Method for semantic analysis of self-checking alarm information of a relay protection device and related apparatus

Abstract

The application discloses a kind of semantic analysis methods of self-checking alarm information of relay protection equipment and related devices.Therein, method includes: the collection of self-checking alarm information feature set of relay protection equipment;Using the low-dimensional feature space of pre-training to the self-checking alarm information feature set Effective feature extraction, determine the predetermined dimension of self-checking alarm information feature subset;Using the two-class classifier of pre-construction to the self-checking alarm information feature subset Classification, determine the semantic label corresponding to self-checking alarm information feature subset.Realize the automatic understanding of original self-checking alarm information semantics, support relay protection online monitoring and operation analysis.

Description

Technical Field

[0001] This invention relates to the field of relay protection technology, and more specifically, to a semantic parsing method and related apparatus for self-test alarm information of relay protection equipment. Background Technology

[0002] Based on the relay protection statistical analysis system, over 10,000 defect records from the State Grid Corporation of China from 2012 to 2019 were selected. Most of these defects occurred at the hardware board level of the device. The defect data in the relay protection statistical analysis system provides a sample set for the semantic understanding of the original self-test alarm information of relay protection. The defect records contain the phenomenon of the defect, involving the time of the defect occurrence and information such as alarms, panel displays, and background messages within the previous time window; the defect records also cover the cause and location of the defect, supporting a further understanding of the semantics of the self-test alarm information. Furthermore, although post-event analysis of defect information can clearly understand the precise semantics of the self-test alarm information of relay protection equipment, the timeliness of reporting defect information is low and the workload is large due to its archival nature. Field applications have a significant need for automatically determining the semantics of self-test alarm information and identifying the defect location and cause. This application, based on relay protection defect record information, completes semantic tag modeling of the self-test alarm information of relay protection equipment, and then automatically extracts semantic tags from the online original self-test alarm information, providing support for operation and maintenance and professional management.

[0003] Statistical analysis of relay protection defects reveals over 90 types of defect locations and over 50 specific causes. For example, the combination of the defect location "CPU plug-in" and the specific cause "storage error"—a CPU board internal storage error—represents a deep semantic understanding of related original self-test alarm information (such as "RAM self-test error"). Similar permutations and combinations result in a wide variety of semantic understandings of self-test alarms. On the other hand, the types of online operational information such as self-test alarms from relay protection equipment far exceed the number of alarm information types specified in the "Substation Relay Protection Information Specification" (DL / T 1782—2017). After removing redundant information and appropriate classification, there are over two hundred alarm information descriptions. The original self-test alarm information is also diverse. Currently, there are no research results in this area regarding the application of conventional methods such as neural networks and deep learning to train historical defect data of relay protection to construct a mapping space from original self-test alarm information to its semantic labels. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a semantic parsing method and related apparatus for self-test alarm information of relay protection equipment.

[0005] According to one aspect of the present invention, a semantic parsing method for self-test alarm information of relay protection equipment is provided, comprising:

[0006] Collect feature sets of self-test alarm information from relay protection equipment;

[0007] Effective feature extraction is performed on the self-check alarm information feature set using a pre-trained low-dimensional feature space to determine a self-check alarm information feature subset of a predetermined dimension.

[0008] A pre-built binary classifier is used to classify the feature subset of self-check alarm information and determine the semantic label corresponding to the feature subset of self-check alarm information.

[0009] Optionally, it also includes: training a low-dimensional feature space through the following steps:

[0010] Based on the pre-acquired training samples, a corresponding class attribute feature space is created for each pre-defined semantic label, where the training samples are the historical self-test alarm information of the relay protection device.

[0011] The class attribute feature space is divided using the random subspace method to determine the class attribute feature subspace;

[0012] The domain rough set method is used to extract the feature set in the class attribute feature subspace and determine the effective low-dimensional features, which are then used to construct a low-dimensional feature space.

[0013] Optionally, the operation of creating a corresponding class attribute feature space for each predefined semantic label based on pre-acquired training samples includes:

[0014] Based on the actual situation of each semantic label, the training samples are split to determine the negative class sample subset and the positive class sample subset, where the negative class sample subset and the positive class sample subset are mutually exclusive sample subsets;

[0015] Clustering algorithms were used to perform cluster analysis on the negative class sample subsets and the positive class sample subsets respectively, and the results of the first cluster analysis and the second cluster analysis were determined.

[0016] Based on the results of the first and second cluster analyses, a class attribute feature space is constructed.

[0017] Optionally, the operation of segmenting the class attribute feature space using a random subspace method to determine the class attribute feature subspace includes:

[0018] A predetermined number of features are randomly selected from the class attribute feature space to form a random subspace;

[0019] Define the proximity relationships between features within a random subspace;

[0020] Based on rough set theory, the correlation between features and corresponding semantic labels in a random subspace is determined.

[0021] Simulated annealing algorithm is used to filter and optimize features in random subspace.

[0022] Optionally, it also includes: constructing a binary classifier based on a low-dimensional feature space, the operation of which includes:

[0023] Obtain semantic labels and effective low-dimensional features to determine the sample set;

[0024] The discriminant function of a binary classifier is trained using a sample set, where the discriminant function is used to determine the semantic labels of the features of the original self-check alarm information.

[0025] According to another aspect of the present invention, a semantic parsing device for self-test alarm information of relay protection equipment is provided, comprising:

[0026] The data acquisition module is used to collect the feature set of self-test alarm information from relay protection equipment.

[0027] The first determining module is used to effectively extract features from the self-check alarm information feature set using a pre-trained low-dimensional feature space, and determine a self-check alarm information feature subset of a predetermined dimension.

[0028] The second determination module is used to classify the feature subset of self-check alarm information using a pre-built binary classifier and determine the semantic label corresponding to the feature subset of self-check alarm information.

[0029] According to another aspect of the present invention, a computer-readable storage medium is provided, the storage medium storing a computer program for performing the methods described in any of the above aspects of the present invention.

[0030] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method described in any of the preceding aspects of the present invention.

[0031] Therefore, based on the understanding of typical self-test alarm information semantic tags, this application extracts the mapping relationship between the original self-test alarm information and semantic tags, realizes the automatic understanding of the semantics of the original self-test alarm information, and supports online monitoring and operation analysis of relay protection. Attached Figure Description

[0032] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0033] Figure 1 This is a flowchart illustrating a semantic parsing method for self-test alarm information of a relay protection device provided in an exemplary embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram illustrating the variation of different timing numbers E(Fi(1)) using the simulated annealing algorithm provided in an exemplary embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram of cluster centers when “AD anomaly”, “zero drift exceeding the limit”, and “dual AD inconsistency” are selected as features of the semantic label of A / D conversion anomaly of the CPU board, provided by an exemplary embodiment of the present invention.

[0036] Figure 4 This is a schematic diagram of the structure of a semantic parsing device for self-test alarm information of relay protection equipment provided in an exemplary embodiment of the present invention;

[0037] Figure 5 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. Detailed Implementation

[0038] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0039] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0040] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of the present invention are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0041] It should also be understood that in the embodiments of the present invention, "multiple" can refer to two or more, and "at least one" can refer to one, two or more.

[0042] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more unless explicitly defined or given contrary instructions in the context.

[0043] Furthermore, the term "and / or" in this invention is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this invention generally indicates that the preceding and following related objects have an "or" relationship.

[0044] It should also be understood that the description of the various embodiments in this invention emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0045] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0046] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0047] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0048] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0049] The embodiments of this invention can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate together with a wide range of other general-purpose or special-purpose computing system environments or configurations. Well-known examples of terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, and servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments including any of the above systems, etc.

[0050] Electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are executed by remote processing devices linked through communication networks. In distributed cloud computing environments, program modules can reside on local or remote computing system storage media, including storage devices.

[0051] Exemplary methods

[0052] Figure 1This is a flowchart illustrating a semantic parsing method for self-test alarm information of a relay protection device according to an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as... Figure 1 As shown, the semantic parsing method 100 for self-test alarm information of relay protection equipment includes the following steps:

[0053] Step 101: Collect the feature set of self-test alarm information of relay protection equipment;

[0054] Step 102: Use a pre-trained low-dimensional feature space to effectively extract features from the self-check alarm information feature set and determine a self-check alarm information feature subset of a predetermined dimension.

[0055] Step 103: Use a pre-built binary classifier to classify the feature subset of the self-check alarm information and determine the semantic label corresponding to the feature subset of the self-check alarm information.

[0056] The defect records of relay protection equipment contain the phenomenon of the defect, including the time of the defect occurrence and information such as alarms, panel displays, and background messages within the previous time window. The defect records also cover the cause and location of the defect, supporting a deeper understanding of the semantics of self-test alarm information. The self-test alarm information of relay protection equipment is used as a feature set for semantic analysis.

[0057] Therefore, based on the understanding of typical self-test alarm information semantic tags, this application extracts the mapping relationship between the original self-test alarm information and semantic tags, realizes the automatic understanding of the semantics of the original self-test alarm information, and supports online monitoring and operation analysis of relay protection.

[0058] Optionally, it also includes: training a low-dimensional feature space through the following steps:

[0059] Based on the pre-acquired training samples, a corresponding class attribute feature space is created for each pre-defined semantic label, where the training samples are the historical self-test alarm information of the relay protection device.

[0060] The class attribute feature space is divided using the random subspace method to determine the class attribute feature subspace;

[0061] The domain rough set method is used to extract the feature set in the class attribute feature subspace and determine the effective low-dimensional features, which are then used to construct a low-dimensional feature space.

[0062] Optionally, the operation of creating a corresponding class attribute feature space for each predefined semantic label based on pre-acquired training samples includes:

[0063] Based on the actual situation of each semantic label, the training samples are split to determine the negative class sample subset and the positive class sample subset, where the negative class sample subset and the positive class sample subset are mutually exclusive sample subsets;

[0064] Clustering algorithms were used to perform cluster analysis on the negative class sample subsets and the positive class sample subsets respectively, and the results of the first cluster analysis and the second cluster analysis were determined.

[0065] Based on the results of the first and second cluster analyses, a class attribute feature space is constructed.

[0066] Optionally, the operation of segmenting the class attribute feature space using a random subspace method to determine the class attribute feature subspace includes:

[0067] A predetermined number of features are randomly selected from the class attribute feature space to form a random subspace;

[0068] Define the proximity relationships between features within a random subspace;

[0069] Based on rough set theory, the correlation between features and corresponding semantic labels in a random subspace is determined.

[0070] Simulated annealing algorithm is used to filter and optimize features in random subspace.

[0071] The final result of the random subspace is a small number of self-check alarm messages that are highly correlated with the semantic labels.

[0072] Optionally, it also includes: constructing a binary classifier based on a low-dimensional feature space, the operation of which includes:

[0073] Obtain semantic labels and effective low-dimensional features to determine the sample set;

[0074] The discriminant function of a binary classifier is trained using a sample set, where the discriminant function is used to determine the semantic labels of the features of the original self-check alarm information.

[0075] Specifically, due to the high dimensionality (multiple types of original self-test alarm information), large amount of redundant information, and strong label sparsity in the semantic label extraction of relay protection self-test alarm information, dimensionality reduction becomes the focus of semantic label extraction. Since there are generally correlations among relay protection self-test alarm information, considering the correlations and differences between different self-test alarm information is a breakthrough point for semantic label extraction. At the same time, each category of semantic label may possess some unique attribute features (or category attributes, i.e., unique original alarms). These features are only related to the corresponding label, which will be more helpful in identifying the corresponding label. Therefore, this application extracts semantic labels by establishing a category feature space and effective low-dimensional features (i.e., a small amount of effective original self-test alarm information related to the semantic label), including two steps: training and prediction of the semantic understanding model for self-test alarm information.

[0076] Training steps

[0077] During the training step, for each different semantic label l k ∈L, train semantic label extraction models respectively:

[0078] (1) Create a corresponding category feature space for each semantic label.

[0079] Generally speaking, in the field of machine learning research, there is a common phenomenon: after a sample set is processed by a similarity-based clustering analysis algorithm, several cluster centers can be obtained. These cluster centers are often highly representative and can effectively reflect some commonalities of their corresponding sample clusters.

[0080] 1) First, for any semantic tag l k ∈L, based on the actual labeling of this tag, the entire training sample set (the set of historical self-test alarm information of relay protection equipment) can be split into negative classes N. k and positive class P k Two disjoint subsets of samples are represented as follows:

[0081] P k ={x i |(x i ,Y i )∈T,l k ∈Y i} (1)

[0082]

[0083] For the corresponding sample x i The set of semantic labels possessed by , where T is the training sample set, and P is the set of semantic labels possessed by . k Indicates that by owning lk A subset N of the labeled samples k It is composed of all those not l k The composition of the labeled samples. Generally, for any sample set X = {x1, x2, ..., x...} n}, all pairs of samples that satisfy the positive constraint relationship should be organized into a positive constraint set, denoted as M = {(x i ,x j )|x i ,x j Belonging to the same category}; correspondingly, the negative constraint set consists of all sample pairs that satisfy the negative constraint relationship, denoted as C={(x i ,x j )|x i ,x j They belong to different categories.

[0084] 2) Then, clustering algorithms are used to analyze and cluster the two sample sets respectively. Under the assumption that constraint sets M and C have the same use value (i.e., the information content of constraint sets M and C is equivalent), the number of clusters in the two sets is set to the same value. Set P... k All samples in m are aggregated into m k There are clusters, and the center of each cluster is denoted as . Set N k It can also be divided into m k There are three distinct clusters, and the center of each cluster is denoted as . Cluster centers can effectively represent the potential structural information of the corresponding set.

[0085] 3) Based on the cluster analysis results, semantic tags are constructed. k The class feature space.

[0086] Semantic tag l k Category attribute characteristics It can be defined as:

[0087]

[0088] Where d(*,*) represents the distance between samples. Semantic label l k The category feature space is based on category attribute features The space formed by each element as a coordinate.

[0089] (2) Using the design concept of random subspaces, the created class feature space is divided.

[0090] Based on the generic feature space in the previous step, this section proposes a method for constructing a random subspace, and based on rough set theory, determines whether the features (original self-check alarm information) covered by this random subspace are strongly correlated with the semantic label l k ∈L (that is, the strength of the directivity of the self-check alarm information obtained by searching to the semantic label lk), laying a foundation for step (3).

[0091] 1) Select P features to form a random subspace

[0092] Using the generic attribute feature space, randomly select P (P < D) features (features composed of original self-check alarm information) in the original sample space with dimension D to create a subspace, denoted by F t to represent any random subspace composed of P-dimensional data samples, that is, F t ={f t,1 , f t,2 , …, f t,N}.

[0093] 2) Define the proximity relationship of samples in the space

[0094] To clearly determine the proximity relationship of samples in the subspace, the distance value between a sample and the center of the clustering cluster of its nearest same-class samples is selected to adaptively select relevant neighbors. That is, in any random subspace F t , for any sample f t,i , its actual proximity relationship depends on the distance value D t,i from the sample f t,i to the center of the clustering cluster of its nearest same-class samples for determination. Among them, the sample spacing is described by the Euclidean distance. If the spacing d t,j from sample f t,i to f t,ij is not greater than the distance value D t,i , then f t,j is regarded as the neighbor of f t,i , otherwise it is considered that there is no proximity relationship between the two. The formula for D t,i is:

[0095]

[0096] 3) Based on rough set theory, quantitatively evaluate the correlation between the features (original self-check alarm information) included in the random subspace and the semantic label l k ∈L

[0097] To measure the effect of random subspaces composed of different features on the extraction of feature labels, we introduce neighborhood rough set theory to calculate the neighborhood dependency of the P-dimensional random subspace. From the perspective of the semantic labels under study, we analyze the classification effect of the random subspace composed of the extracted features, that is, whether the projections of samples that are close to each other in the original space are also close to each other in the low-dimensional space; on the other hand, for non-nearest neighbor samples, we identify whether their projection points in the low-dimensional space are spread out as much as possible.

[0098] With any random subspace F t Based on this, and according to the distance between samples, an approximate neighborhood graph is adaptively constructed. Approximate graph in the neighborhood The samples are displayed as nodes in the relationship graph, and the nearest neighbor relationships between adjacent nodes are represented by the edges between them. Based on the above relationship graph, construct the following two weight matrices, where: Used to reflect the approximate relationship in the neighborhood of different samples, indicating whether any two samples are close to each other in terms of distance; matrix Used to represent the approximate relationship in the neighborhood of different samples, indicating that when sample f t,i with f t,j When the labels are inconsistent, can the distance information between the two samples be used to effectively separate them? The specific weight values ​​in the above matrix are set as follows:

[0099]

[0100]

[0101] In the formula: d t,ij Used to reflect sample f t,i to f t,j The spacing is represented here by Euclidean distance.

[0102] Based on the upper and lower neighborhood approximation graphs, the average weight of each edge in the graph is calculated to obtain the combined weight of the upper and lower neighborhood approximation graphs. They are respectively:

[0103]

[0104]

[0105] (3) Integrate the proximity relationships of samples in each subspace (samples are the sample data described in "2. Background Technology", which covers self-check alarm information and its corresponding semantic labels), and extract effective low-dimensional features (i.e. a small amount of effective original self-check alarm information related to semantic labels) by combining simulated annealing-neighborhood rough set method (one semantic label corresponds to one low-dimensional feature space, and the low-dimensional space is obtained through steps (1)-(3)).

[0106] The goal of effective feature extraction is to select the most effective low-dimensional feature cluster (original self-check alarm information) for judging a semantic label. By training a classifier on this feature cluster (original self-check alarm information), the accuracy reduction problem caused by the interrelationships between original self-check alarm information during the semantic label selection process (establishing a mapping from original self-check alarm information to multiple semantic labels) can be avoided. Simulated annealing algorithm can quickly and effectively search for random self-check alarm information in the label space, maximizing the neighborhood dependency of the searched P-dimensional random subspace (the stronger the directionality of the searched self-check alarm information to a certain self-check alarm semantic label).

[0107] Simulated annealing is an optimization of the greedy algorithm. Its goal is to escape local optima and search for the optimal solution in the global space. In key feature extraction applications, simulated annealing can accept several non-optimal feature combinations to collect a complete set of better features, thus achieving effective key feature extraction. Its operation involves accepting a solution that is worse than the current one with a certain probability, sacrificing iteration steps to strive for a better solution.

[0108] The steps for extracting effective low-dimensional features (highly correlated original self-check alarm information) based on the simulated annealing algorithm are as follows:

[0109] 1) Let the iteration step number g = 0, and the annealing temperature τ = τ(g) = τ(0), which is the initial temperature at which annealing begins. Generate h P-dimensional random subspaces F1(0), F2(0), ..., F h (0);

[0110] 2) Based on the definition of neighborhood dependency, calculate the label l to be studied. k Regarding the random subspaces F1(g), F2(g), ..., F h (g) Neighborhood dependency E(F) i (g)), in, The calculation is based on formulas (7) and (8);

[0111] 3) Let τ be equal to the next value τ(g+1) in the cooling schedule;

[0112] 4) Based on the current random subspace F i(g) Perturb the space to generate a new random subspace F. i (g+1), calculate the corresponding objective function value E(F) i (g+1)), we get ΔE=E(F i (g+1))-E(F i (g));

[0113] 5) If ΔE < 0, then the new solution F i (g+1) is accepted as the new current solution; if ΔE > 0, then the new solution F i (g+1) is accepted with probability exp(-ΔE / τ(g+1)), where τ(g+1) is the current temperature;

[0114] 6) At temperature τ(g+1), repeat the perturbation and acceptance process L times, i.e., perform steps 4) and 5);

[0115] 7) Determine if τ has been reached. f If yes, then terminate the algorithm; otherwise, g = g + 1, and proceed to step 2) to continue execution. The random subspace with the highest neighborhood dependency is the one with label l. k The P features with the strongest correlation;

[0116] 8) Update the dimension P of the random subspace, and repeat steps 2) to 7). For different dimensions P, determine the optimal random subspace under different dimensions P by calculating the neighborhood dependency E, and take the features in the random subspace with the largest E as the features related to the label l. k The P features with the strongest correlation.

[0117] (4) Based on the reduced class feature space, learn to construct the corresponding binary classifier.

[0118] To determine the semantic tags of self-test alarm information k Essentially, this is a pattern classification problem. Based on the key feature set (i.e., self-test alarm information) extracted in steps (1) to (3), a sample set covering different original self-test alarm information (i.e., features) and semantic labels is established; then, a discriminant function is trained based on the sample set; finally, the discriminant function is applied to determine the corresponding semantic label of the original self-test alarm information. Label extraction is to determine whether a semantic label matches the self-test alarm information of the relay protection device, which is a typical binary classification problem. This application uses the Fisher classifier.

[0119] When considering projecting samples from a multidimensional space onto a straight line to form a one-dimensional space, the samples form several compact and mutually separate clusters in the P-dimensional space. If we project them onto an arbitrary straight line, we might find a direction along which the projected samples are best separated. The Fisher classifier finds this best and easiest-to-classify projection line based on the samples, and its steps are as follows:

[0120] 1) Calculate the mean vector m of each type of sample i N i It is w i The number of samples in class (i = 1, 2). For the semantic label extraction problem, the two classes represent samples that have the semantic label (the original self-check alarm information corresponds to the semantic label) and samples that do not have the semantic label (the original self-check alarm information does not correspond to the semantic label).

[0121]

[0122] 2) Calculate the within-class scatter matrix S of the samples. i The total scatter matrix S within the class w .

[0123]

[0124] S w =S1+S2 (11)

[0125] 3) Calculate the inter-class scatter matrix S of the samples. b

[0126] S b =(m1-m2)(m1-m2) T (12)

[0127] 4) Find the vector w*. For this, define the Fisher criterion function.

[0128]

[0129] Make J F (w) The maximum value of w* is

[0130]

[0131] 5) Project all samples in the training set.

[0132] Y = (w * ) T x (15)

[0133] 6) Calculate the segmentation threshold y0 in the projection space. Different methods can be used to select the threshold; one commonly used method is...

[0134]

[0135] in, The mean of each type of sample in one-dimensional space:

[0136] Within-class dispersion of samples The total intra-class dispersion is for

[0137]

[0138]

[0139] 7) For a given x, calculate its projection point Y' on w*.

[0140] Y'=(w * ) T x (19)

[0141] 8) According to the classification of decision-making rules, there are

[0142]

[0143] Prediction Step

[0144] Extract self-test alarm information (features i.e. self-test alarm information) from relay protection equipment, and rely on low-dimensional feature space and binary classifier to determine the semantic labels of self-test alarm information, thereby realizing semantic understanding of self-test alarm information.

[0145] Furthermore, with the goal of extracting semantic labels corresponding to the original self-check alarm information, a category feature space is created for each semantic label. The category feature space is partitioned using a random subspace, and an effective low-dimensional feature space is searched using the simulated annealing-neighborhood rough set algorithm. Taking the construction of the random subspace for the semantic label "CPU board A / D conversion anomaly" as an example, with P=5, the trajectory of max(E(Fi(h))) as a function of iteration step h is calculated using the simulated annealing algorithm, as shown below. Figure 2 As shown. From Figure 2 As can be seen from the initial iteration, because the features in the random subspace (the original self-check alarm information) are not effective key features of the semantic label "CPU board A / D conversion anomaly", these features cannot classify the label as positive class P. k With negative class N k Separately, according to the definition of the approximate graph in the neighborhood, for the positive class P k With negative class N kFor any node in the neighborhood approximation graph formed by any sample in the graph, the weight of the connection between the node and other nodes of the same class in the neighborhood approximation graph is 1 only if all other nodes that are not of the same class as the node can be effectively separated by the features in the random subspace. Otherwise, the weight is 0, as shown in Equation (6). As the iteration proceeds, the features in the random subspace are gradually replaced by effective key features. These key features can play the role of distinguishing positive class Pk and negative class Nk samples. Therefore, the maximum neighborhood dependency max(E(Fi(h))) increases stepwise with the number of iteration steps. For the semantic label "CPU board A / D conversion abnormal", when P=5, the features in the random subspace that obtain the maximum neighborhood dependency (the features are the original self-check alarm information) are "AD abnormal", "zero drift limit", "double AD inconsistency", "AD communication interruption" and "AD sampling error". Figure 3 The positive class P was plotted when "AD anomaly", "zero drift exceeding the limit", and "double AD inconsistency" were used as features of the random subspace. k Cluster center As shown in the red "+" in the picture. Figure 3 In the diagram, coordinate = 0 indicates that the sample covers the self-check alarm information corresponding to that coordinate, and coordinate = 1 indicates that the sample does not cover the self-check alarm information corresponding to that coordinate. Negative class N k The coordinates of the sample are generally (0,0,0), and the positive class P k The cluster centers are closer to the positive class sample points (as shown by the blue dots in the figure), while they are closer to the negative class N. k The large distance between sample points indicates that "AD anomaly", "zero drift exceeding the limit", and "double AD inconsistency" are effective features.

[0146] In the derived low-dimensional feature space, a binary classifier is trained using the Fisher discriminant method to obtain the inference logic for determining whether the original self-check alarm information conforms to a specific semantic label based on the low-dimensional feature space. The inference logic for several typical labels is shown in Table 1.

[0147] Table 1

[0148]

[0149]

[0150] Using the idea of ​​random subspaces, the neighborhood dependency of different random subspaces is calculated. The simulated annealing algorithm is used to iterate step by step to obtain effective key features (features are the original self-check alarm information). The semantic labels are obtained by training the Fisher classifier.

[0151] Thus, based on the understanding of typical self-test alarm information semantic tags, the mapping relationship between the original self-test alarm information and semantic tags is extracted, realizing the automatic understanding of the semantics of the original self-test alarm information and supporting online monitoring and operation analysis of relay protection.

[0152] Exemplary device

[0153] Figure 4 This is a schematic diagram of the structure of a semantic parsing device for self-test alarm information of a relay protection device provided in an exemplary embodiment of the present invention. For example... Figure 4 As shown, the device 400 includes:

[0154] The data acquisition module 410 is used to acquire the feature set of self-test alarm information of relay protection equipment;

[0155] The first determining module 420 is used to effectively extract features from the self-check alarm information feature set using a pre-trained low-dimensional feature space, and determine a self-check alarm information feature subset of a predetermined dimension.

[0156] The second determining module 430 is used to classify the feature subset of self-check alarm information using a pre-built binary classifier and determine the semantic label corresponding to the feature subset of self-check alarm information.

[0157] Optionally, the device 400 further includes a training module for training a low-dimensional feature space through the following steps:

[0158] A submodule is created to create a corresponding class attribute feature space for each predefined semantic label based on pre-acquired training samples, where the training samples are historical self-test alarm information of relay protection devices.

[0159] The first determining submodule is used to segment the class attribute feature space using the random subspace method to determine the class attribute feature subspace;

[0160] The second determination submodule is used to extract the feature set in the class attribute feature subspace using the domain rough set method, and determine the effective low-dimensional features, wherein the effective low-dimensional features construct a low-dimensional feature space.

[0161] Optionally, submodules can be created, including:

[0162] The first determining unit is used to split the training samples according to the actual situation of each semantic label, and determine the negative sample subset and the positive sample subset, wherein the negative sample subset and the positive sample subset are mutually disjoint sample subsets.

[0163] The second determining unit is used to perform cluster analysis on the negative sample subset and the positive sample subset respectively using a clustering algorithm to determine the first cluster analysis result and the second cluster analysis result;

[0164] The first constituent unit is used to construct the class attribute feature space based on the results of the first cluster analysis and the results of the second cluster analysis.

[0165] Optionally, the first determined submodule includes:

[0166] The second constituent unit is used to randomly select a predetermined number of features from the class attribute feature space to form a random subspace;

[0167] Define a unit to define the proximity relationships between features within a random subspace;

[0168] The third determining unit is used to determine the correlation between features and corresponding semantic labels in a random subspace based on rough set theory.

[0169] The optimization unit is used to filter and optimize features in a random subspace using the simulated annealing algorithm.

[0170] Optionally, the device 400 further includes: a construction module for constructing a binary classifier based on a low-dimensional feature space, the module comprising:

[0171] The third determination submodule is used to obtain semantic labels and effective low-dimensional features to determine the sample set;

[0172] The training submodule is used to train the discriminant function of the binary classifier using the sample set. The discriminant function is used to determine the semantic labels of the features of the original self-check alarm information.

[0173] Exemplary electronic devices

[0174] Figure 5 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. For example... Figure 5 As shown, the electronic device 50 includes one or more processors 51 and memory 52.

[0175] The processor 51 may be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0176] The memory 52 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 51 may execute the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above, and / or other desired functions. In one example, the electronic device may also include an input device 53 and an output device 54, these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0177] In addition, the input device 53 may also include, for example, a keyboard, a mouse, etc.

[0178] The output device 54 can output various information to the outside. The output device 54 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0179] Of course, for the sake of simplicity, Figure 5 Only some of the components of the electronic device relevant to the present invention are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0180] Exemplary computer program products and computer-readable storage media

[0181] In addition to the methods and apparatus described above, embodiments of the present invention may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0182] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0183] Furthermore, embodiments of the present invention may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the methods for information mining of historical change records according to various embodiments of the present invention as described in the "Exemplary Methods" section above.

[0184] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0185] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the aforementioned specific details.

[0186] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0187] The block diagrams of devices, systems, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, systems, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0188] The methods and systems of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of the present invention are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the present invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the present invention. Thus, the present invention also covers recording media storing programs for performing the methods according to the present invention.

[0189] It should also be noted that in the systems, apparatus, and methods of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered equivalents of the present invention. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0190] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for semantic analysis of self-checking alarm information of a protective relay device, characterized in that, The method comprises the following steps: collecting a self-check alarm information feature set of a relay protection device; performing effective feature extraction on the self-check alarm information feature set by using a pre-trained low-dimensional feature space to determine a self-check alarm information feature subset of a predetermined dimension; classifying the self-check alarm information feature subset by using a pre-constructed two-class classifier to determine a semantic label corresponding to the self-check alarm information feature subset; training the low-dimensional feature space by the following steps: creating a class attribute feature space corresponding to each pre-set semantic label according to pre-acquired training samples, wherein the training samples are historical self-check alarm information of the relay protection device; segmenting the class attribute feature space by using a random subspace method to determine a class attribute feature subspace; extracting features in the class attribute feature subspace by using a domain rough set method to determine effective low-dimensional features, wherein the effective low-dimensional features construct the low-dimensional feature space; segmenting the class attribute feature space by using a random subspace method to determine a class attribute feature subspace, which comprises the following steps: randomly selecting a predetermined number of features in the class attribute feature space to form a random subspace; defining a neighborhood relationship between features in the random subspace; determining the correlation between features in the random subspace and corresponding semantic labels based on rough set theory; screening and optimizing the features in the random subspace by using a simulated annealing algorithm.

2. The method of claim 1, wherein, The operation of creating a class attribute feature space corresponding to each pre-set semantic label according to pre-acquired training samples comprises the following steps: splitting the training samples according to the actual situation of each semantic label to determine a negative class sample subset and a positive class sample subset, wherein the negative class sample subset and the positive class sample subset are mutually exclusive sample subsets; performing clustering analysis on the negative class sample subset and the positive class sample subset respectively by using a clustering algorithm to determine a first clustering analysis result and a second clustering analysis result; constructing the class attribute feature space according to the first clustering analysis result and the second clustering analysis result.

3. The method of claim 1, wherein, The method further comprises the following steps: constructing the two-class classifier based on the low-dimensional feature space, which comprises the following steps: acquiring the semantic labels and the effective low-dimensional features to determine a sample set; training a discriminant function of the two-class classifier by using the sample set, wherein the discriminant function is used to determine the semantic label of original self-check alarm information features.

4. A device for semantic analysis of self-checking alarm information of a protective relay, characterized in that, The method comprises the following steps: a collection module is configured to collect a self-check alarm information feature set of a relay protection device; a first determination module is configured to perform effective feature extraction on the self-check alarm information feature set by using a pre-trained low-dimensional feature space to determine a self-check alarm information feature subset of a predetermined dimension; a second determination module is configured to classify the self-check alarm information feature subset by using a pre-constructed two-class classifier to determine a semantic label corresponding to the self-check alarm information feature subset; a training module is configured to train the low-dimensional feature space by the following steps: The creating submodule is configured to create a class attribute feature space corresponding to each semantic label according to pre-acquired training samples, wherein the training samples are historical self-check alarm information of the relay protection device; The first determining submodule is configured to segment the class attribute feature space by using a random subspace method to determine a class attribute feature subspace; The second determining submodule is configured to extract effective low-dimensional features from the feature set in the class attribute feature subspace by using a domain rough set method to determine the low-dimensional feature space; The first determining submodule comprises: The second constituent unit is configured to randomly select a predetermined number of features in the class attribute feature space to constitute a random subspace; The defining unit is configured to define a neighborhood relationship between the features in the random subspace; The third determining unit is configured to determine the correlation between the features in the random subspace and the corresponding semantic labels based on rough set theory; The optimization unit is configured to filter and optimize the features in the random subspace by using a simulated annealing algorithm.

5. The apparatus of claim 4, wherein, The creating submodule comprises: The first determining unit is configured to split the training samples according to the actual situation of each semantic label to determine a negative class sample subset and a positive class sample subset, wherein the negative class sample subset and the positive class sample subset are mutually exclusive sample subsets; The second determining unit is configured to perform clustering analysis on the negative class sample subset and the positive class sample subset respectively by using a clustering algorithm to determine a first clustering analysis result and a second clustering analysis result; The first constituent unit is configured to constitute the class attribute feature space according to the first clustering analysis result and the second clustering analysis result.

6. The apparatus of claim 4, wherein, Further comprising: The constructing module is configured to construct the two-class classifier according to the low-dimensional feature space, and the module comprises: The third determining submodule is configured to acquire the semantic labels and the effective low-dimensional features to determine a sample set; The training submodule is configured to train a discriminant function of the two-class classifier by using the sample set, wherein the discriminant function is used to determine the semantic label of original self-check alarm information features.

7. A computer readable storage medium characterized in that, The storage medium stores a computer program, and the computer program is used to execute the method in any one of claims 1-3.

8. An electronic device, comprising: The electronic device comprises: A processor; A memory for storing executable instructions of the processor; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method in any one of claims 1-3.

Images