Method, device and computer equipment for cleaning status data of substation equipment

The method using the Koopman operator and self-organizing neural networks enhances the accuracy of substation equipment data cleaning and status determination by estimating and replacing dirty data, addressing the inaccuracies in existing methods.

CN114238271BActive Publication Date: 2024-12-03GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202111526441.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-14
Publication Date
2024-12-03
Estimated Expiration
2041-12-14

AI Technical Summary

Technical Problem

Existing methods for cleaning status data of substation equipment are not accurate enough due to fluctuations in data, leading to misjudgments in the operating status of substation equipment.

Method used

A method using the Koopman operator, dimension reduction matrix, and target basis function to estimate and replace dirty data in substation equipment status data, combined with self-organizing neural networks for data clustering and state transition probability analysis.

Benefits of technology

Improves the accuracy of data cleaning and determination of substation equipment operating status by effectively handling fluctuating data, providing more accurate estimated state data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114238271B_ABST
    Figure CN114238271B_ABST
Patent Text Reader

Abstract

The present application relates to a method, device, computer equipment, storage medium and computer program product for cleaning the status data of a substation equipment. The method comprises: obtaining a sequence of status data to be detected of the substation equipment; wherein each status data in the sequence of status data to be detected is data including status values ​​of multiple parameters; determining dirty data in the sequence of status data to be detected; obtaining target status data at the previous moment corresponding to the dirty data; using the target status data, Koopman operator, dimensionality reduction matrix and target basis function, calculating the estimated status data at the moment corresponding to the dirty data, and replacing the dirty data with the estimated status data. The present method can be used to obtain more accurate estimated status data of the substation equipment, thereby improving the accuracy of cleaning the status data of the substation equipment and the accuracy of determining the operating status of the substation equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of substation equipment, and in particular to a method, device, computer equipment, storage medium and computer program product for cleaning status data of substation equipment. Background Art

[0002] The safe operation of the power grid is the basis for ensuring a stable and reliable power supply. Once a large-scale power outage occurs in the power system, it will cause huge economic losses and adverse social impacts, endanger public safety, and cause serious social losses. Therefore, the stable operation of substation equipment and the correct judgment and timely handling of abnormal conditions are the prerequisites for the strong and reliable operation of the power grid; collecting the status data of substation equipment and analyzing the collected status data is a common means to determine whether the substation equipment is in an abnormal state. However, in actual operation, due to interference from factors such as sensor abnormalities, transmission equipment abnormalities, and external environmental interference, the acquired status data will be interrupted, jumped, exceeded the limit, and unreasonable values, that is, dirty data will appear. Determining the abnormality of substation equipment based on the data abnormality of dirty data will cause misjudgment of the operating status of substation equipment. Therefore, it is necessary to determine the dirty data in the status data of substation equipment and clean the status data. Data cleaning refers to the process of discovering and correcting identifiable errors in data files. In the traditional technical solution, a method based on the Pearson correlation coefficient combined with a regression model is used to interpolate the missing distribution transformer data, and a relay protection system fault diagnosis model based on a least squares vector machine-Bayesian network decision tree is used to solve the problems of missing and fuzzy data. However, such methods are essentially linear regression and can only estimate data for steadily changing data, while the status data of substation equipment fluctuates. Therefore, the existing technical methods are not accurate enough in cleaning the status data of substation equipment, and therefore the determined operating status of the substation equipment is not accurate enough.

[0003] Therefore, how to improve the accuracy of cleaning the status data of substation equipment and improve the accuracy of determining the operating status of substation equipment is a technical problem that technical personnel in this field currently need to solve. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for cleaning status data of substation equipment to improve the accuracy of cleaning status data of substation equipment and improve the accuracy of determining the operating status of substation equipment in response to the above-mentioned technical problems.

[0005] In a first aspect, the present application provides a method for cleaning status data of a substation device. The method comprises:

[0006] Acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0007] Determining dirty data in the state data sequence to be detected;

[0008] Acquire target state data at a previous moment corresponding to the dirty data;

[0009] The target state data, the Koopman operator, the dimension reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0010] In one embodiment, determining the Koopman operator, the dimension reduction matrix, and the target basis function includes:

[0011] Using sample parameters of the basis function and a deep neural network to train target parameters of the basis function, and determining the target basis function according to the target parameters;

[0012] Determine the historical state data sequence of two adjacent moments according to the historical normal state data set, and use the target basis function to perform a dimension-upgrading operation on the two historical state data sequences to obtain the historical state data sequence after dimension-upgrading;

[0013] The historical state data sequence after dimensionality increase is used to perform extended dynamic mode decomposition to obtain the Koopman operator and the dimensionality reduction matrix.

[0014] In one embodiment, the determining of dirty data in the sequence of state data to be detected includes:

[0015] Determine whether each state data in the state data sequence to be detected belongs to a preset data cluster; the data cluster is obtained by clustering historical normal state data sets;

[0016] Determine a state transition probability matrix corresponding to each of the parameters, and determine a parameter data sequence to be detected corresponding to each of the parameters according to the state data sequence to be detected;

[0017] Determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each of the parameters using each of the state transition probability matrices;

[0018] The dirty data in the state data sequence to be detected is determined according to the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each state transition probability sequence.

[0019] In one embodiment, determining the dirty data in the state data sequence to be detected according to the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each state transition probability sequence includes:

[0020] Determine a first quantity of the state data that does not belong to the data cluster in the state data sequence to be detected;

[0021] Determine a second number of the state transition probability sequences having a value of 0 in each of the state transition probability sequences and a third number of the values ​​of 0 in each of the state transition probability sequences;

[0022] If the first number is less than a first number threshold, the second number is less than a second number threshold, and the third number is greater than a third number threshold, the state data in the state data sequence to be detected that does not belong to the data cluster is determined as the dirty data.

[0023] In one embodiment, the determining of the state transition probability sequence of the to-be-detected parameter data sequence corresponding to each of the parameters by using each of the state transition probability matrices includes:

[0024] Inputting a target parameter data sequence into a self-organizing neural network model, and using the self-organizing neural network model to quantize each state value in the target parameter data sequence to obtain a state quantization value; the target parameter data sequence is a parameter data sequence to be detected corresponding to the target parameter in the parameter data sequence to be detected, and the target parameter is any one of the parameters;

[0025] According to the corresponding relationship between the state quantization value and the transition probability in the target state transfer probability matrix, the target state transfer probability sequence corresponding to the target parameter data sequence is determined; the target state transfer probability matrix is ​​a state transfer probability matrix corresponding to the target parameters of the target parameter data sequence.

[0026] In one embodiment, determining the data cluster according to the historical normal state data set clustering includes:

[0027] Calculating the data local density of each historical state data in the historical normal state data set and the distance between each historical state data;

[0028] The data cluster is determined according to the size relationship of each local density of the data and the size relationship of each distance.

[0029] In one embodiment, the determining whether each state data in the sequence of state data to be detected belongs to a preset data cluster includes:

[0030] Whether each state data belongs to the data cluster is determined according to the local density and current distance between each state data in the state data sequence to be detected and the data cluster.

[0031] In one embodiment, determining the state transition probability matrix corresponding to each of the parameters includes:

[0032] Determine, according to the historical normal state data set, a historical normal parameter data sequence corresponding to each of the parameters;

[0033] Quantifying each target historical parameter data in the target historical normal parameter data sequence by using the pre-trained self-organizing neural network model, and outputting the target historical quantization value corresponding to each target historical parameter data; the target historical normal parameter data sequence is a historical normal parameter data sequence corresponding to the target parameter, and the target parameter is any one of the parameters;

[0034] According to the probability distribution of the target historical quantized value, the probability of transfer between neurons in the self-organizing neural network model is determined, and a state transfer probability matrix corresponding to the target parameter is obtained.

[0035] In a second aspect, the present application also provides a device for cleaning status data of a substation device. The device comprises:

[0036] A first acquisition module is used to acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0037] A first determination module, used to determine dirty data in the state data sequence to be detected;

[0038] A second acquisition module, used to acquire target state data at a previous moment corresponding to the dirty data;

[0039] The second determination module is used to calculate the estimated state data at the time corresponding to the dirty data by using the target state data, Koopman operator, dimensionality reduction matrix and target basis function, and replace the dirty data with the estimated state data.

[0040] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0041] Acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0042] Determining dirty data in the state data sequence to be detected;

[0043] Acquire target state data at a previous moment corresponding to the dirty data;

[0044] The target state data, the Koopman operator, the dimension reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0045] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0046] Acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0047] Determining dirty data in the state data sequence to be detected;

[0048] Acquire target state data at a previous moment corresponding to the dirty data;

[0049] The target state data, the Koopman operator, the dimension reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0050] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0051] Acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0052] Determining dirty data in the state data sequence to be detected;

[0053] Acquire target state data at a previous moment corresponding to the dirty data;

[0054] The target state data, the Koopman operator, the dimension reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0055] The above-mentioned state data cleaning method, device, computer equipment, storage medium and computer program product of substation equipment, after determining the dirty data in the state data sequence to be detected of the substation equipment, the method obtains the target state data of the previous moment corresponding to the dirty data, and uses the target state data, Koopman operator, dimensionality reduction matrix and target basis function to calculate the estimated state data at the moment corresponding to the dirty data, and replaces the dirty data with the estimated state data to achieve data cleaning of the state data of the substation equipment; the Koopman algorithm can perform data estimation on the fluctuating state data, so that more accurate estimated state data of the substation equipment can be obtained, thereby improving the accuracy of state data cleaning of the substation equipment and improving the accuracy of determining the operating state of the substation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a flow chart of a method for cleaning status data of a substation device in one embodiment;

[0057] Figure 2 is a schematic diagram of a state transition probability matrix in one embodiment;

[0058] Figure 3a to Figure 3c A schematic diagram of a process of determining a state transition probability sequence corresponding to a parameter data sequence to be detected by using a state transition probability matrix in one embodiment;

[0059] Figure 4 A schematic diagram of a process of a method for cleaning status data of a substation device in another embodiment;

[0060] Figure 5 is a curve diagram of a data sequence of parameters to be detected corresponding to three parameters in another embodiment;

[0061] Figure 6 A curve diagram showing a result of data clustering determination of a state data sequence to be detected in another embodiment;

[0062] Figure 7a is a graph of the state transition probability sequence of the top oil temperature; Figure 7b It is a curve diagram of the state transition probability sequence of the load; Figure 7c is a graph of the state transition probability sequence of ambient temperature;

[0063] Figure 8 is another curve diagram of a sequence of state data to be detected after dirty data is corrected in this embodiment;

[0064] Fig. 9 A structural block diagram of a state data cleaning device for a substation device in one embodiment;

[0065] Fig.10FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] In one embodiment, Figure 1 As shown, a method for cleaning status data of substation equipment is provided. This embodiment takes the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0068] Step 102, obtaining a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters.

[0069] Among them, the substation is an electric power facility in the power system that transforms voltage, receives and distributes electric energy, controls the flow of electricity and adjusts voltage. It connects power grids of various voltage levels through transformers; substation equipment refers to functional equipment in the substation. State data refers to data describing the operating status of substation equipment. The state data in this embodiment is multidimensional data. Multidimensional data refers to data including state values ​​of multiple parameters, such as a three-dimensional array of state values ​​of three parameters including top oil temperature, load and ambient temperature as state data. State data sequence refers to a collection of state data of substation equipment at various time points within a period of time. The state data sequence to be detected refers to the state data sequence in the substation equipment that needs to be cleaned. The state data sequence to be detected in this embodiment can be a currently acquired data sequence or a pre-stored data sequence, which is not limited in this embodiment.

[0070] Step 104: determine the dirty data in the state data sequence to be detected.

[0071] Specifically, dirty data refers to abnormalities in the collected state data due to sensor abnormalities, transmission equipment abnormalities, external environmental interference, etc. In actual operation, dirty data in the state data sequence to be detected can be determined through data clusters and self-organizing neural network models. This embodiment does not limit the specific method of determining dirty data.

[0072] Step 106: Obtain the target state data at the previous moment corresponding to the dirty data.

[0073] Specifically, after determining the dirty data of the substation equipment, the time point corresponding to the dirty data is obtained; the time point is the moment of obtaining the state data corresponding to the dirty data. According to the time point of the dirty data, the state data of the previous moment adjacent to the time point of the dirty data is determined from the sequence of state data to be detected, and the target state data is determined.

[0074] Step 108, using the target state data, the Koopman operator, the dimension reduction matrix and the target basis function, calculate the estimated state data at the time corresponding to the dirty data, and replace the dirty data with the estimated state data.

[0075] Specifically, the Koopman operator is an infinite-dimensional linear operator that can transform state data from the original state space to another space to achieve the purpose of approximation. It is a powerful tool for nonlinear dynamic system analysis and decomposition. The target basis function refers to a function used to increase the dimension of the state data, which increases the dimension of the original state data to a high-dimensional observable function space. In this embodiment, the target state data, the Koopman operator, the dimensionality reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data. Among them, the estimated state data refers to the state data calculated at the same time as the dirty data, and the calculated estimated state data is used as the state data at the corresponding time, that is, the dirty data is replaced by the estimated state data, thereby realizing data cleaning of the state data sequence to be detected.

[0076] In the above-mentioned state data cleaning method of substation equipment, after determining the dirty data in the state data sequence to be detected of the substation equipment, the target state data of the previous moment corresponding to the dirty data is obtained, and the target state data, Koopman operator, dimensionality reduction matrix and target basis function are used to calculate the estimated state data at the moment corresponding to the dirty data, and the dirty data is replaced with the estimated state data to achieve data cleaning of the state data of the substation equipment; the Koopman algorithm can perform data estimation on the fluctuating state data, so that more accurate estimated state data of the substation equipment can be obtained, thereby improving the accuracy of the state data cleaning of the substation equipment and improving the accuracy of determining the operating state of the substation equipment.

[0077] Based on the above embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically, in this embodiment, the Koopman operator, the dimension reduction matrix and the target basis function are determined, including:

[0078] Step 1: Use the sample parameters of the basis function and the deep neural network to train the target parameters of the basis function, and determine the target basis function based on the target parameters.

[0079] In this embodiment, a BP (Back Propagation) neural network is used as an approximator of the basis function, and the form of the basis function Ψ(x) is set as:

[0080] Ψ(x)=W out h+b out ;

[0081] h = tanh(Wx+b);

[0082] Where W∈R 16×2 and W out ∈R 40×16 are the weight and output weight of the BP neural network, b∈R 16×1 and b out ∈R 40×1 are the bias term and output bias term of the BP neural network respectively; that is, the parameters of the basis function in this embodiment include the weight, output weight, bias term and output bias term of the BP neural network, that is, the parameter θ={W,b,W out ,b out The sample parameters of the basis function are obtained in advance, and a large number of sample parameters are input into the deep neural network for learning and training to obtain the target parameters of the basis function, and then the target basis function is determined according to the target parameters.

[0083] Step 2: Determine the historical state data sequence of two adjacent moments based on the historical normal state data set, and use the target basis function to perform a dimension-upgrading operation on the two historical state data sequences to obtain the historical state data sequence after dimension-upgrading.

[0084] Obtain a historical normal state data set, and use the historical normal state data set to construct historical state data sequences X and Y at two adjacent moments; where X is the historical state data sequence with the earlier time series, and Y is the historical state data sequence with the later time series:

[0085]

[0086] Wherein, p and q respectively represent the number of rows and columns when converting the historical normal state data set into the corresponding historical state data sequence. This embodiment does not limit the specific number of rows and columns, and can be set accordingly according to the number of historical state data in the historical state data sequence, that is, the value of p×q is equal to the total number of historical state data in the historical state data sequence; Indicates the historical status data with a previous time series. Representation and The corresponding time series is the historical status data later.

[0087] Then, the two historical state data sequences X and Y are respectively input into the target basis function, and the target basis function is used to perform dimension-upgrading operations on the two historical state data sequences X and Y to obtain the dimension-upgraded historical state data sequence X lift and Y lift .

[0088] Step 3: Use the historical state data sequence after dimensionality increase to perform extended dynamic mode decomposition to obtain the Koopman operator and dimensionality reduction matrix.

[0089] Specifically, using the obtained dimensionality-upgraded historical state data sequence X lift and Y lift , the Koopman operator finite-dimensional approximation is obtained by extending the dynamic mode decomposition algorithm, that is, solving the following minimization problem to obtain the Koopman operator A and the dimension reduction matrix C:

[0090]

[0091] The analytical formula for solving the minimum value is:

[0092] Among them, A∈R N×N , C∈R s×N , and N represents the dimension after dimensionality increase, s represents the number of bits of state data, that is, the number of parameters; Represents the pseudoinverse operation of a matrix.

[0093] After determining the Koopman operator, dimensionality reduction matrix, and target basis function, the process of calculating the estimated state data corresponding to the dirty data is as follows:

[0094] Assuming that the time corresponding to the dirty data is m, the previous moment corresponding to the dirty data is m-1, and the target state data at the previous moment corresponding to the dirty data is x m-1 ;

[0095] The estimated state data at the predicted time m Right now Ψ(x m )=AΨ(x m-1 ).

[0096] It can be seen that the operation process of determining the Koopman operator, the dimensionality reduction matrix and the target basis function in the manner of this embodiment is more convenient, thereby improving the convenience of calculating the estimated state data.

[0097] Based on the above embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically, in this embodiment, determining dirty data in the state data sequence to be detected includes:

[0098] Step 1: determine whether each state data in the state data sequence to be detected belongs to a preset data cluster; the data cluster is obtained by clustering the historical normal state data set.

[0099] Specifically, the historical normal state data set refers to a normal state data set obtained by pre-screening based on historical state data, that is, the historical state data in the historical normal state data set are all normal state data. Clustering operations are performed using the historical normal state data set to obtain multiple data clusters. It should be noted that this embodiment does not limit the specific type of clustering analysis method for obtaining data clusters based on clustering of historical normal state data sets, for example, it can be a k-means clustering method, a CLARA (Clustering LARge Application) clustering method, a grid-based clustering method, etc. After obtaining the state data sequence to be detected, each state data in the state data sequence to be detected is judged separately to determine whether each state data belongs to a preset data cluster.

[0100] Step 2: determine the state transition probability matrix corresponding to each parameter, and determine the parameter data sequence to be detected corresponding to each parameter based on the state data sequence to be detected.

[0101] Specifically, for each parameter in the state data, a corresponding state transition probability matrix is ​​set, that is, the state transition probability matrix corresponds to the parameter one-to-one. Specifically, the state data of the state data sequence to be detected includes a time series of state values ​​of multiple parameters, so it is necessary to convert the state data sequence to be detected into a parameter data sequence to be detected corresponding to each parameter, and the parameter data sequence to be detected corresponds to each parameter one-to-one. For example, assuming that each state data in the state data sequence to be detected is a state value corresponding to three parameters: top oil temperature, load and ambient temperature, the parameter data sequences to be detected are determined to be a parameter data sequence to be detected corresponding to the top oil temperature, a parameter data sequence to be detected corresponding to the load, and a parameter data sequence to be detected corresponding to the ambient temperature.

[0102] Step three, using each state transition probability matrix to determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each parameter.

[0103] For each parameter data sequence to be detected, the corresponding parameter is determined, and then the state transition probability matrix corresponding to the parameter is determined. The determined state transition probability matrix is ​​used to determine the transition probabilities corresponding to each state value in the parameter data sequence to be detected, thereby obtaining the state transition probability sequence corresponding to the parameter data sequence to be detected. According to the above method, the state transition probability sequence corresponding to each parameter data sequence to be detected is determined.

[0104] Step 4: Determine the dirty data in the state data sequence to be detected based on the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each state transition probability sequence.

[0105] Among them, not belonging to the data cluster means that the state data in the state data sequence to be detected and the historical state data in the pre-determined data cluster do not belong to the same data cluster. The 0 value in the state transition probability sequence refers to the situation where the probability value in the state transition probability sequence is 0. Specifically, determine the state data that does not belong to the data cluster in the state data sequence to be detected, and judge whether there is a 0 value in each state transition probability sequence. If there is a 0 value in the state transition probability sequence, further determine the number of 0 values ​​in the state transition probability sequence with 0 values, and then determine whether there is dirty data in the state data sequence to be detected, and determine the dirty data based on the state data that does not belong to the data cluster in the state data sequence to be detected.

[0106] According to the method of this embodiment, dirty data in the state data sequence to be detected of the substation equipment is determined, and the determination method is convenient and accurate.

[0107] In one embodiment, determining dirty data in the state data sequence to be detected according to state data that does not belong to the data cluster in the state data sequence to be detected and a 0 value in each state transition probability sequence includes:

[0108] Determine a first quantity of state data that does not belong to the data cluster in the state data sequence to be detected;

[0109] Determine a second number of state transition probability sequences having a value of 0 in each state transition probability sequence and a third number of 0 values ​​in each state transition probability sequence;

[0110] If the first number is less than the first number threshold, the second number is less than the second number threshold, and the third number is greater than the third number threshold, the state data that does not belong to the data cluster in the state data sequence to be detected is determined as dirty data.

[0111] Specifically, the first cumulative count of the counter is initialized to 0, and whether each state data belongs to a preset data cluster is determined in turn according to the arrangement order of each state data in the state data sequence to be detected; if the state data does not belong to any data cluster, the first cumulative count of the counter is increased by 1; if the state data belongs to any data cluster, the first cumulative count of the counter remains unchanged; after judging each state data in the state data sequence to be detected, the first cumulative count of the current counter is determined as the first quantity.

[0112] Specifically, the second cumulative count of the counter is initialized to 0, and it is determined in turn whether there is a 0 value in each state transition sequence; if it is determined that there is a 0 value in a state transition sequence, the second cumulative count is increased by 1, and the next state transition sequence is judged; if it is determined that there is no 0 value in a state transition sequence after the probability values ​​in the state transition sequence are judged respectively, the second cumulative count remains unchanged, and the next state transition sequence is continuously judged, until all state transition sequences are judged, and the second cumulative count is determined to be the second number.

[0113] Specifically, for each state transition sequence, the third cumulative count of the counter is initialized to 0, and each state value is judged to be 0 in turn according to the arrangement order of each state value in the state transition sequence; if the probability value is determined to be 0, the third cumulative count of the counter is increased by 1; if the probability value is determined to be not 0, the second cumulative count of the counter remains unchanged; after judging each probability value in the state transition sequence, the third cumulative count is determined to be the third number.

[0114] Specifically, the first quantity threshold, the second quantity threshold and the third quantity threshold are all set according to actual needs, and this embodiment does not limit this; among them, the second quantity threshold is generally set to 2. That is to say, in this embodiment, if the first quantity of state data that does not belong to the data cluster in the state data sequence to be detected is less than the first quantity threshold, there are only 0 values ​​in the state transition sequence of a single parameter, and the third quantity of 0 values ​​in the state transition sequence is greater than the third quantity threshold, the state data that does not belong to the data cluster in the state data sequence to be detected is determined as dirty data.

[0115] It can be seen that this embodiment can quickly and accurately determine the dirty data in the state sequence to be detected.

[0116] In one embodiment, the method for cleaning status data of a substation device further includes:

[0117] If the first number is greater than a first number threshold, the second number is greater than a second number threshold, and the third number is greater than a fourth number threshold, the state data that does not belong to the data cluster in the state data sequence to be detected is determined as abnormal state data.

[0118] Among them, abnormal state data refers to state data abnormality caused by the operation failure of the substation equipment itself, that is, the abnormal state data characterizes that the substation equipment is in an abnormal operation state. In this embodiment, if the first number of state data that does not belong to the data cluster in the state data sequence to be detected is greater than the first number threshold, there is a situation where the probability value is 0 in the state transition sequence of multiple parameters, and the third number of probability values ​​​​of 0 in the state transition sequence is greater than the fourth number threshold, it means that there is abnormal state data in the state data sequence to be detected, and the state data that does not belong to the data cluster in the state data sequence to be detected is determined as abnormal state data.

[0119] In addition, in other embodiments, the fourth number of continuous state data that do not belong to the data cluster in the state data sequence to be detected, and the fifth number of probability values ​​that are continuously 0 in the state transition probability sequence can be further determined; if the second number is greater than the second number threshold, the fourth number is greater than the fifth number threshold, and the fifth number is greater than the sixth number threshold, then the state data that does not belong to the data cluster in the state data sequence to be detected is determined as abnormal state data. That is to say, in other embodiments, when there is a continuous section of state data that does not belong to the data cluster in the state data sequence to be detected, and there is a continuous section of 0 values ​​in the state transition probability sequence of multiple parameters, the state data that does not belong to the data cluster in the state data sequence to be detected is determined as abnormal state data.

[0120] It can be seen that according to the method of this embodiment, not only the dirty data in the state data sequence to be detected can be determined, but also the abnormal state data in the state data sequence to be detected can be determined, and the detection of each state data in the state data sequence to be detected is more comprehensive.

[0121] On the basis of the above-mentioned embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically, in this embodiment, each state transition probability matrix is ​​used to determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each parameter, including:

[0122] Step 1: Input the target parameter data sequence into the self-organizing neural network model, and use the self-organizing neural network model to quantize each state value in the target parameter data sequence to obtain a state quantization value; the target parameter data sequence is a parameter data sequence to be detected that corresponds to the target parameter in the parameter data sequence to be detected, and the target parameter is any parameter.

[0123] In this embodiment, a target parameter data sequence is first determined from a plurality of parameter data sequences to be detected, the target parameter data sequence being any one of the plurality of parameter data sequences to be detected, and the target parameter data sequence includes the state value of the target parameter. Then the target parameter data sequence is input into a pre-trained self-organizing neural network (SOM, Self-organizing feature Map) model, and the state value of the target parameter in the target parameter data sequence is quantized using the self-organizing neural network model to obtain the state quantization values ​​corresponding to each state value.

[0124] Step 2: Determine the target state transfer probability sequence corresponding to the target parameter data sequence based on the correspondence between the state quantization value and the transfer probability in the target state transfer probability matrix; the target state transfer probability matrix is ​​the state transfer probability matrix corresponding to the target parameters of the target parameter data sequence.

[0125] Among them, the state transition probability matrix represents the transition probability between neurons in the trained self-organizing neural network model; the target state transition probability matrix is ​​the state transition probability matrix corresponding to the target parameters of the target parameter data sequence; the state transition probability sequence refers to the sequence corresponding to the transition probability corresponding to the parameter data sequence to be detected.

[0126] Specifically, according to the target parameters of the target parameter data sequence, the target state transition probability matrix corresponding to the target parameter data sequence is determined; the target state transition probability sequence corresponding to the target parameter data sequence is determined by utilizing the correspondence between the state quantization value and the transition probability in the target state transition probability matrix; in accordance with the above method, the state transition probability sequences corresponding to each parameter data sequence to be detected are determined.

[0127] Figure 2 A schematic diagram of a state transition probability matrix provided in this embodiment, wherein the self-organizing neural network model in this embodiment has a total of 10 neurons, and Q(i,j)(i,j∈{1,2,…,10}) in the state transition probability matrix represents the transition probability from neuron i to neuron j; Figure 3a to Figure 3c A schematic diagram of a process of using a state transition probability matrix to determine a state transition probability sequence corresponding to a parameter data sequence to be detected is provided in this embodiment. Specifically, assuming that the parameter data sequence to be detected is a section of top oil temperature data, and the data curve of the top oil temperature data is as follows: Figure 3a As shown in FIG. 1 ; the obtained top oil temperature data is input into the trained self-organizing neural network model to obtain the state quantization value (quantization value Q) corresponding to each top oil temperature data, and then the quantization value sequence corresponding to the top oil temperature data is obtained. The quantization value curve corresponding to the quantization value sequence of the top oil temperature data is shown in FIG. Figure 3bAs shown; in the quantized value sequence, t n The state quantization value at the moment is Q i , t n+1 The state quantization value at the moment is Q j , then according to the state transfer relationship, the state quantization value is Q i Transfer to Q j The state transition probability is Figure 2 The transition probability corresponding to Q(i,j) in the state transition probability matrix shown in FIG. 1 is determined in sequence, and the state transition probability corresponding to each adjacent state quantization value in the quantization value sequence is obtained to obtain the state transition probability sequence; that is, the dynamic change of the top oil temperature data over time is represented by the state transition probability sequence, such as Figure 3c shown.

[0128] It can be seen that the state transition probability sequence corresponding to each parameter data sequence to be detected is determined in accordance with the method of this embodiment, and the operation is convenient and easy.

[0129] In one embodiment, determining a state transition probability matrix corresponding to each parameter includes:

[0130] Determine the historical normal parameter data sequence corresponding to each parameter according to the historical normal state data set;

[0131] Quantify each target historical parameter data in the target historical normal parameter data sequence by using the pre-trained self-organizing neural network model, and output the target historical quantization value corresponding to each target historical parameter data; the target historical normal parameter data sequence is the historical normal parameter data sequence corresponding to the target parameter, and the target parameter is any parameter;

[0132] According to the probability distribution of the target historical quantized value, the probability of transfer between neurons in the self-organizing neural network model is determined, and the state transfer probability matrix corresponding to the target parameter is obtained.

[0133] Among them, by converting the historical normal state data set, the historical normal parameter data sequence corresponding to each parameter is determined.

[0134] Among them, the self-organizing neural network is a kind of unsupervised learning neural network; the process of training the self-organizing neural network model includes:

[0135] ①Process the historical normal parameter data sequence into a zero-mean stationary sequence {x 1 ,x 2 ...x t ,...x T}, through the first-order AR(1) model for the zero-mean stationary series {x 1 ,x 2 ...x t,...x T}Fitting: Where φ is the regression coefficient, Represents historical parameter data x t For historical parameter data x t-1 The degree of dependence; t is white noise with mean 0 and unknown variance Normal distribution, T represents the amount of data in the historical normal parameter data series; the parameters of the AR(1) model are Through matrix estimation, we get:

[0136]

[0137] Among them, γ 0 is the variance of the zero-mean stationary series, γ 1 is the first-order autocovariance function;

[0138] ② The zero-mean stationary series {x 1 ,x 2 ...x t ,...x T} is input into a self-organizing neural network containing N neurons, and the weight vector W corresponding to all neurons in the competition layer t Perform similarity comparison and judge the most similar weight vector as the winning neuron; the output of the winning neuron j* is set to 1, and the outputs of other neurons are 0; the winning neuron can adjust its weight vector W j , the adjusted weight vector is: Here, α is the learning rate.

[0139] ③ After the winning neuron adjusts the weight vector, the zero-mean stationary sequence {x 1 ,x 2 ...x t ,...x T}The weight vector W corresponding to all neurons in the competitive layer t Perform similarity comparison, that is, perform iterative calculation, and when the iteration is terminated, it means that the self-organizing neural network model is trained; the termination condition of the iteration is that the learning rate decays to 0 or a very small preset value.

[0140] The state transition probability matrix represents the transition probability between neurons in the self-organizing neural network model. Specifically, when the self-organizing neural network model is trained, for each target parameter, the zero-mean stationary sequence {x 1 ,x 2 ...x t ,...x T} is input into the pre-trained self-organizing neural network model, and each target historical parameter data x in the target historical normal parameter data sequence is converted into t Transformed into a time series quantity Q in linear space t , each weight vector W t Represents the historical parameter data of the target closest to it; that is, W t As the target historical parameter data x t The target historical quantitative value of is expressed as

[0141] According to x t and W t The similarity of can be concluded that at time t, q t The value is Q I The probability is:

[0142] P[q t =Q I ]=P[i(x t )=I]=P[x t ∈(a,b)] (1)

[0143] Where a is the target historical quantitative value Q I and Q I+1 The average value between I +Q I+1 ) / 2, b is the target historical quantitative value Q I-1 and Q I The average value between I +Q I-1 ) / 2; and 1 <I<N。

[0144] Therefore, P[i(x t )=I]=P[b <x t <a]=F x (a)-F x (b), where F x is the variable x t The probability distribution function of .

[0145] Due to x t is a normal distribution, then the probability of formula (1) can be expressed as The standard normal distribution function of is:

[0146]

[0147] Under the same conditions, q t The second-order probability distribution of is:

[0148] P[q t =Q I1 |qt+k =Q I2 ]=P[x t ∈(b 1 ,a 1 ),x t+k ∈(b 2 ,a 2 )] (3)

[0149] Among them, Q I1 ,Q I2 ∈{W 1 ,...,W 10}, I 1 =(a 1 ,b 1 ), I 2 =(a 2 ,b 2 ).

[0150] x t The second-order probability distribution is:

[0151]

[0152] in, is the autocorrelation coefficient of the first-order AR(1) model;

[0153] According to equation (1) and equation (2), equation (3) can be simplified as:

[0154]

[0155] According to the above formula (4), the state transition probability matrix between self-organizing neurons is calculated, that is, the state transition probability matrix corresponding to the target parameter is determined; the above operation is performed on each parameter respectively to determine the state transition probability matrix corresponding to each parameter.

[0156] It can be seen that this embodiment uses the self-organizing neural network model and the historical normal parameter data sequence to determine the state transition probability matrix corresponding to each parameter, and derives the law of change of each state value of the substation equipment over time, so as to use the state transition probability matrix to determine the state transition probability sequence corresponding to the parameter data sequence to be detected, and determine the dynamic change of the state value over time, thereby improving the accuracy of determining the dirty data of the substation equipment, and the determination method is more convenient.

[0157] Based on the above embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically, in this embodiment, data clusters are determined based on historical normal state data sets, including:

[0158] Calculate the data local density of each historical state data and the distance between each historical state data in the historical normal state data set;

[0159] The data clusters are determined according to the size relationship of the local density of each data and the size relationship of each distance.

[0160] In this embodiment, the historical normal state data set is clustered by the CFSFDP algorithm (Clustering by fast search and find of density peaks) to determine the corresponding data clusters. The CFSFDP algorithm determines the clustering center and the number of data clusters by calculating the local density and distance of the data.

[0161] Specifically, obtain the historical normal state data set U = {x i}={x 1 ,x 2 ...x i ,...x T}, i = 1, 2, ... T; for any historical state data x in the historical normal state data set U i , whose local density ρ i for:

[0162] in, x=d ij -d c ;d ij For other historical status data x j With historical status data x i The Euclidean distance between c is the distance threshold, usually selected as d ij Arrange the distance values ​​at the 1% to 2% position in ascending order; use this calculation method to obtain the historical status data x i The local density of data ρ i , which is usually equal to the historical state data x i The surrounding Euclidean distance is less than d c The number of historical status data.

[0163] Among them, the distance δ between historical state data i is calculated by calculating the historical state data x i Compared with historical status data x i Local density ρ i Other historical status data x j The distance δ i It can be calculated as follows:

[0164]

[0165] Generally speaking, the distance δ between historical state data i is the historical state data x i Compared with the historical status data x i Local density ρ i The minimum value of the distance between the other historical state data is high, but for the local density ρ i The highest point, the distance δ between historical state data i is other historical state data and the highest historical state data x i The maximum value of the distance.

[0166] After calculating the local density of each historical state data and the distance between each historical state data, the data clusters are determined according to the size relationship of each local density of each data and the size relationship of each distance. Specifically, for all historical state data, there are four types of data points: one is δ i Large and ρ i Large, this type of historical state data is determined as the cluster center of the data cluster; one is δ i Big ρ i Small, this type of historical state data is determined as an outlier; one is δ i Small i This type of historical status data is the data point close to the cluster center; the remaining data points are the data points around the cluster center, and these data points are divided into corresponding data clusters.

[0167] Correspondingly, after determining the data cluster, determining whether each state data in the state data sequence to be detected belongs to the preset data cluster includes:

[0168] According to the local density and current distance between each state data in the state data sequence to be detected and the current data of the data cluster, it is determined whether each state data belongs to the data cluster.

[0169] Specifically, after determining the data clusters based on the historical normal state data set, the critical density and distance threshold of each data cluster are obtained. For each state data in the state data sequence to be detected, the current data local density and current distance corresponding to the state data and each data cluster are calculated. If the current data local density is less than the critical density and the current distance is greater than the distance threshold, it means that the state data does not belong to any predetermined data cluster.

[0170] It can be seen that judging whether the status data to be detected belongs to a preset data cluster in accordance with the method of this embodiment can reduce the amount of calculation, thereby improving the detection efficiency of abnormal status data of substation equipment.

[0171] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application are described in detail below in combination with actual application scenarios. Figure 4 A process diagram of a state data cleaning method for a substation device is shown; the present application embodiment takes the substation device as a 220kV main transformer as an example for explanation, each historical state data in the historical normal state data set is a three-dimensional array including three parameters: the top oil temperature of the transformer, the load, and the ambient temperature, and the state data in the state data sequence to be detected is also a three-dimensional array including three parameters: the top oil temperature of the main transformer, the load, and the ambient temperature. The specific steps of the state data cleaning method for the substation device in this embodiment are as follows:

[0172] The historical normal state data set is clustered by the CFSFDP algorithm to determine four data clusters; the historical normal parameter data sequences corresponding to each parameter are determined based on the historical normal state data set;

[0173] The historical normal parameter data sequences are calculated respectively by the first-order AR(1) model and the self-organizing neural network model to obtain the state transition probability matrix corresponding to each parameter. The number of neurons in the self-organizing neural network model is 10 and the training cycle is 100 times.

[0174] According to the above steps, the state transition probability matrix corresponding to the three parameters (top oil temperature, load, and ambient temperature) is established, and four data clusters are generated by clustering.

[0175] A sequence of state data to be detected is obtained, which includes 500 groups of state data to be detected, wherein the sampling period of top oil temperature, load, and ambient temperature is 1 min / group; Figure 5 As shown, it is a curve diagram of the parameter data sequence to be detected corresponding to the three parameters of top oil temperature, load and ambient temperature in this embodiment.

[0176] The CFSFDP algorithm is used to perform clustering calculation on each state data in the state data sequence to be detected, and it is determined whether the state data at each time point in the state data sequence to be detected belongs to one of the four data clusters, and the judgment result is output, 1 represents belonging, and 0 represents not belonging;

[0177] Determine the parameter data sequences to be detected corresponding to each parameter according to the state data sequence to be detected, quantify each state value in the parameter data sequence to be detected by using the self-organizing neural network model, and calculate the state transition probability sequences corresponding to each parameter data sequence to be detected according to the state transition probability matrix;

[0178] Comprehensively judge and determine the normal data, dirty data and abnormal data in the state data sequence to be detected:

[0179] Determine a first number of state data that does not belong to the four predetermined data clusters in the state data sequence to be detected, determine a second number of state transition probability sequences having a value of 0 in each state transition probability sequence, and determine a third number of 0 values ​​in each state transition probability sequence;

[0180] When there is no state data that does not belong to the four predetermined data clusters in the state data sequence to be detected, and there is no 0 value in the state transition probability sequence of each parameter, it means that the state data in the state transition probability sequence are all normal data;

[0181] If the first quantity is not 0 and is less than the first quantity threshold, that is, there are a small number of outliers in the state data sequence to be detected, and the second quantity and the third quantity are both 0, it means that the state value change of a certain parameter causes the corresponding state data to not belong to the data cluster of the historical normal state data set, but the state value change is not large, so the state transition probability sequence is normal, so the outliers in the state data sequence to be detected are determined to be normal data that are misjudged, so the state data of the outliers are added to the historical normal state data set to expand the historical normal state data set;

[0182] If the first number is not 0 and is less than the first number threshold, that is, there are a small number of outliers in the state data sequence to be detected, there is a 0 value in the state transition probability sequence of a single parameter, and the third number of 0 values ​​in the state transition probability sequence is greater than the third number threshold, it is determined that the outliers are caused by sensor abnormalities, transmission line abnormalities, external environmental interference, etc., which lead to unreasonable numerical problems such as jumps, missing values, and over-values ​​in the state data. The state data (outliers) that do not belong to the data cluster in the state data sequence to be detected are marked as dirty data.

[0183] If the fourth number of continuous state data that do not belong to the data cluster in the state data sequence to be detected is greater than the fifth number threshold, and the fifth number of probability values ​​that are continuously 0 in the state transition probability sequence of multiple parameters is greater than the sixth number threshold, the state data (outliers) that do not belong to the data cluster in the state data sequence to be detected are determined as abnormal state data, and it is determined that the main transformer is in an abnormal operating state;

[0184] By judging whether each state data in the state data sequence to be detected belongs to a predetermined data cluster, a judgment result curve of the state data sequence to be detected is obtained as follows: Figure 6 As shown; the curve corresponding to the state transition probability sequence of each parameter is as follows Figures 7a to 7c As shown; among them, Figure 7a It is the curve graph corresponding to the state transition probability sequence of the top oil temperature; Figure 7bIt is the curve graph corresponding to the state transition probability sequence of the load; Figure 7c It is the curve graph corresponding to the state transition probability sequence of ambient temperature.

[0185] from Figure 7a and Figure 7b It can be seen that at the time point T=400min, the state transition probabilities of the top oil temperature and load are both 0, indicating that the transfer probability sequence of the top oil temperature and load at this moment is not stable, and most of the state data between T=400 and 500min do not belong to any data cluster. Therefore, it can be determined that the main transformer has been in an abnormal operating state since T=400min, and the top oil temperature and load have increased rapidly relative to the normal values.

[0186] from Figure 7c It can be seen that at the time point T=100min, the transition probability sequence of the ambient temperature has a value of 0, and the state data at this time point does not belong to any data cluster, and there is no such situation around the time point T=100min. Therefore, it can be determined that the ambient temperature at this time point is dirty data.

[0187] For the detected dirty data, obtain the target state data of the previous moment corresponding to the dirty data, according to Predict the estimated state data at time m, where x m-1 is the target state data of the previous moment corresponding to the dirty data, A is the Koopman operator, C is the dimension reduction matrix, and Ψ(x) is the target basis function. The dirty data is replaced by the estimated state data, and the state data sequence to be detected is updated to obtain the correct state data sequence of the main transformer. Figure 8 As shown in FIG. 1 , it is a curve diagram corresponding to the state data sequence to be detected after the dirty data is corrected in this embodiment, wherein: Figure 8 The data corresponding to the “·” is the calculated estimated state data; the estimated state data is added to the historical normal state data set to expand the historical normal state data set and provide data support for subsequent data analysis and mining.

[0188] For the detected abnormal state data, that is, determining that the main transformer is in an abnormal operating state, according to the state values ​​with continuous probability values ​​of 0 in the state transition probability sequence, the occurrence time of the abnormal operating state of the main transformer is determined and an alarm is issued.

[0189] According to the method of this embodiment, dirty data and abnormal status data in the substation equipment can be quickly and accurately determined, and the dirty data can be corrected to obtain a normal status data sequence, and an alarm can be issued for abnormal status data to prompt staff to pay attention to the operating status of the substation equipment.

[0190] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0191] Based on the same inventive concept, the embodiment of the present application also provides a state data cleaning device for substation equipment for implementing the state data cleaning method for substation equipment involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the state data cleaning device for one or more substation equipment provided below can refer to the limitations of the state data cleaning method for substation equipment above, and will not be repeated here.

[0192] In one embodiment, Fig. 9 As shown, a state data cleaning device for a substation device is provided, comprising a first acquisition module 902, a first determination module 904, a second acquisition module 906 and a second determination module 908, wherein:

[0193] The first acquisition module 902 is used to acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is data including state values ​​of multiple parameters;

[0194] A first determination module 904, configured to determine dirty data in the state data sequence to be detected;

[0195] The second acquisition module 906 is used to acquire the target state data at the previous moment corresponding to the dirty data;

[0196] The second determination module 908 is used to calculate the estimated state data at the time corresponding to the dirty data by using the target state data, the Koopman operator, the dimension reduction matrix and the target basis function, and replace the dirty data with the estimated state data.

[0197] A state data cleaning device for a substation device provided in an embodiment of the present invention has the same beneficial effects as the state data cleaning method for a substation device described above.

[0198] In one embodiment, the second determining module includes:

[0199] A first determination submodule is used to train target parameters of the basis function using sample parameters of the basis function and a deep neural network, and determine a target basis function according to the target parameters;

[0200] A dimension-raising operation submodule is used to determine the historical state data sequence of two adjacent moments according to the historical normal state data set, and perform a dimension-raising operation on the two historical state data sequences using the target basis function to obtain the historical state data sequence after dimension-raising;

[0201] The computing submodule is used to perform extended dynamic mode decomposition using the historical state data sequence after dimensionality increase to obtain the Koopman operator and dimensionality reduction matrix.

[0202] In one embodiment, the first determining module includes:

[0203] The second determination submodule is used to determine whether each state data in the state data sequence to be detected belongs to a preset data cluster; the data cluster is obtained by clustering the historical normal state data set;

[0204] The third determination submodule is used to determine the state transition probability matrix corresponding to each parameter, and determine the parameter data sequence to be detected corresponding to each parameter according to the state data sequence to be detected;

[0205] A fourth determination submodule is used to determine the state transition probability sequence of the to-be-detected parameter data sequence corresponding to each parameter using each state transition probability matrix;

[0206] The fifth determination submodule is used to determine the dirty data in the state data sequence to be detected according to the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each state transition probability sequence.

[0207] In one embodiment, the fifth determining submodule includes:

[0208] A first quantity determining unit, used to determine a first quantity of state data that does not belong to a data cluster in the state data sequence to be detected;

[0209] A second number determining unit, used to determine a second number of state transition probability sequences having a value of 0 in each state transition probability sequence and a third number of 0 values ​​in each state transition probability sequence;

[0210] A dirty data determination unit is used to determine the state data that does not belong to the data cluster in the state data sequence to be detected as dirty data if the first number is less than a first number threshold, the second number is less than a second number threshold, and the third number is greater than a third number threshold.

[0211] In one embodiment, the fourth determining submodule includes:

[0212] A first quantization unit is used to input the target parameter data sequence into the self-organizing neural network model, and use the self-organizing neural network model to quantize each state value in the target parameter data sequence to obtain a state quantization value; the target parameter data sequence is a parameter data sequence to be detected corresponding to the target parameter in the parameter data sequence to be detected, and the target parameter is any parameter;

[0213] A determination unit is used to determine a target state transfer probability sequence corresponding to a target parameter data sequence according to a correspondence between a state quantization value and a transfer probability in a target state transfer probability matrix; the target state transfer probability matrix is ​​a state transfer probability matrix corresponding to a target parameter of the target parameter data sequence.

[0214] In one embodiment, the second determining submodule includes:

[0215] A calculation unit, used for calculating the data local density of each historical state data and the distance between each historical state data in the historical normal state data set;

[0216] The clustering unit is used to determine the data clusters according to the size relationship of the local density of each data and the size relationship of each distance.

[0217] In one embodiment, the second determining submodule includes:

[0218] The judgment subunit is used to determine whether each state data belongs to the data cluster according to the local density and current distance between each state data in the state data sequence to be detected and the current data of the data cluster.

[0219] In one embodiment, the third determining submodule includes:

[0220] A sequence determination unit, used to determine the historical normal parameter data sequence corresponding to each parameter according to the historical normal state data set;

[0221] The second quantization unit is used to quantize each target historical parameter data in the target historical normal parameter data sequence by using the pre-trained self-organizing neural network model, and output the target historical quantization value corresponding to each target historical parameter data; the target historical normal parameter data sequence is the historical normal parameter data sequence corresponding to the target parameter, and the target parameter is any parameter;

[0222] The matrix determination unit is used to determine the probability of transfer between neurons in the self-organizing neural network model according to the probability distribution of the target historical quantization value, and obtain the state transfer probability matrix corresponding to the target parameter.

[0223] Each module in the above-mentioned state data cleaning device of substation equipment can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0224] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for cleaning the status data of a substation equipment is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0225] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0226] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0227] Acquire a state data sequence to be detected of the substation equipment; wherein each state data in the state data sequence to be detected is data including state values ​​of multiple parameters;

[0228] Determine dirty data in the state data sequence to be detected;

[0229] Obtain the target state data at the previous moment corresponding to the dirty data;

[0230] The target state data, Koopman operator, dimension reduction matrix and target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0231] A computer device provided by an embodiment of the present invention has the same beneficial effects as the above-mentioned method for cleaning status data of a substation device.

[0232] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0233] Acquire a state data sequence to be detected of the substation equipment; wherein each state data in the state data sequence to be detected is data including state values ​​of multiple parameters;

[0234] Determine dirty data in the state data sequence to be detected;

[0235] Obtain the target state data at the previous moment corresponding to the dirty data;

[0236] The target state data, Koopman operator, dimension reduction matrix and target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0237] A computer-readable storage medium provided by an embodiment of the present invention has the same beneficial effects as the above-mentioned method for cleaning status data of substation equipment.

[0238] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0239] Acquire a state data sequence to be detected of the substation equipment; wherein each state data in the state data sequence to be detected is data including state values ​​of multiple parameters;

[0240] Determine dirty data in the state data sequence to be detected;

[0241] Obtain the target state data at the previous moment corresponding to the dirty data;

[0242] The target state data, Koopman operator, dimension reduction matrix and target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

[0243] A computer program product provided by an embodiment of the present invention has the same beneficial effects as the above-mentioned method for cleaning status data of substation equipment.

[0244] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0245] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0246] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for cleaning status data of substation equipment, characterized in that: The method comprises: Acquire a state data sequence to be detected of the substation equipment; wherein each state data in the state data sequence to be detected is a three-dimensional array of state values ​​of three parameters including top oil temperature, load and ambient temperature; the state data sequence refers to a collection of state data of the substation equipment at each time point within a period of time; Determine whether each state data in the state data sequence to be detected belongs to a preset data cluster; the data cluster is obtained by clustering historical normal state data sets; Determine a state transition probability matrix corresponding to each of the parameters, and determine a parameter data sequence to be detected corresponding to each of the parameters according to the state data sequence to be detected; Determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each of the parameters using each of the state transition probability matrices; determine the dirty data in the state data sequence to be detected according to the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each of the state transition probability sequences; Acquire target state data at a previous moment corresponding to the dirty data; Using sample parameters of the basis function and a deep neural network to train target parameters of the basis function, and determining a target basis function according to the target parameters; Determine the historical state data sequence of two adjacent moments according to the historical normal state data set, and use the target basis function to perform a dimension-upgrading operation on the two historical state data sequences to obtain the historical state data sequence after dimension-upgrading; Using the dimension-upgraded historical state data sequence to perform extended dynamic modal decomposition to obtain a Koopman operator and a dimension-reduction matrix; The target state data, the Koopman operator, the dimension reduction matrix and the target basis function are used to calculate the estimated state data at the time corresponding to the dirty data, and the dirty data is replaced by the estimated state data.

2. The method according to claim 1, characterized in that The determining the dirty data in the state data sequence to be detected according to the state data that does not belong to the data cluster in the state data sequence to be detected and the 0 value in each state transition probability sequence includes: Determine a first quantity of the state data that does not belong to the data cluster in the state data sequence to be detected; Determine a second number of the state transition probability sequences having a value of 0 in each of the state transition probability sequences and a third number of the values ​​of 0 in each of the state transition probability sequences; If the first number is less than a first number threshold, the second number is less than a second number threshold, and the third number is greater than a third number threshold, the state data in the state data sequence to be detected that does not belong to the data cluster is determined as the dirty data.

3. The method according to claim 1, characterized in that The method of using each of the state transition probability matrices to determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each of the parameters includes: Inputting a target parameter data sequence into a self-organizing neural network model, and using the self-organizing neural network model to quantize each state value in the target parameter data sequence to obtain a state quantization value; the target parameter data sequence is a parameter data sequence to be detected corresponding to the target parameter in the parameter data sequence to be detected, and the target parameter is any one of the parameters; According to the corresponding relationship between the state quantization value and the transition probability in the target state transfer probability matrix, the target state transfer probability sequence corresponding to the target parameter data sequence is determined; the target state transfer probability matrix is ​​a state transfer probability matrix corresponding to the target parameters of the target parameter data sequence.

4. The method according to claim 1, characterized in that Determining the data cluster according to the historical normal state data set by clustering includes: Calculating the data local density of each historical state data in the historical normal state data set and the distance between each historical state data; Determining the data cluster according to the magnitude relationship of each local density of the data and the magnitude relationship of each distance; The determining whether each state data in the sequence of state data to be detected belongs to a preset data cluster includes: Whether each state data belongs to the data cluster is determined according to the local density and current distance between each state data in the state data sequence to be detected and the data cluster.

5. The method according to claim 3, characterized in that: The determining of the state transition probability matrix corresponding to each of the parameters comprises: Determine, according to the historical normal state data set, a historical normal parameter data sequence corresponding to each of the parameters; Quantifying each target historical parameter data in the target historical normal parameter data sequence by using the pre-trained self-organizing neural network model, and outputting the target historical quantization value corresponding to each target historical parameter data; the target historical normal parameter data sequence is a historical normal parameter data sequence corresponding to the target parameter, and the target parameter is any one of the parameters; According to the probability distribution of the target historical quantized value, the probability of transfer between neurons in the self-organizing neural network model is determined, and a state transfer probability matrix corresponding to the target parameter is obtained.

6. A state data cleaning device for substation equipment, characterized in that: The device comprises: The first acquisition module is used to acquire a sequence of state data to be detected of the substation equipment; wherein each state data in the sequence of state data to be detected is a three-dimensional array of state values ​​of three parameters including top oil temperature, load and ambient temperature; the state data sequence refers to a collection of state data of the substation equipment at each time point within a period of time; The first determination module is used to determine whether each of the state data in the state data sequence to be detected belongs to a preset data cluster; the data cluster is obtained by clustering the historical normal state data set; determine the state transition probability matrix corresponding to each of the parameters, and determine the parameter data sequence to be detected corresponding to each of the parameters according to the state data sequence to be detected; use each of the state transition probability matrices to determine the state transition probability sequence of the parameter data sequence to be detected corresponding to each of the parameters; determine the dirty data in the state data sequence to be detected according to the state data in the state data sequence to be detected that does not belong to the data cluster and the 0 value in each of the state transition probability sequences; A second acquisition module is used to acquire target state data at a previous moment corresponding to the dirty data; The second determination module is used to train the target parameters of the basis function using the sample parameters of the basis function and the deep neural network, and determine the target basis function according to the target parameters; determine the historical state data sequence of two adjacent moments according to the historical normal state data set, and use the target basis function to perform a dimensionality increase operation on the two historical state data sequences to obtain the historical state data sequence after dimensionality increase; use the historical state data sequence after dimensionality increase to perform extended dynamic mode decomposition to obtain the Koopman operator and the dimensionality reduction matrix; use the target state data, the Koopman operator, the dimensionality reduction matrix and the target basis function to calculate the estimated state data at the moment corresponding to the dirty data, and use the estimated state data to replace the dirty data.

7. The device according to claim 6, characterized in that The first determining module is further used for: Determine a first quantity of the state data that does not belong to the data cluster in the state data sequence to be detected; Determine a second number of the state transition probability sequences having a value of 0 in each of the state transition probability sequences and a third number of the values ​​of 0 in each of the state transition probability sequences; If the first number is less than a first number threshold, the second number is less than a second number threshold, and the third number is greater than a third number threshold, the state data in the state data sequence to be detected that does not belong to the data cluster is determined as the dirty data.

8. The device according to claim 6, characterized in that The first determining module is further used for: Inputting a target parameter data sequence into a self-organizing neural network model, and using the self-organizing neural network model to quantize each state value in the target parameter data sequence to obtain a state quantization value; the target parameter data sequence is a parameter data sequence to be detected corresponding to the target parameter in the parameter data sequence to be detected, and the target parameter is any one of the parameters; According to the corresponding relationship between the state quantization value and the transition probability in the target state transfer probability matrix, the target state transfer probability sequence corresponding to the target parameter data sequence is determined; the target state transfer probability matrix is ​​a state transfer probability matrix corresponding to the target parameters of the target parameter data sequence.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Citation Information

Patent Citations

  • Power grid load prediction method and system based on neural network and dynamic mode decomposition

    CN113537649A

  • Power system state estimation method, device and equipment

    CN113553755A

  • Systems and methods for modeling continuous stochastic processes with dynamic normalizing flows

    US20210256358A1