A fault diagnosis method and system for a power concentration communication control device

Abstract

The application discloses a fault diagnosis method and system of a power centralized communication control device, which is used for accurately identifying cascading faults to improve the safety and stability of a power system. The method comprises the following steps: obtaining operation data of the power centralized communication control device, and extracting a frequency feature set from the operation data; calculating the correlation strength between each variable in the frequency feature set by using a conditional probability table, screening out a key feature vector set, and generating an interaction topology between different devices in the power system; calculating the propagation path and propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and screening out a candidate fault propagation path set based on the propagation path and propagation speed; calculating the fault propagation probability of the candidate fault propagation path, determining a cascading fault propagation feature set, constructing a high-dimensional space classification boundary according to the cascading fault propagation feature set, and performing fault mode classification to obtain a fault diagnosis result.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power systems, in particular to a fault diagnosis method and system of a power centralized communication control device. BACKGROUND

[0002] The power centralized communication control device is a key equipment for realizing information transmission, centralized control and coordinated operation between each part in the power system, which is widely used in substation, distribution network and power system dispatching scenes. In the power system, the power centralized communication control device undertakes the important responsibilities of real-time monitoring, data transmission and instruction coordination, and its running state is crucial to the stability of the entire power grid.

[0003] In the existing fault diagnosis method of the power centralized communication control device, the fault recognition is usually dependent on the monitoring parameters of each individual device in the power system. However, with the continuous expansion of the scale and the increasing complexity of the structure of the power system, when facing complex fault scenarios, it is difficult to comprehensively consider the complex interaction relationship and fault propagation characteristics between devices in the power system, so that the cascading failure is difficult to be discovered in time, thereby reducing the safety and stability of the power system. SUMMARY

[0004] The application provides a fault diagnosis method and system of a power centralized communication control device, which can accurately identify cascading failures to improve the safety and stability of the power system.

[0005] The first aspect of the application provides a fault diagnosis method of a power centralized communication control device, comprising:

[0006] Obtaining the running data of the power centralized communication control device, and extracting a frequency feature set from the running data;

[0007] Calculating the correlation strength between each variable in the frequency feature set by using a conditional probability table, and screening out a key feature vector set to generate an interaction topology between different devices in the power system;

[0008] Calculating the propagation path and propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and screening out a candidate fault propagation path set based on the propagation path and propagation speed;

[0009] Calculating the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set;

[0010] Constructing a high-dimensional space classification boundary according to the cascading fault propagation feature set to perform fault mode classification, and obtaining a fault diagnosis result.

[0011] Optionally, the high-dimensional space classification boundary is constructed according to the cascade fault propagation feature set to perform fault mode classification, and a fault diagnosis result is obtained, comprising:

[0012] A cascade fault propagation feature vector in the cascade fault propagation feature set is extracted;

[0013] The cascade fault propagation feature vector is taken as an input of a support vector machine to construct a high-dimensional space classification boundary;

[0014] The interval of the high-dimensional space classification boundary is adjusted by a kernel function;

[0015] A distance measure of the high-dimensional space classification boundary after the interval adjustment is calculated;

[0016] It is judged whether the distance measure is less than a preset distance threshold;

[0017] If yes, it is determined that the fault mode is a cascade fault.

[0018] Optionally, the propagation path and the propagation speed of each key feature vector in the interaction topology are calculated according to a path reasoning algorithm, and a candidate fault propagation path set is screened out based on the propagation path and the propagation speed, comprising:

[0019] The propagation path of each key feature vector in the interaction topology is calculated according to a path reasoning algorithm;

[0020] The propagation speed of each propagation path is calculated based on a propagation speed influencing factor and a corresponding assigned weight;

[0021] A propagation path with a propagation speed greater than a preset speed threshold is determined as a candidate sub-path;

[0022] The path coverage range of the candidate sub-path is calculated;

[0023] A candidate sub-path with a path coverage range greater than a preset coverage threshold is determined as a candidate fault propagation path, so as to obtain a candidate fault propagation path set.

[0024] Optionally, the fault propagation probability of the candidate fault propagation path is calculated to determine the cascade fault propagation feature set, comprising:

[0025] An initial state transition matrix is constructed according to the candidate fault propagation path;

[0026] An evidence updating mechanism is adopted to adjust a conditional probability distribution in the initial state transition matrix;

[0027] A convergence index of the adjusted conditional probability distribution is calculated to obtain a convergence constraint result;

[0028] Based on the convergence constraint results, a set of target paths is selected from the set of candidate fault propagation paths;

[0029] Construct a path feature matrix based on the target path set;

[0030] The failure propagation probability of each target path is calculated based on the path feature matrix, and the cascade failure propagation feature set is determined based on the failure propagation probability.

[0031] Optionally, the step of calculating the correlation strength between variables in the frequency feature set using a conditional probability table and filtering out a set of key feature vectors to generate the interaction topology between different devices in the power system includes:

[0032] Calculate the conditional probabilities of state associations among all variables in the frequency feature set, and construct a complete conditional probability table;

[0033] The correlation strength between variables in the frequency feature set is determined based on the conditional probability table.

[0034] Variable combinations with a correlation strength greater than a preset correlation threshold are identified as key feature vectors;

[0035] Using each device in the power system as a node and key feature vectors as edges, an interaction topology between different devices in the power system is generated.

[0036] Optionally, acquiring the operating data of the centralized power communication control device and extracting a set of frequency features from the operating data includes:

[0037] Acquire operational data from the centralized power communication control device;

[0038] The running data is segmented to obtain a segmented data set;

[0039] Extract the dynamic data set from the fragmented data set;

[0040] The frequency components and periodic features in the dynamic data set are extracted using the small Bode transform.

[0041] A frequency feature set is generated based on the frequency components and the periodic features.

[0042] The second aspect of this application provides a fault diagnosis system for a centralized power communication control device, comprising:

[0043] The acquisition unit is used to acquire the operating data of the power centralized communication control device and extract a set of frequency features from the operating data.

[0044] a first calculation unit, configured to calculate the correlation strength between variables in the frequency feature set by using a conditional probability table, and to screen a key feature vector set to generate an interaction topology between different devices in the power system;

[0045] a second calculation unit, configured to calculate a propagation path and a propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and to screen a candidate fault propagation path set based on the propagation path and the propagation speed;

[0046] a determination unit, configured to calculate a fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set;

[0047] a classification unit, configured to construct a high-dimensional space classification boundary according to the cascading fault propagation feature set to perform fault mode classification and obtain a fault diagnosis result.

[0048] Optionally, the classification unit is specifically configured to:

[0049] extract a cascading fault propagation feature vector in the cascading fault propagation feature set;

[0050] use the cascading fault propagation feature vector as an input of a support vector machine to construct the high-dimensional space classification boundary;

[0051] adjust the interval of the high-dimensional space classification boundary by using a kernel function;

[0052] calculate a distance measure of the high-dimensional space classification boundary after the interval adjustment;

[0053] determine whether the distance measure is less than a preset distance threshold;

[0054] if yes, determine that the fault mode is a cascading fault.

[0055] Optionally, the second calculation unit is specifically configured to:

[0056] calculate the propagation path of each key feature vector in the interaction topology according to the path reasoning algorithm;

[0057] calculate the propagation speed of each propagation path based on a propagation speed influencing factor and a corresponding assigned weight;

[0058] determine a propagation path with a propagation speed greater than a preset speed threshold as a candidate sub-path;

[0059] calculate a path coverage range of the candidate sub-path;

[0060] determine a candidate sub-path with a path coverage range greater than a preset coverage threshold as a candidate fault propagation path to obtain the candidate fault propagation path set.

[0061] Optionally, the determining unit is specifically used for:

[0062] Construct an initial state transition matrix based on candidate fault propagation paths;

[0063] An evidence update mechanism is used to adjust the conditional probability distribution in the initial state transition matrix;

[0064] Calculate the convergence index of the adjusted conditional probability distribution to obtain the convergence constraint results;

[0065] Based on the convergence constraint results, a set of target paths is selected from the set of candidate fault propagation paths;

[0066] Construct a path feature matrix based on the target path set;

[0067] The failure propagation probability of each target path is calculated based on the path feature matrix, and the cascade failure propagation feature set is determined based on the failure propagation probability.

[0068] As can be seen from the above technical solutions, this application has the following effects:

[0069] This process involves acquiring operational data from a centralized power communication control device and extracting a set of frequency features from this data. Conditional probability tables are used to calculate the correlation strength between variables in the frequency feature set, selecting a set of key feature vectors to generate an interaction topology between different devices in the power system. A path reasoning algorithm is then used to calculate the propagation path and speed of each key feature vector in the interaction topology, and a set of candidate fault propagation paths is selected based on these paths and speeds. The fault propagation probability of the candidate fault propagation paths is calculated to determine the set of cascading fault propagation features. A high-dimensional space classification boundary is constructed based on the cascading fault propagation feature set for fault mode classification, yielding fault diagnosis results. In this way, the correlation strength between corresponding variables of various devices in the power system can be used to select correlated key feature vectors, and the propagation path and speed of these key feature vectors can be obtained to determine the set of cascading fault path features that may lead to cascading faults. Finally, the specific fault mode is determined based on the high-dimensional space boundary. This allows for a comprehensive consideration of the complex interaction relationships and fault propagation characteristics between devices in the power system, enabling accurate identification of cascading faults and improving the safety and stability of the power system. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of an embodiment of a fault diagnosis method for a centralized power communication control device according to this application;

[0071] Figure 2-1 , Figure 2-2 as well as Figure 2-3Figure 1 shows a schematic diagram of another embodiment of the method for diagnosing faults of a power centralized communication control device according to the present application;

[0072] Figure 3 Figure 2 shows a schematic diagram of an embodiment of the system for diagnosing faults of a power centralized communication control device according to the present application. DETAILED DESCRIPTION

[0073] The present application provides a method and system for diagnosing faults of a power centralized communication control device, which can accurately identify cascading faults and improve the safety and stability of a power system.

[0074] The method for diagnosing faults of a power centralized communication control device according to the present application can be implemented on a system, a server or a terminal with logical analysis capability.

[0075] Referring to Figure 1, Figure 1 An embodiment of the method for diagnosing faults of a power centralized communication control device according to the present application includes:

[0076] 101. Obtain operation data of the power centralized communication control device, and extract a frequency feature set from the operation data;

[0077] In this embodiment, the power centralized communication control device collects operation parameters of each node in the power system in real time, such as voltage, current and power parameters of devices in the power system, such as distributed power sources, smart meters, transformers, switching devices and protection devices, etc. during operation. The power centralized communication control device transmits the collected operation parameters of each node to the system through a communication network, thereby obtaining the operation data of the power centralized communication control device. The operation data is converted from time domain to frequency domain by Fourier transform or wavelet transform, etc. to obtain the frequency feature set.

[0078] 102. Calculate the correlation strength between each variable in the frequency feature set using a conditional probability table, filter out a key feature vector set, and generate an interaction topology between different devices in the power system;

[0079] In this embodiment, the conditional probability refers to the probability of occurrence of event A under the condition that event B occurs. For each variable in the frequency feature set, according to its value range and actual application requirements, it is divided into several meaningful intervals. For example, for the output frequency variable of the transformer, it can be divided according to the normal operation frequency range, slight fluctuation range, abnormal fluctuation range, etc. For another frequency-related variable, such as the frequency change rate of the power grid node, similar interval division is also performed. The division of these intervals should consider the operation characteristics of the power system and the actual monitoring accuracy. For each pair of variables, the number of samples when one variable is in a particular interval and the other variable is in different intervals is counted. According to the number of samples counted, the conditional probability is calculated, that is, the probability that when one variable takes a particular value (is in a particular interval), the other variable takes different values (is in different intervals). Arranging these conditional probabilities into a table form, the conditional probability table is obtained, which can clearly show the mutual relationship between different variables in the probability level.

[0080] The dependence degree between variables is evaluated through the conditional probability table. The higher the dependence degree, the higher the correlation strength between different variables. For example: if the change of the value of one variable causes a significant change in the conditional probability distribution of another variable, it indicates that there is a strong dependence relationship between the two variables; on the contrary, if the conditional probability distribution changes little, it indicates that the dependence relationship between the variables is weak. Then, according to the actual operation requirements of the power system and experience, combined with the analysis results of the correlation strength, a reasonable screening standard is set. According to the set screening standard, the variable pairs with strong correlation are selected from all variable pairs as key feature vectors. These key feature vectors represent the parts of different devices in the power system that have close contact in terms of frequency characteristics. Finally, according to the correlation strength between the key feature vectors, the interaction topology between different devices in the power system is constructed to clearly show the mutual influence relationship between the devices.

[0081] 103、According to the path reasoning algorithm, the propagation path and propagation speed of each key feature vector in the interaction topology are calculated, and a candidate fault propagation path set is selected based on the propagation path and propagation speed;

[0082] In this embodiment, the interaction topology describes the connection relationship between each node, and the key feature vector represents the starting point or key factor that may cause fault propagation. In the process of using path reasoning algorithm to explore the propagation path, starting from the node where the key feature vector is located, according to the connection relationship between nodes and certain rules, such as finding the shortest path or the path meeting certain conditions, the propagation trajectory of signals or influences from the key feature vector node to other nodes is determined to simulate the propagation mode that the fault is most likely to follow in the actual power system. After obtaining the propagation path, the propagation speed of the propagation path is calculated in combination with other factors affecting the propagation speed, such as time, delay, etc. It can be understood that the propagation speed reflects the speed of the key feature vector spreading in the interaction topology. According to the calculated propagation path and propagation speed, some screening conditions can be set to determine the candidate fault propagation path set

[0083] 104. Calculate the fault propagation probability of the candidate fault propagation path to determine the cascade fault propagation feature set;

[0084] In this embodiment, the fault propagation probability of the candidate fault propagation path is calculated by using the total probability formula or Markov chain, and the key path is identified by using the fault propagation probability, and the cascade fault propagation features are extracted from the key path. The cascade fault propagation features can include fault occurrence sequence, time interval and other features. Among them, the cascade fault is usually not a single and isolated event, but a series of faults. The sequence of fault occurrence is an important propagation feature, which can reveal the propagation path and logical relationship of the fault in the system. For example, in a power system, a certain transmission line may be overloaded and tripped first, causing the voltage of the related bus to fluctuate, and then triggering the overcurrent of other lines, eventually leading to the sequential failure of multiple components. The time interval between adjacent faults is also an important part of the cascade fault propagation feature. Different time intervals may reflect the speed and mechanism of fault propagation.

[0085] 105. Construct a high-dimensional space classification boundary according to the cascade fault propagation feature set to classify the fault mode and obtain the fault diagnosis result.

[0086] In order to classify the fault mode, the cascade fault propagation feature set needs to be mapped to a high-dimensional space. Each cascade fault propagation feature is regarded as a dimension in the high-dimensional space, so that each fault sample can be represented by a point in the high-dimensional space. Then, through support vector machine, decision tree and other algorithms, the boundary that can distinguish different fault modes in the high-dimensional space is found. After the classification boundary in the high-dimensional space is constructed, the new fault sample is projected into the high-dimensional space, and according to the region it is located in, it is determined which fault mode it belongs to. It should be noted that the result of fault mode classification is the result of fault diagnosis.

[0087] In this embodiment, by doing so, the correlation strength between the corresponding variables of each device in the power system can be used to screen out the key feature vectors with correlation, and the propagation path and propagation speed of the key feature vectors can be obtained, so as to determine the cascade failure path feature set of the cascade failure that is likely to occur, and finally determine the specific failure mode according to the high-dimensional space boundary. Thus, the complex interaction relationship and failure propagation characteristics between devices in the power system can be comprehensively considered to accurately identify the cascade failure, thereby improving the safety and stability of the power system.

[0088] Please refer to Figure 2-1 , Figure 2-2 and Figure 2-3 , another embodiment of the fault diagnosis method of the power centralized communication control device in the application includes:

[0089] 201, obtaining operation data of the power centralized communication control device;

[0090] 202, performing sharding processing on the operation data to obtain a sharded data set;

[0091] 203, extracting a dynamic data set in the sharded data set;

[0092] 204, extracting frequency components and periodic characteristics in the dynamic data set by using wavelet transform;

[0093] 205, generating a frequency feature set based on the frequency components and the periodic characteristics.

[0094] Optionally, in this embodiment, since the operation data can be a large amount of continuous data stream, in order to facilitate subsequent processing, it can be divided into several segments according to certain rules. The sharding manner can be determined according to time interval, data size or other logic, and the overall operation data is divided into a sharded data set that is relatively independent and has certain characteristics. For example, the collected operation data is continuous voltage and current values recorded by seconds, and it is set to be sharded according to every 10 minutes as a time interval, such as the operation data from 9:00 to 9:10 in the morning as a shard, which contains the voltage and current values every second within the 10 minutes, forming a sharded data. In the sharded data set, some data can be relatively stable, and some can be dynamic data that changes significantly over time or other factors. The judgment of dynamic data can be based on statistical quantities such as variance and coefficient of variation of the sharded data. The variance is used to measure the dispersion degree of the data, and the larger the variance, the more violent the data fluctuation; the coefficient of variation is the ratio of the standard deviation to the mean, which eliminates the influence of the data magnitude and more accurately reflects the relative fluctuation degree of the data. By setting a certain threshold, the data with variance or coefficient of variation greater than the threshold is screened out from the sharded data set and determined as dynamic data.

[0095] After the dynamic data is acquired, the wavelet transform is used to decompose the dynamic data. By calculating the energy distribution of the wavelet coefficients, the frequency components of the signal can be determined. If the signal has periodicity, the coefficients after the wavelet transform will present regular distribution at certain scales and positions. By analyzing the distribution regularity, the periodicity feature can be extracted. Each frequency component corresponds to a dictionary item, which contains the frequency value, energy amplitude, position in the original data, and the like; the periodicity feature records the period length, data variation regularity in the period, and the like. By structuring and sorting the frequency components and the periodicity feature, the frequency feature set can be obtained.

[0096] 206、Calculate the conditional probability of the state association between all variables in the frequency feature set, and construct a complete conditional probability table;

[0097] 207、Determine the association strength between variables in the frequency feature set according to the conditional probability table;

[0098] 208、Determine the key feature vector as the variable combination whose association strength is greater than a preset association threshold;

[0099] 209、Take each device in the power system as a node, and take the key feature vector as an edge, to generate the interaction topology between different devices in the power system.

[0100] Optionally, in the embodiment, the frequency feature set contains multiple variables. To calculate the conditional probability of the state association between the variables, the probability of another variable being in a specific state is calculated when a certain state of one variable is known. The conditional probability table gives the probability dependence relationship between the variables, but this relationship needs to be further quantified into a specific numerical value to represent the association strength, so as to be further compared and analyzed, in which the information gain and mutual information can be used to determine the association strength. It should be noted that the information gain refers to the degree of reduction of uncertainty of one variable due to the knowledge of another variable. For discrete variables, the information gain can be obtained by calculating the change of entropy. The entropy is a measure of the uncertainty of a random variable. The greater the entropy value, the higher the uncertainty of the variable. Mutual information is an index for measuring the mutual dependence between two variables. It is equal to the information gain of one variable to another variable, and its value ranges from 0 to positive infinity. The greater the value, the higher the association strength between the two variables. By calculating the information gain and mutual information, the quantified association strength between the variables can be extracted from the conditional probability table.

[0101] The preset correlation threshold can be set by comprehensively considering various factors, such as stability requirements of the system, sensitivity to key features, and the like. When the correlation strength between variables is greater than the preset correlation threshold, it indicates that there is a strong dependent relationship between the variables, and at this time, the combination of the variables can be determined as a key feature vector. In the construction process of the interaction topology, since the power system is a complex network formed by a large number of devices connected to each other. Therefore, each device in the power system can be abstracted as a node, and the key feature vector is used as an edge to generate an interaction topology to intuitively show the internal relationship and interaction between devices. For example: the key feature vector {A, B} represents a strong correlation between the generator output frequency A and the transmission line frequency fluctuation amplitude B, and the key feature vector {B, C} represents a strong correlation between the transmission line frequency fluctuation amplitude B and the load device frequency response C. Then, in the generated interaction topology diagram, the generator, transformer, transmission line and load device are taken as nodes, an edge is drawn between the generator and the transmission line to represent the key feature vector {A, B}, and an edge is drawn between the transmission line and the load device to represent the key feature vector {B, C}.

[0102] 210. Calculate the propagation path of each key feature vector in the interaction topology according to the path reasoning algorithm;

[0103] 211. Calculate the propagation speed of each propagation path based on the propagation speed influencing factors and the corresponding assigned weights;

[0104] 212. Determine the propagation path with a propagation speed greater than a preset speed threshold as a candidate sub-path;

[0105] 213. Calculate the path coverage range of the candidate sub-path;

[0106] 214. Determine the candidate sub-path with a path coverage range greater than a preset coverage threshold as a candidate fault propagation path, to obtain a candidate fault propagation path set.

[0107] Optionally, in this embodiment, a path reasoning algorithm such as depth-first search, breadth-first search or Dijkstra algorithm can be used to calculate the propagation path of each key feature vector in the interaction topology. For example, if the interaction topology is a network with obvious hierarchical structure, breadth-first search can be preferred, which can explore nodes in hierarchical order and quickly find the shortest path from the starting point to other nodes, which is helpful for initially determining the range of possible impact of the fault. If the interaction topology is a more complex network with multiple paths with different weights, Dijkstra algorithm can be more optimal, which can quickly find the shortest path from the source node to all other nodes in the weighted graph. After obtaining the propagation path, the propagation speed can be calculated according to the propagation speed influencing factors, such as the interaction frequency between devices, data transmission delay, processing time, etc. A weight is assigned to each edge in the interaction topology, which represents the speed-related information of the feature vector propagating on this edge. The propagation speed is calculated according to the weight assigned to the edge and the combination of edges on the path. For example, if the sum of the weights of all edges on a path is small, the propagation speed on this path is relatively fast. At this time, the propagation speed can be calculated by a specific formula, such as taking the sum of the reciprocals of the weights of all edges on the path as the calculation result of the propagation speed. Then, by comparing the calculated propagation speed with the preset speed threshold, the path with faster propagation speed is selected to obtain the candidate sub-path. In the interaction topology graph, the path coverage range can be understood as the number of nodes, edges, or the system function modules, business areas covered by the candidate sub-path. Specifically, graph traversal algorithm can be used to calculate the path coverage range, that is, starting from the starting node, exploring as deeply as possible along a path, until it cannot continue or reaches the target node, then backtracking to the last node and continuing to explore other paths, recording the nodes and edges passed in the traversal process to determine the path coverage range. Finally, the path coverage range is compared with the preset coverage threshold to determine the candidate sub-path with wide path coverage range as the candidate fault propagation path.

[0108] 215. Constructing an initial state transition matrix according to the candidate fault propagation paths;

[0109] 216. Adjusting the conditional probability distribution in the initial state transition matrix using an evidence updating mechanism;

[0110] 217. Calculating the convergence index of the adjusted conditional probability distribution to obtain a convergence constraint result;

[0111] 218. Screening a target path set from the candidate fault propagation path set based on the convergence constraint result;

[0112] 219. Constructing a path feature matrix based on the target path set;

[0113] 220. Calculate the fault propagation probability of each target path according to the path feature matrix, and determine the cascading fault propagation feature set based on the fault propagation probability;

[0114] Optionally, in the embodiment, after the candidate fault propagation path is obtained, an initial state transition matrix is constructed according to the candidate fault propagation path. The initial probability value can be assigned according to historical fault data, expert experience or theoretical model. It can be understood that different operating states of each device are defined as rows and columns of the matrix, the state transition probability represents the possibility of changing from one state to another, and the sum of the probabilities of each row in the matrix is 1. The evidence updating mechanism can use Bayesian theory or other data-driven methods. As the power system continues to operate, new evidence such as real-time monitoring data and online detection results can reflect the current actual operating conditions and fault occurrence trends of each device. Using these new evidence, the probabilities in the initial state transition matrix are updated through the Bayesian formula, so that the probability distribution is more in line with the actual situation. For example, in a power system, real-time monitored data such as line load, device temperature, and voltage fluctuation can be used as evidence to update the probability. When the load of a line continuously exceeds the rated value, according to historical experience, the probability of failure of this line will increase, and at this time the state transition probability related to the failure of this line needs to be adjusted.

[0115] It should be noted that the convergence index can include trace distance, KL divergence, etc. The trace distance can measure the difference between two matrices by calculating the square root of the sum of the square of the difference between the elements of the two matrices. The KL divergence is used to measure the difference between two probability distributions. The conditional probability distribution is updated through multiple iterations, and the convergence index is calculated after each iteration. When the convergence index is less than a pre-set threshold, it is considered that the conditional probability distribution converges. After the conditional probability distribution converges, the probabilities corresponding to the candidate fault propagation paths are evaluated to select the target path set. Then, the path feature matrix is constructed according to the length of the target path, the importance of the nodes on the target path, the type of the devices on the target path, the time delay of the fault propagation, etc. Finally, the relationship between the fault propagation probability and the path feature is established by using algorithms such as support vector machine, random forest or grey prediction model, to calculate the fault propagation probability of each target path. Finally, the cascading fault propagation features are analyzed according to the fault propagation probability, path features, etc. For example: the common features of high probability paths are that the cascading fault starting points are all key devices in the power system, and the fault propagation time is short, so the cascading fault propagation feature set can be determined based on the above information.

[0116] 221. Extract the cascading fault propagation feature vector in the cascading fault propagation feature set;

[0117] 222. Use the cascaded fault propagation feature vector as input to the support vector machine to construct a high-dimensional classification boundary.

[0118] 223. Adjusting the margin of classification boundaries in high-dimensional space using kernel functions;

[0119] 224. Calculate the distance metric of the high-dimensional spatial classification boundary after the adjustment interval;

[0120] 225. Determine if the distance metric is less than the preset distance threshold. If so, proceed to step 226.

[0121] 226. The fault mode is determined to be a cascading fault.

[0122] Optionally, in this embodiment, algorithms such as the Small Porter Transform or Fourier Transform can be used to extract cascade fault propagation feature vectors from the cascade fault propagation feature set. These feature vectors can include node state change rates, time intervals, and parameter change rates. Since cascade fault propagation feature vectors may exhibit complex nonlinear relationships, it is difficult to find effective classification boundaries directly in the original low-dimensional space. Mapping them to a high-dimensional space using a Support Vector Machine (SVM) increases the linear separability of the data. Therefore, each cascade fault propagation feature vector can be used as input to the SVM, which constructs a classification boundary in the high-dimensional space based on these input data. This classification boundary is determined based on the inherent patterns and data distribution characteristics of the cascade fault propagation feature vectors.

[0123] After determining the high-dimensional classification boundary, the margin of the boundary needs to be adjusted by optimizing the kernel parameters. If the margin is too large, some data points that should be correctly classified may be misclassified into other categories, because an overly loose classification boundary cannot accurately capture the subtle differences between different failure modes. If the margin is too small, the classification model will become too complex and prone to overfitting. Specifically, the kernel parameters can be optimized based on cascading failure propagation features. By continuously adjusting the kernel parameters, the margin of the classification boundary can adapt to the distribution characteristics of the data, thereby better distinguishing the feature data of different failure modes and further improving the accuracy of classification. By calculating the distance from each cascading failure propagation feature vector to the classification boundary, a distance metric can be obtained, which can be achieved using methods such as Euclidean distance or Manhattan distance. This reflects the degree of fit between the current failure feature and the constructed cascading failure classification boundary, providing an important basis for subsequent failure mode judgment. The smaller the distance metric value, the more similar the failure situation represented by the cascading failure propagation feature vector is to the cascading failure mode; the larger the distance metric value, the greater the difference between the two. When the distance metric is greater than a preset distance threshold, the fault mode of the cascaded fault propagation feature vector can be identified as a cascaded fault mode.

[0124] Please refer to Figure 3 An embodiment of the fault diagnosis system of the power centralized communication control device in the application includes:

[0125] The acquisition unit 301 is configured to acquire operation data of the power centralized communication control device and extract a frequency feature set from the operation data.

[0126] The first calculation unit 302 is configured to calculate the correlation strength between variables in the frequency feature set by using a conditional probability table, filter out a key feature vector set, and generate an interaction topology between different devices in the power system.

[0127] The second calculation unit 303 is configured to calculate the propagation path and the propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and filter out a candidate fault propagation path set based on the propagation path and the propagation speed.

[0128] The determination unit 304 is configured to calculate the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set.

[0129] The classification unit 305 is configured to construct a high-dimensional space classification boundary according to the cascading fault propagation feature set, perform fault mode classification, and obtain a fault diagnosis result.

[0130] In the embodiment, the acquisition unit 301 acquires operation data of the power centralized communication control device and extracts a frequency feature set from the operation data. The first calculation unit 302 calculates the correlation strength between variables in the frequency feature set by using a conditional probability table, filters out a key feature vector set, and generates an interaction topology between different devices in the power system. The second calculation unit 303 calculates the propagation path and the propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and filters out a candidate fault propagation path set based on the propagation path and the propagation speed. The determination unit 304 calculates the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set. The classification unit 305 constructs a high-dimensional space classification boundary according to the cascading fault propagation feature set, performs fault mode classification, and obtains a fault diagnosis result. In this way, the correlation strength between variables corresponding to each device in the power system can be used to filter out key feature vectors with correlation, and the propagation path and the propagation speed of the key feature vectors can be obtained to determine a cascading fault path feature set in which cascading faults are likely to occur. Finally, the specific fault mode is determined according to the high-dimensional space boundary. Thus, the complex interaction relationship and the fault propagation characteristics between devices in the power system can be comprehensively considered to accurately identify cascading faults, thereby improving the safety and stability of the power system.

[0131] Another embodiment of the fault diagnosis system of the power centralized communication control device in the application includes:

[0132] The acquisition unit 301 is specifically configured to acquire operation data of the power centralized communication control device; perform fragmentation processing on the operation data to obtain a fragmented data set; extract a dynamic data set in the fragmented data set; extract frequency components and periodic characteristics in the dynamic data set by using wavelet transform; and generate a frequency feature set based on the frequency components and the periodic characteristics.

[0133] The first calculation unit 302 is specifically configured to calculate conditional probabilities of state correlations between all variables in the frequency feature set, to construct a complete conditional probability table; determine correlation strengths between variables in the frequency feature set according to the conditional probability table; determine a variable combination with a correlation strength greater than a preset correlation threshold as a key feature vector; and generate an interaction topology between different devices in the power system, with each device in the power system as a node and the key feature vector as an edge.

[0134] The second calculation unit 303 is specifically configured to calculate propagation paths of each key feature vector in the interaction topology according to a path reasoning algorithm; calculate propagation speeds of each propagation path based on propagation speed influencing factors and corresponding assigned weights; determine a propagation path with a propagation speed greater than a preset speed threshold as a candidate sub-path; calculate a path coverage range of the candidate sub-path; and determine a candidate sub-path with a path coverage range greater than a preset coverage threshold as a candidate fault propagation path, to obtain a candidate fault propagation path set.

[0135] The determination unit 304 is specifically configured to construct an initial state transition matrix according to the candidate fault propagation path; adjust conditional probability distributions in the initial state transition matrix by using an evidence updating mechanism; calculate a convergence index of the adjusted conditional probability distributions to obtain a convergence constraint result; screen out a target path set from the candidate fault propagation path set based on the convergence constraint result; construct a path feature matrix based on the target path set; calculate fault propagation probabilities of each target path according to the path feature matrix, and determine a cascading fault propagation feature set based on the fault propagation probabilities.

[0136] The classification unit 305 is specifically configured to extract a cascading fault propagation feature vector in the cascading fault propagation feature set; construct a high-dimensional space classification boundary by taking the cascading fault propagation feature vector as an input of a support vector machine; adjust a separation of the high-dimensional space classification boundary by using a kernel function; calculate a distance measure of the high-dimensional space classification boundary after the separation; and determine that the fault mode is a cascading fault if the distance measure is greater than a preset distance threshold.

[0137] In this embodiment, the functions of each unit are similar to those of steps 201 to 226 in the foregoing embodiments shown in Figure 2-1 , Figure 2-2 and Figure 2-3 , and will not be described here in detail.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0139] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0140] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0141] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0142] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that makes a contribution or the whole or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

Claims

1. A failure diagnosis method of a power concentration communication control device, characterized by, The method comprises the following steps: acquiring operation data of a power centralized communication control device, and extracting a frequency feature set from the operation data; calculating the correlation strength between variables in the frequency feature set by using a conditional probability table, screening out a key feature vector set, and generating an interaction topology between different devices in a power system; calculating the propagation path and propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and screening out a candidate fault propagation path set based on the propagation path and propagation speed; calculating the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set; constructing a high-dimensional space classification boundary based on the cascading fault propagation feature set to perform fault mode classification and obtain a fault diagnosis result; the method of calculating the propagation path and propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and screening out a candidate fault propagation path set based on the propagation path and propagation speed, comprises the following steps: calculating the propagation path of each key feature vector in the interaction topology according to a path reasoning algorithm; calculating the propagation speed of each propagation path based on the propagation speed influencing factors and corresponding assigned weights; determining the propagation path with a propagation speed greater than a preset speed threshold as a candidate sub-path; calculating the path coverage range of the candidate sub-path; determining the candidate sub-path with a path coverage range greater than a preset coverage threshold as a candidate fault propagation path to obtain a candidate fault propagation path set.

2. The failure diagnosis method of the power concentration communication control device according to claim 1, characterized by, the method of constructing a high-dimensional space classification boundary based on the cascading fault propagation feature set to perform fault mode classification and obtain a fault diagnosis result, comprises the following steps: extracting a cascading fault propagation feature vector in the cascading fault propagation feature set; taking the cascading fault propagation feature vector as the input of a support vector machine to construct a high-dimensional space classification boundary; adjusting the interval of the high-dimensional space classification boundary through a kernel function; calculating the distance measure of the high-dimensional space classification boundary after the interval is adjusted; judging whether the distance measure is less than a preset distance threshold; if yes, determining that the fault mode is a cascading fault.

3. The failure diagnosis method of the power concentration communication control device according to claim 1, characterized by, the method of calculating the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set, comprises the following steps: constructing an initial state transition matrix according to the candidate fault propagation path; adjusting the conditional probability distribution in the initial state transition matrix by using an evidence updating mechanism; calculating the convergence index of the adjusted conditional probability distribution to obtain a convergence constraint result; screening out a target path set from the candidate fault propagation path set based on the convergence constraint result; constructing a path feature matrix based on the target path set; calculating the fault propagation probability of each target path according to the path feature matrix, and determining a cascading fault propagation feature set based on the fault propagation probability.

4. The failure diagnosis method of the power concentration communication control device according to claim 1, characterized by, the method of calculating the correlation strength between variables in the frequency feature set by using a conditional probability table, screening out a key feature vector set, and generating an interaction topology between different devices in a power system, comprises the following steps: calculating the conditional probability of the state correlation between all variables in the frequency feature set to construct a complete conditional probability table; determine the correlation strength between variables in the frequency feature set according to the conditional probability table; determine a variable combination with a correlation strength greater than a preset correlation threshold as a key feature vector; generate an interaction topology between different devices in the power system, with each device in the power system as a node and the key feature vector as an edge.

5. The failure diagnosis method of the power concentration communication control device according to any one of claims 1 to 4, characterized by, The operation data of the power centralized communication control device is obtained, and a frequency feature set is extracted from the operation data, including: obtaining operation data of a power centralized communication control device; performing sharding processing on the operation data to obtain a sharded data set; extracting a dynamic data set from the sharded data set; extracting frequency components and periodic features in the dynamic data set using wavelet transform; generating a frequency feature set based on the frequency components and the periodic features.

6. A fault diagnosis system for a centralized power communication control device, characterized in that, including: an obtaining unit configured to obtain operation data of a power centralized communication control device and extract a frequency feature set from the operation data; a first calculation unit configured to calculate the correlation strength between variables in the frequency feature set using a conditional probability table, filter out a key feature vector set, and generate an interaction topology between different devices in the power system; a second calculation unit configured to calculate the propagation path and propagation speed of each key feature vector in the interaction topology according to a path reasoning algorithm, and filter out a candidate fault propagation path set based on the propagation path and propagation speed; a determination unit configured to calculate the fault propagation probability of the candidate fault propagation path to determine a cascading fault propagation feature set; a classification unit configured to construct a high-dimensional space classification boundary based on the cascading fault propagation feature set to perform fault mode classification and obtain a fault diagnosis result; The second calculation unit is specifically configured to: calculate the propagation path of each key feature vector in the interaction topology according to the path reasoning algorithm; calculate the propagation speed of each propagation path based on the propagation speed influencing factors and the corresponding assigned weights; determine the propagation path with a propagation speed greater than a preset speed threshold as a candidate sub-path; calculate the path coverage range of the candidate sub-path; determine the candidate sub-path with a path coverage range greater than a preset coverage threshold as a candidate fault propagation path to obtain a candidate fault propagation path set.

7. The failure diagnosis system for the power concentration communication control device according to claim 6, characterized by The classification unit is specifically configured to: extract a cascading fault propagation feature vector from the cascading fault propagation feature set; use the cascading fault propagation feature vector as the input of a support vector machine to construct a high-dimensional space classification boundary; adjust the interval of the high-dimensional space classification boundary through a kernel function; calculate the distance measure of the high-dimensional space classification boundary after adjusting the interval; determine whether the distance measure is less than a preset distance threshold; if yes, determine that the fault mode is a cascading fault.

8. The failure diagnosis system for the power concentration communication control device according to claim 6, characterized by The determination unit is specifically configured to: construct an initial state transition matrix according to the candidate fault propagation path; adjust the conditional probability distribution in the initial state transition matrix using an evidence updating mechanism; calculate the convergence index of the adjusted conditional probability distribution to obtain a convergence constraint result; filter out a target path set from the candidate fault propagation path set based on the convergence constraint result; constructing a path feature matrix based on the target path set; calculating a failure propagation probability of each target path according to the path feature matrix, and determining a cascading failure propagation feature set based on the failure propagation probability.

Images