
[0047] In formula 8, λ(t) represents the failure rate, F(t) represents the failure density value,
[0048] In a second aspect, the present invention further provides a power distribution terminal module fault analysis system, comprising:
[0049] An acquisition module is used to acquire several fault samples of the power distribution terminal module, calculate the feature vector corresponding to each fault sample according to a preset formula, obtain several fault samples and their corresponding feature vectors, and construct a training data set;
[0050] The training module is used to use the feature vector as input and the corresponding fault sample as output, substitute the training data set into the DBN neural network for training, and obtain the fault recognition network model;
[0051] A sample generation module, used to randomly generate a number of extended feature vectors and substitute them into the fault identification network model, output corresponding extended fault samples, and construct an extended data set according to the extended feature vectors and their corresponding extended fault samples;
[0052] A fitting module, used for performing linear fitting on each of the expanded feature vectors in the expanded data set and its corresponding expanded fault samples to obtain a plurality of linear regression functions and their regression coefficients;
[0053] A fault analysis module, used to calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type obeyed by the power distribution terminal module;
[0054] The failure rate calculation module calculates a failure density value according to a failure density function of the failure distribution function type to be obeyed, and calculates the failure rate of the power distribution terminal module according to the failure density value through a failure rate calculation formula.
[0055] Preferably, the acquisition module specifically includes:
[0056] The acquisition submodule is used to obtain the fault samples of the distribution terminal module at multiple fault time points, sort the fault samples in ascending order according to the time sequence, and sequentially substitute each fault time point and its fault sample into the median rank formula to calculate the feature vector corresponding to each fault sample, and obtain a number of fault samples and their corresponding feature vectors, where the median rank formula is,
[0057]
[0058] In formula 1, F(t k) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples;
[0059] The training set construction submodule is used to construct a training data set based on the mapping relationship between each fault sample and its corresponding feature vector.
[0060] Preferably, the training module specifically includes:
[0061] A construction submodule is used to construct a DBN neural network. The DBN neural network uses an RBM neural network as the backbone neural network structure. Each layer of the RBM neural network consists of a visible layer and a hidden layer. The hidden layer of the previous RBM neural network serves as the visible layer of the adjacent next RBM neural network.
[0062] The training submodule is used to substitute the feature vector as input into the visual layer of the DBN neural network to obtain the visual layer state vector, add corresponding weights to the visual layer state vector and pass it to the hidden layer of the DBN neural network to generate the hidden layer state vector H, perform separate unsupervised training on each layer of the RBM neural network until convergence, and obtain the fault recognition network model through training.
[0063] Preferably, the fitting module specifically includes:
[0064] The function construction submodule is used to construct the initial linear regression function:
[0065] Y=A+BX Formula 2
[0066] In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients;
[0067] The assignment submodule is used to assign values to the initial linear regression function according to a preset assignment rule to obtain a plurality of linear regression functions and their regression coefficients, wherein the preset assignment rule is:
[0068] X=ln(t), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0069] X=ln(t-μ), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0070] X=t,Y=φ -1 [F(t k )], A = -μ / σ, B = 1 / σ;
[0071] X=lgt,Y=φ -1[F(lgt)],A=-μ 1gt / σ lgt , B=1 / σ lgt ;
[0072] Where x = F(t k ).
[0073] It can be seen from the above technical solutions that the present invention has the following advantages:
[0074] The present invention constructs a training data set including several fault samples and their corresponding feature vectors by observing the running state of the power distribution terminal module; trains the training data set through a DBN neural network to construct a fault identification network model; and obtains an extended data set by randomly generating several extended feature vectors and substituting them into the fault identification network model, linearly fitting the extended feature vectors in the extended data set and their corresponding extended fault samples to obtain several linear regression functions and their regression coefficients, calculates the observed value of each linear regression function, matches the linear regression function with the smallest observed value, determines the type of fault distribution function obeyed by the power distribution terminal module, calculates the fault density value through the fault density function of the fault distribution function type obeyed, and calculates the fault rate of the power distribution terminal module. The accuracy of fault analysis of the power distribution terminal module is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0075] Figure 1 A flow chart of a method for analyzing a power distribution terminal module failure provided by an embodiment of the present invention;
[0076] Figure 2 A schematic diagram of the structure of a power distribution terminal module fault analysis system provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0077] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0078] For easier understanding, see Figure 1 The present invention provides a method for analyzing a power distribution terminal module failure, comprising the following steps:
[0079] S1. Obtain several fault samples of the power distribution terminal module, calculate the feature vector corresponding to each fault sample according to a preset formula, obtain several fault samples and their corresponding feature vectors, and construct a training data set.
[0080] It should be noted that by observing the operation of the power distribution terminal module, if the data collected by the metering module is far from the mean and no data is collected; the communication signal of the communication module is weak, does not respond to the general call, repeatedly establishes links or initializes frequently; the power module PT burns out and explodes, the backup power supply starts frequently, the backup power supply voltage is abnormal, and the terminal is not online; the CPU module terminal is not online, the terminal software restarts, the switch refuses to move, etc., then the module is considered to have failed and its fault samples are collected.
[0081] S2. Using the feature vector as input and the corresponding fault sample as output, the training data set is substituted into the DBN neural network for training to obtain a fault recognition network model.
[0082] S3. Randomly generate several extended feature vectors and substitute them into the fault identification network model, output the corresponding extended fault samples, and construct an extended data set according to the extended feature vectors and their corresponding extended fault samples.
[0083] It should be noted that several extended feature vectors are randomly generated Then by random number Input the trained neural network to obtain the expanded failure samples that meet the median rank formula and the data characteristics reflected by the small sample
[0084] S4. Perform linear fitting on each expanded feature vector in the expanded data set and its corresponding expanded fault sample to obtain several linear regression functions and their regression coefficients.
[0085] S5. Calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type that the distribution terminal module obeys.
[0086] S6. Calculate a fault density value according to a fault density function of the fault distribution function type to be obeyed, and calculate the failure rate of the power distribution terminal module according to the fault density value using a failure rate calculation formula.
[0087] The present embodiment provides a method for analyzing the fault of a power distribution terminal module. By observing the running state of the power distribution terminal module, a training data set including several fault samples and their corresponding feature vectors is constructed; the training data set is trained by a DBN neural network to construct a fault identification network model; and several extended feature vectors are randomly generated and substituted into the fault identification network model to obtain an extended data set, and the extended feature vectors in the extended data set and their corresponding extended fault samples are linearly fitted to obtain several linear regression functions and their regression coefficients, and the observed value of each linear regression function is calculated, and the linear regression function with the smallest observed value is matched to determine the type of fault distribution function obeyed by the power distribution terminal module, and the fault density value is calculated by the fault density function of the type of fault distribution function obeyed, and the fault rate of the power distribution terminal module is calculated. The accuracy of the fault analysis of the power distribution terminal module is greatly improved.
[0088] The following is a detailed description of an embodiment of the method for analyzing a power distribution terminal module failure provided by the present invention.
[0089] In a specific embodiment, step S1 specifically includes:
[0090] S101, obtaining fault samples of the power distribution terminal module at multiple fault time points, sorting the fault samples in ascending order according to the time sequence, substituting each fault time point and its fault sample into the median rank formula in order to calculate the feature vector corresponding to each fault sample, and obtaining a number of fault samples and their corresponding feature vectors, wherein the median rank formula is,
[0091]
[0092] In formula 1, F(t k ) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples;
[0093] S102: construct a training data set based on the mapping relationship between each fault sample and its corresponding feature vector.
[0094] In a specific embodiment, step S2 specifically includes:
[0095] S201. Construct a DBN neural network. The DBN neural network uses the RBM neural network as the backbone neural network structure. Each layer of the RBM neural network consists of a visible layer and a hidden layer. The hidden layer of the previous RBM neural network serves as the visible layer of the adjacent next RBM neural network.
[0096] It is understandable that the DBN (Deep Belief Network) neural network is composed of a three-layer restricted Boltzmann machine (RBM) and a single-layer backpropagation (BP) neural network. RBM is the basic component of DBN. Each layer of RBM consists of a visible layer and a hidden layer. The hidden layer of the previous RBM serves as the visible layer of the next RBM.
[0097] S202, substitute the feature vector as input into the visible layer of the DBN neural network to obtain the visible layer state vector, add corresponding weights to the visible layer state vector and pass it to the hidden layer of the DBN neural network to generate the hidden layer state vector H, perform separate unsupervised training on each layer of the RBM neural network until convergence, and obtain the fault recognition network model through training.
[0098] It should be noted that by performing unsupervised training on each layer of the RBM neural network separately, it is ensured that the feature vector can be mapped to different feature subspaces and retain as much feature data as possible. In this process, the input is input to the visual layer of the RBM to obtain the vector V, and the vector V is added with the corresponding weight w to pass it to the hidden layer to generate the vector H. Since there is no connection between the hidden layers, the values of all nodes in the hidden layer can be obtained.
[0099] The PSO algorithm is used to optimize DBN hyperparameters. In the PSO algorithm, the solution to each optimization problem is a particle in the search space. All particles have a fitness, speed, and flight direction and distance determined by the optimization function. PSO initializes a group of particles, and then searches for the optimal solution based on the solution space where the current optimal particle in the particle swarm is located. The particle swarm iteratively updates itself based on two extreme values. One is the optimal solution found by the individual particle, called the individual extreme value p best ; The second is the optimal solution found by the entire group, called the global extreme value g best The PSO algorithm requires fewer parameters to be adjusted and has faster computing speed and good global search capabilities in dynamic, multi-objective optimization environments.
[0100] In a specific embodiment, step S4 specifically includes:
[0101] S401, construct the initial linear regression function as follows:
[0102] Y=A+BX Formula 2
[0103] In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients;
[0104] S402, assigning values to the initial linear regression function through a preset assignment rule to obtain a number of linear regression functions and their regression coefficients, wherein the preset assignment rule is:
[0105] X=ln(t), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0106] X=ln(t-μ), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0107] X=t,Y=φ -1 [F(t k )], A = -μ / σ, B = 1 / σ;
[0108] X=lgt,Y=φ -1 [F(lgt)],A=-μ 1gt / σ lgt , B=1 / σ lgt ;
[0109] Where x = F(t k ).
[0110] Among them, t is the fault sample after expansion, F(t k ) is the expanded fault sample and is substituted into the median rank formula to calculate the eigenvector corresponding to each fault sample to obtain the values of X and Y.
[0111] It should be noted that there are many types of fault distribution functions for distribution terminals, including two-parameter Weibull distribution, three-parameter Weibull distribution, normal distribution, and lognormal distribution. By analyzing these common types, the most appropriate type of distribution terminal equipment is finally determined. Among them, the above assignment of the initial linear regression function corresponds to the fault distribution type, and the linear regression function assignment parameter comparison table is shown in Table 1.
[0112] Table 1
[0113]
[0114] In a specific embodiment, step S5 specifically includes:
[0115] S501, calculating the observation value of each linear regression function through the observation equation is,
[0116]
[0117] In Formula 3, L represents the observed value, γi represents the straight-line distance between the i-th observation point and the corresponding linear regression function, X i , Y i represents the observation point, d represents the total number of observation points, i=1,...,d;
[0118] It can be understood that for a certain straight line, if the sum of the squares of the distances from each observation point to the straight line can be minimized, then this line is the regression line that can represent the linear relationship between X and Y.
[0119] Wherein, Formula 3 indicates that there are discrete points (X i ,Y i ), we need to draw a straight line to satisfy the distribution of these scattered points as much as possible. The linear fitting is as follows:
[0120] The data obtained after expansion {t, F(t k )} into the above table to get {X,Y}, find all the points (X i ,Y i ) to the straight line Y = AX + B, the values of A and B can determine the parameters A and B, which can be determined by the least squares method. All these points can be regarded as observation points, as follows:
[0121] Taking the partial derivative of L with respect to A and B, we get,
[0122]
[0123] According to the least squares method, we can know that when the partial derivative of the function L is 0, we can find the extreme point of the square loss function and get:
[0124]
[0125]
[0126] After determining A and B, we can further reverse the solution of μ, σ, and m. In the two-parameter Weibull distribution and the three-parameter Weibull distribution, m = B and σ = e -A , in normal distribution and lognormal distribution, σ=1 / B, μ=-A / B, while μ in three-parameter Weibull distribution can be obtained by correlation coefficient estimation method, as follows:
[0127] When μ is estimated correctly, X and Y are in a linear relationship, and the maximum correlation coefficient R is calculated between X and Y. The relationship between the estimated value of μ and the correlation coefficient R is: as μ increases, the correlation coefficient R gradually increases. When μ reaches the theoretical value, R also reaches the maximum value. When μ increases further, R gradually decreases. It can be seen that when the correlation coefficient R is the largest, μ = R.
[0128] S502. Determine the linear regression function with the smallest observation value based on the observed values of all linear regression functions, match the linear regression function with the smallest observation value based on a preset fault distribution function type library, and determine the fault distribution function type obeyed by the distribution terminal module. The preset fault distribution function type library includes a mapping relationship between fault distribution function types and regression coefficients. The fault distribution function types include two-parameter Weibull distribution, three-parameter Weibull distribution, normal distribution and lognormal distribution.
[0129] It can be understood that each fault distribution function type has a specific regression coefficient, and when the linear regression function with the smallest observed value is determined, the corresponding regression coefficient can be determined, and then the corresponding fault distribution function type can be determined through the regression coefficient through Table 1.
[0130] In a specific embodiment, step S6 specifically includes:
[0131] S601. Calculate the fault density value according to the fault density function of the fault distribution function type. If the fault distribution function type is a two-parameter Weibull distribution, the fault density function is:
[0132]
[0133] S602. If the type of fault distribution function obeyed is three-parameter Weibull distribution, then the fault density function is:
[0134]
[0135] S603. If the fault distribution function type is normal distribution, then the fault density function is:
[0136]
[0137] S604. If the fault distribution function type obeyed is lognormal distribution, then the fault density function is:
[0138]
[0139] S605. Calculate the failure rate of the power distribution terminal module according to the failure density value using a failure rate calculation formula. The failure rate calculation formula is:
[0140]
[0141] In formula 8, λ(t) represents the failure rate, F(t) represents the failure density value,
[0142] The above is a detailed description of an embodiment of a method for analyzing a fault of a power distribution terminal module provided by the present invention. The following is a detailed description of an embodiment of a system for analyzing a fault of a power distribution terminal module provided by the present invention.
[0143] For easier understanding, see Figure 2 The present invention provides a power distribution terminal module fault analysis system, comprising:
[0144] An acquisition module 100 is used to acquire a number of fault samples of the power distribution terminal module, calculate a feature vector corresponding to each fault sample according to a preset formula, obtain a number of fault samples and their corresponding feature vectors, and construct a training data set;
[0145] The training module 200 is used to use the feature vector as input and the corresponding fault sample as output, substitute the training data set into the DBN neural network for training, and obtain a fault recognition network model;
[0146] The sample generation module 300 is used to randomly generate a number of extended feature vectors and substitute them into the fault identification network model, output the corresponding extended fault samples, and construct an extended data set with the extended feature vectors and their corresponding extended fault samples;
[0147] A fitting module 400 is used to perform linear fitting on each expanded feature vector in the expanded data set and its corresponding expanded fault sample to obtain a plurality of linear regression functions and their regression coefficients;
[0148] The fault analysis module 500 is used to calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type that the power distribution terminal module obeys;
[0149] The failure rate calculation module 600 calculates a failure density value according to a failure density function of the failure distribution function type to be obeyed, and calculates the failure rate of the power distribution terminal module according to the failure density value using a failure rate calculation formula.
[0150] In a specific embodiment, the acquisition module specifically includes:
[0151] The acquisition submodule is used to obtain the fault samples of the distribution terminal module at multiple fault time points, sort the fault samples in ascending order according to the time sequence, and substitute each fault time point and its fault sample into the median rank formula in order to calculate the feature vector corresponding to each fault sample, and obtain several fault samples and their corresponding feature vectors, where the median rank formula is,
[0152]
[0153] In formula 1, F(t k ) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples;
[0154] The training set construction submodule is used to construct a training data set based on the mapping relationship between each fault sample and its corresponding feature vector.
[0155] In a specific embodiment, the training module specifically includes:
[0156] The construction submodule is used to construct a DBN neural network. The DBN neural network uses the RBM neural network as the backbone neural network structure. Each layer of the RBM neural network consists of a visible layer and a hidden layer. The hidden layer of the previous RBM neural network serves as the visible layer of the adjacent next RBM neural network.
[0157] The training submodule is used to substitute the feature vector as input into the visual layer of the DBN neural network to obtain the visual layer state vector, add corresponding weights to the visual layer state vector and pass it to the hidden layer of the DBN neural network to generate the hidden layer state vector H, perform separate unsupervised training on each layer of the RBM neural network until convergence, and obtain the fault recognition network model through training.
[0158] In a specific embodiment, the fitting module specifically includes:
[0159] The function construction submodule is used to construct the initial linear regression function:
[0160] Y=A+BX Formula 2
[0161] In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients;
[0162] The assignment submodule is used to assign values to the initial linear regression function according to a preset assignment rule to obtain several linear regression functions and their regression coefficients, wherein the preset assignment rule is:
[0163] X=ln(t), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0164] X=ln(t-μ), Y=ln{ln[1 / (1-F(t k ))]}, A = -lnσ, B = m;
[0165] X=t,Y=φ -1 [F(t k )], A = -μ / σ, B = 1 / σ;
[0166] X=lgt,Y=φ -1 [F(lgt)],A=-μ 1gt / σ lgt , B=1 / σ lgt ;
[0167] Where x = F(t k ).
[0168] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0169] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0170] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0171] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0172] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for executing all or part of the steps of the method described in each embodiment of the present invention through a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk, etc. Various media that can store program codes.
[0173] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for analyzing a power distribution terminal module fault, characterized in that: The following steps are involved: Obtain several fault samples of the power distribution terminal module, calculate the feature vector corresponding to each fault sample according to the preset formula, obtain several fault samples and their corresponding feature vectors, and construct a training data set, which specifically includes: Obtain fault samples of the power distribution terminal module at multiple fault time points, sort the fault samples in ascending order according to the time sequence, substitute each fault time point and its fault sample into the median rank formula in order to calculate the feature vector corresponding to each fault sample, and obtain several fault samples and their corresponding feature vectors, where the median rank formula is, In formula 1, F(t k ) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples; Construct a training data set based on the mapping relationship between each fault sample and its corresponding feature vector; Taking the feature vector as input and the corresponding fault sample as output, the training data set is substituted into the DBN neural network for training to obtain the fault recognition network model; Randomly generate a number of extended feature vectors and substitute them into the fault identification network model, output corresponding extended fault samples, and construct an extended data set according to the extended feature vectors and their corresponding extended fault samples; Performing linear fitting on each of the expanded feature vectors in the expanded data set and its corresponding expanded fault sample to obtain a plurality of linear regression functions and their regression coefficients; Calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type that the power distribution terminal module obeys; The fault density value is calculated according to the fault density function of the fault distribution function type to be obeyed, and the failure rate of the power distribution terminal module is calculated according to the fault density value through the failure rate calculation formula.
2. The power distribution terminal module fault analysis method according to claim 1, characterized in that: Taking the feature vector as input and the corresponding fault sample as output, the training data set is substituted into the DBN neural network for training. The steps of obtaining the fault recognition network model specifically include: Constructing a DBN neural network, wherein the DBN neural network uses an RBM neural network as a backbone neural network structure, each layer of the RBM neural network consists of a visible layer and a hidden layer, and the hidden layer of the previous layer of the RBM neural network serves as the visible layer of the adjacent next layer of the RBM neural network; The feature vector is substituted as the input into the visible layer of the DBN neural network to obtain the visible layer state vector. The corresponding weight is added to the visible layer state vector and transmitted to the hidden layer of the DBN neural network to generate the hidden layer state vector H. Each layer of the RBM neural network is trained separately and unsupervised until convergence, and the fault recognition network model is obtained through training.
3. The method for analyzing power distribution terminal module failure according to claim 1, characterized in that: The step of performing linear fitting on each of the expanded feature vectors in the expanded data set and the corresponding expanded fault samples to obtain a plurality of linear regression functions and their regression coefficients specifically includes: Construct the initial linear regression function as, Y=A+BX Formula 2 In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients; The initial linear regression function is assigned a value by a preset assignment rule to obtain a plurality of linear regression functions and their regression coefficients, wherein the preset assignment rule is: X=ln(t),Y=ln{ln[1 / (1-F(t k ))]},A=-lnσ,B=m; X=ln(t-μ),Y=ln{ln[1 / (1-F(t k ))]},A=-lnσ,B=m; X=t, Y=φ -1 [F(t k )], A=-μ / σ, B=1 / σ; X=lgt,Y=φ -1 [F(lgt)], A=-μ 1gt / s lgt ,B=1 / σ lgt ; Among them, μ represents the location parameter, σ represents the scale parameter, and m represents the shape parameter. and φ are the density function and distribution function of the standard normal distribution N(0,1), respectively, where Where x = F(t k ).
4. The method for analyzing the fault of a power distribution terminal module according to claim 3, characterized in that: The steps of calculating the observed value of each linear regression function, obtaining the linear regression function with the smallest observed value, matching the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determining the fault distribution function type obeyed by the power distribution terminal module specifically include: The observation value of each linear regression function is calculated by the observation equation as follows: In Formula 3, L represents the observed value, γ i represents the straight-line distance between the i-th observation point and the corresponding linear regression function, X i , Y i represents the observation point, d represents the total number of observation points, i=1,...,d; A linear regression function with the smallest observed value is determined according to the observed values of all linear regression functions, and the linear regression function with the smallest observed value is matched based on a preset fault distribution function type library to determine the fault distribution function type obeyed by the power distribution terminal module. The preset fault distribution function type library includes a mapping relationship between fault distribution function types and regression coefficients, and the fault distribution function types include two-parameter Weibull distribution, three-parameter Weibull distribution, normal distribution and lognormal distribution.
5. The method for analyzing power distribution terminal module failure according to claim 4, characterized in that: The steps of calculating a fault density value according to a fault density function of the fault distribution function type to be obeyed, and calculating the failure rate of the power distribution terminal module according to the fault density value by using a failure rate calculation formula specifically include: The fault density value is calculated according to the fault density function of the fault distribution function type. If the fault distribution function type is a two-parameter Weibull distribution, the fault density function is: If the fault distribution function type obeyed is the three-parameter Weibull distribution, then the fault density function is, If the fault distribution function type is normal distribution, then the fault density function is: If the fault distribution function type obeyed is lognormal distribution, then the fault density function is, The failure rate of the power distribution terminal module is calculated according to the failure density value by using a failure rate calculation formula, and the failure rate calculation formula is: In formula 8, λ(t) represents the failure rate, F(t) represents the failure density value, 6. A power distribution terminal module fault analysis system, characterized in that: include: An acquisition module is used to acquire several fault samples of the power distribution terminal module, calculate the feature vector corresponding to each fault sample according to a preset formula, obtain several fault samples and their corresponding feature vectors, and construct a training data set; The acquisition module specifically includes: The acquisition submodule is used to obtain the fault samples of the distribution terminal module at multiple fault time points, sort the fault samples in ascending order according to the time sequence, and sequentially substitute each fault time point and its fault sample into the median rank formula to calculate the feature vector corresponding to each fault sample, and obtain a number of fault samples and their corresponding feature vectors, where the median rank formula is, In formula 1, F(t k ) is the characteristic vector, t k is the kth fault time point, k is the kth sample, and n is the number of fault samples; A training set construction submodule is used to construct a training data set based on the mapping relationship between each fault sample and its corresponding feature vector; The training module is used to use the feature vector as input and the corresponding fault sample as output, substitute the training data set into the DBN neural network for training, and obtain the fault recognition network model; A sample generation module, used to randomly generate a number of extended feature vectors and substitute them into the fault identification network model, output corresponding extended fault samples, and construct an extended data set according to the extended feature vectors and their corresponding extended fault samples; A fitting module, used for performing linear fitting on each of the expanded feature vectors in the expanded data set and its corresponding expanded fault samples to obtain a plurality of linear regression functions and their regression coefficients; A fault analysis module, used to calculate the observed value of each linear regression function, obtain the linear regression function with the smallest observed value, match the linear regression function with the smallest observed value based on a preset fault distribution function type library, and determine the fault distribution function type obeyed by the power distribution terminal module; The failure rate calculation module calculates a failure density value according to a failure density function of the failure distribution function type to be obeyed, and calculates the failure rate of the power distribution terminal module according to the failure density value through a failure rate calculation formula.
7. The power distribution terminal module fault analysis system according to claim 6, characterized in that: The training module specifically includes: A construction submodule is used to construct a DBN neural network. The DBN neural network uses an RBM neural network as the backbone neural network structure. Each layer of the RBM neural network consists of a visible layer and a hidden layer. The hidden layer of the previous RBM neural network serves as the visible layer of the adjacent next RBM neural network. The training submodule is used to substitute the feature vector as input into the visual layer of the DBN neural network to obtain the visual layer state vector, add corresponding weights to the visual layer state vector and pass it to the hidden layer of the DBN neural network to generate the hidden layer state vector H, perform separate unsupervised training on each layer of the RBM neural network until convergence, and obtain the fault recognition network model through training.
8. The power distribution terminal module fault analysis system according to claim 6, characterized in that: The fitting module specifically includes: The function construction submodule is used to construct the initial linear regression function: Y=A+BX Formula 2 In Formula 2, X represents the expanded feature vector, Y represents the expanded fault sample, and A and B both represent regression coefficients; The assignment submodule is used to assign values to the initial linear regression function according to a preset assignment rule to obtain a plurality of linear regression functions and their regression coefficients, wherein the preset assignment rule is: X=ln(t),Y=ln{ln[1 / (1-F(t k ))]},A=-lnσ,B=m; X=ln(t-μ),Y=ln{ln[1 / (1-F(t k ))]},A=-lnσ,B=m; X=t, Y=φ -1 [F(t k )], A=-μ / σ, B=1 / σ; X=lgt,Y=φ -1 [F(lgt)], A=-μ 1gt / s lgt ,B=1 / σ lgt ; Among them, μ represents the location parameter, σ represents the scale parameter, and m represents the shape parameter. and φ are the density function and distribution function of the standard normal distribution N(0,1), respectively, where Where x = F(t k ).
Citation Information
Patent Citations
Deep-belief-network-based diagnosis method of distribution network
CN108089099A
Mechanical equipment fault diagnosis method and system based on neural network
CN109062177A
Power electronic circuit fault diagnosis method based on optimizing deep belief network
US20200285900A1
Method and apparatus for functional relationship approximation through nonparametric regression using R-functions
US7933850B1
