A tensor decomposition-based fault diagnosis and self-healing method for merging units of smart substations

Abstract

The application discloses a kind of intelligent substation merging unit fault diagnosis and self-healing method based on tensor decomposition, comprising: from the sensor network of intelligent substation, the operation data of merging unit is collected in real time, and is handled, and the data after processing is organized into tensor form, and output multi-dimensional data;Tensor decomposition method is used to decompose the multi-dimensional data, extract fault characteristics, and detect the fault characteristics, obtain detection result;Determine the position and type of fault based on the detection result, start self-healing control strategy, realize the automatic repair of fault by taking corresponding measures.The application realizes the automatic repair of fault by adaptive control technology, enhances the self-healing ability of intelligent substation, improves the reliability and stability of system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a method for fault diagnosis and self-recovery of a merging unit in a smart substation based on tensor decomposition. BACKGROUND

[0002] The merging unit (MU) in a smart substation is responsible for collecting electrical quantity data from a mutual inductor and converting it into a digital signal for transmission to protection devices, measuring instruments and other intelligent devices. The performance and reliability of the MU are crucial to ensuring the stable operation of the smart substation system. However, in practical applications, the MU may face various faults such as data acquisition errors, time synchronization inaccuracies and data transmission interruptions. If these problems are not identified and resolved in a timely and effective manner, they may cause the protection device to malfunction or refuse to act, thereby threatening the safety and stability of the entire smart substation system.

[0003] In actual operation, the fault diagnosis of the MU faces many challenges. For example, during a routine maintenance check, the operation and maintenance team found that the current data of a specific line was fluctuating abnormally, but it was difficult to determine whether this was caused by a data acquisition problem of the MU or a fault in the line itself. In addition, the smart substation system has recorded a significant increase in the data packet loss rate, but it is uncertain whether this is due to a data transmission fault of the MU or the result of network congestion. Furthermore, the MU has been subjected to external interference, causing time synchronization to deviate and further leading to inaccurate data timestamps. These not only highlight the specific problems that the MU may encounter in actual operation, but also reflect the limitations of existing MU fault diagnosis, which mainly relies on pre-set rules and experience and is difficult to adapt to complex and variable power grid environments, especially when facing new and unknown fault types, its accuracy and efficiency need to be improved.

[0004] The prior art discloses a state evaluation calculation method suitable for a merging unit of a smart substation. The method can evaluate the state of the merging unit and system and calculate evaluation indexes according to the needs of the state maintenance work of the relay protection system of the smart substation. The method combines expert analysis and engineering actual needs to determine state quantities such as mercury fixed time delay drift, sampling value message sending time interval dispersion deviation, sampling value sequence number error and frame loss, zero drift offset, sampling value failure, environmental temperature and humidity, insulation condition and dust prevention, infrared temperature measurement, sampling precision, insulation condition and dust prevention, environmental temperature and humidity, and infrared temperature measurement for monitoring and evaluation, and establishes corresponding calculation methods and indexes, thereby providing an important guiding role for the state maintenance work of the relay protection system of the smart substation. However, there is no disclosure of how to solve the limitation of the existing fault diagnosis method in processing multi-dimensional and nonlinear data, nor is there disclosure of using tensor decomposition technology to solve the technical problem.

[0005] A kind of monitoring method of the state of intelligent substation merging unit, the running state and synchronization state of merging unit are monitored by fault recording device, comprising the following steps: step S1, fault recording device receives the sampling value message quantity sent by merging unit in each unit time, the signal of sampling value message sent by merging unit;And the sampling value message quantity forms first recording data, and the signal of sampling value message sent by merging unit forms second recording data;Step S2, first recording data and second recording data are formed into first waveform chart and second waveform chart after analysis processing respectively, and whether the running of merging unit is normal is judged by first waveform chart, whether the signal of sampling value message sent by merging unit is synchronized is judged by second waveform chart, so that the running state and synchronization state of merging unit are monitored in advance, the condition of merging unit can be understood in time, and the blind area of operation, maintenance, debugging and accident maintenance is avoided.However, it still does not disclose how to solve the technical problem of the limitation of existing fault diagnosis method when processing multi-dimensional, nonlinear data, and does not disclose the use of tensor decomposition technology to solve the technical problem.

[0006] The main problem of the prior art is that it mainly relies on preset rules and experience, and it is difficult to adapt to complex and variable power grid environment, especially when facing new unknown fault types, its accuracy and efficiency need to be improved. SUMMARY

[0007] To solve the above technical problems, the present application provides a kind of intelligent substation merging unit fault diagnosis and self-healing method based on tensor decomposition, which can effectively extract merging unit fault characteristics, improve the accuracy and efficiency of fault diagnosis, and realize the automatic repair of fault through adaptive control technology, so as to improve the reliability and stability of the whole power grid.

[0008] To achieve the above purpose, the present application provides a kind of intelligent substation merging unit fault diagnosis and self-healing method based on tensor decomposition, comprising:

[0009] Real-time collection of each operation data of merging unit from sensor network of intelligent substation is carried out, and processing is carried out, and the processed data is organized into tensor form, and multi-dimensional data is outputted;

[0010] The multi-dimensional data is decomposed using tensor decomposition method, the fault characteristics are extracted, and the fault characteristics are detected to obtain detection results;

[0011] The position and type of fault are determined based on the detection results, the self-healing control strategy is started, and the automatic repair of fault is realized by taking corresponding measures.

[0012] Preferably, the processing comprises normalizing the operation data; wherein the operation data comprises current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature.

[0013] Preferably, the method of organizing the processed data into a tensor form comprises:

[0014] The current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature collected per second are organized into a three-dimensional tensor, specifically:

[0015] T=(t ijk )

[0016] wherein T represents the tensor, t ijk represents an element in the tensor, i is the first dimension, j is the second dimension, and k is the third dimension.

[0017] wherein the dimensions in the three-dimensional tensor are:

[0018] Time: representing the time point of data collection;

[0019] Data type: including current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature;

[0020] Collection point: representing different data collection points.

[0021] Preferably, the method of decomposing the multi-dimensional data using a tensor decomposition method comprises:

[0022] T≈G×1A×2B×3C

[0023] wherein G is the core tensor, representing the low-rank structure of the data; A is the factor matrix of the first dimension, representing the characteristics of the time dimension; B is the factor matrix of the second dimension, representing the characteristics of the data type; C is the factor matrix of the third dimension, representing the characteristics of the collection point; ×1 represents the product along the first dimension; ×2 represents the product along the second dimension; and ×3 represents the product along the third dimension.

[0024] Preferably, the detecting the fault feature comprises:

[0025] determining whether there is a fault by calculating the residual tensor norm between the original tensor and the reconstructed tensor, and determining that there is a fault if the residual tensor norm exceeds a preset threshold;

[0026] wherein the reconstructed tensor is obtained by recombining the core tensor G and the factor matrices A, B and C of Tucker decomposition;

[0027] The method of calculating the residual tensor norm comprises:

[0028]

[0029] In the formula, ||Residual|| F For the residual tensor norm, Residual ijk Let Residual be the element of the residual tensor at time point i, data type j, and acquisition point k.

[0030] The residual tensor Residual is calculated as follows:

[0031]

[0032] In the formula, t ijk Let T be the element of the original tensor T at time point i, data type j, and acquisition point k. The reconstructed tensor obtained through Tucker decomposition Elements at the same position.

[0033] Preferably, determining the location of the fault based on the detection results includes:

[0034] Calculate the residual tensor norm for each combination of time points and acquisition points, and determine the initial fault location by finding the combination of time points and acquisition points with the largest norm.

[0035] By introducing the contributions of time and data collection point dimensions, the comprehensive residual norm is weighted to determine the final fault location after weighting.

[0036] Preferably, determining the type of fault based on the detection results includes:

[0037] Determine the data type dimension, and by calculating the residual tensor norm of each data type dimension, determine the data type with the largest norm as the initial fault type;

[0038] By introducing the contribution of the data type dimension, the data type norm is weighted to output the final fault type.

[0039] Preferably, the self-healing control strategy is as follows:

[0040] Increase data redundancy and adjust the time synchronization mechanism;

[0041] If the fault persists after adjusting the time synchronization mechanism, switch to the backup merging unit to isolate the faulty component and simultaneously attempt to repair the faulty component and restore its functionality.

[0042] Preferably, automatic fault repair is achieved by taking corresponding measures, including:

[0043] Increasing the bandwidth of data transmission, optimizing the data transmission protocol, increasing the data retransmission mechanism, optimizing the network topology, using a more accurate time synchronization protocol, checking the power supply line, increasing the current protection device, checking the power supply equipment, increasing the voltage stabilizer, checking the cooling system and increasing the environmental monitoring.

[0044] Compared with the prior art, the present application has the following advantages and technical effects:

[0045] (1) The present application can improve the accuracy of fault detection. By organizing multi-dimensional data into tensor form and using Tucker decomposition technology, the characteristics of data in multiple dimensions such as time, type and collection point can be fully captured, thereby significantly improving the accuracy of fault detection. Moreover, by calculating the residual tensor between the original tensor and the reconstructed tensor, the specific location of the fault can be accurately located. By analyzing the values of the residual tensor on different data types, the fault type can be accurately identified, including data transmission failure, packet loss rate failure, time synchronization failure, current failure, voltage failure and temperature failure, providing a scientific basis for subsequent fault handling.

[0046] (2) The present application improves the timeliness of fault detection. Through real-time data acquisition and processing mechanism, the system can quickly detect abnormalities when a fault occurs, timely issue an alarm, avoid further expansion of the fault, and ensure stable operation of the system.

[0047] (3) The present application can automatically adjust data transmission parameters, time synchronization mechanism, etc. according to the fault type and location through an efficient self-healing control strategy, reduce data loss and time synchronization error, and ensure the accuracy and integrity of the data. When a certain merging unit fails, the system can automatically switch to a backup merging unit to ensure the continuity of data acquisition and transmission, and avoid system downtime due to single-point failure. Through fault isolation and repair mechanism, the system can quickly isolate the faulty component and attempt to repair it, reducing maintenance time and cost, and improving the reliability and availability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The illustrative embodiments of the present application and their description serve to explain the present application. They do not, however, limit the present application unduly.

[0049] Figure 1 A flow chart of an intelligent substation merging unit fault diagnosis and self-healing method based on tensor decomposition according to an embodiment of the present application. DETAILED DESCRIPTION

[0050] It should be noted that the embodiments and features in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0051] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0052] The present application provides a tensor decomposition-based intelligent substation merging unit fault diagnosis and self-healing method, which comprises Figure 1 , including:

[0053] Real-time collection of various operation data of the merging unit from the sensor network of the intelligent substation, and processing, organizing the processed data into a tensor form, and outputting multi-dimensional data;

[0054] The multi-dimensional data is decomposed by using a tensor decomposition method, the fault features are extracted, and the fault features are detected to obtain a detection result;

[0055] Based on the detection result, the specific position and type of the fault are determined, the self-healing control strategy is started, and the automatic repair of the fault is realized by taking corresponding measures.

[0056] The present application effectively solves the limitations of the existing fault diagnosis method in processing multi-dimensional and nonlinear data by introducing the tensor decomposition technology, improves the accuracy and efficiency of fault diagnosis. At the same time, through the adaptive control technology, the automatic repair of the fault is realized, the self-healing ability of the intelligent substation is enhanced, and the reliability and stability of the system are improved.

[0057] Further, the processing includes standardizing the operation data; wherein the operation data includes current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature.

[0058] Specifically, the various operation data of the merging unit from the sensor network of the intelligent substation are real-time collected, including current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature. These data are standardized to ensure the consistency and accuracy of the data.

[0059] Current data standardization:

[0060]

[0061] Wherein, u I and σ I are the mean and standard deviation of the current data, respectively.

[0062] Voltage data standardization:

[0063]

[0064] where u V and σ V are the mean and standard deviation of the voltage data, respectively.

[0065] Similarly, the data transmission rate, packet loss rate, time synchronization error, and temperature data can be standardized.

[0066] Further, the standardized data is organized into a tensor form. The current, voltage, data transmission rate, packet loss rate, time synchronization error, and temperature information collected every second is organized into a three-dimensional tensor, with the first dimension representing time, the second dimension representing data type (current, voltage, data transmission rate, packet loss rate, time synchronization error, temperature), and the third dimension representing the collection point.

[0067] Specifically, the tensor is defined as:

[0068] T = (t ijk )

[0069] where T represents the tensor, t ijk represents an element in the tensor, i is the first dimension representing time, j is the second dimension representing data type (current, voltage, data transmission rate, packet loss rate, time synchronization error, temperature), and k is the third dimension representing the collection point. For example, t 123 represents the data of the second data type (such as voltage) collected at the third collection point in the first second.

[0070] Tensor structure:

[0071] First dimension (time): represents the time point of data collection, for example, collecting data once every second.

[0072] Second dimension (data type): includes current, voltage, data transmission rate, packet loss rate, time synchronization error, and temperature.

[0073] Third dimension (collection point): represents different data collection points, for example, sensors at different positions.

[0074] Through the above steps, data collection, standardization, and organization are completed, providing a high-quality data basis for subsequent tensor decomposition and fault detection.

[0075] Further, the method for decomposing the multi-dimensional data using tensor decomposition is:

[0076] T ≈ G × 1A × 2B × 3C

[0077] In the formula, G is a core tensor, representing the low-rank structure of the data; A is a factor matrix of the first dimension, representing the characteristics of the time dimension; B is a factor matrix of the second dimension, representing the characteristics of the data type; C is a factor matrix of the third dimension, representing the characteristics of the collection points; x1 represents the product along the first dimension; x2 represents the product along the second dimension; and x3 represents the product along the third dimension.

[0078] Specifically, the collected multi-dimensional data is decomposed by using a tensor decomposition technique, and a fault feature is extracted, and whether a fault exists is detected by calculating a residual tensor. Specifically, the tensor data is decomposed by using a tensor decomposition technique (Tucker decomposition) to extract a fault feature.

[0079] Through the above decomposition, for a 3-dimensional tensor T, it can be decomposed into a core tensor G and three factor matrices A, B and C.

[0080] Further, the fault feature is detected, including:

[0081] Whether a fault exists is determined by calculating the residual tensor norm between the original tensor and the reconstructed tensor, and if the residual tensor norm exceeds a preset threshold, it is determined that a fault exists.

[0082] Specifically, the reconstructed tensor is obtained by recombining the core tensor G and the factor matrices A, B and C of the Tucker decomposition.

[0083] The reconstruction process is:

[0084]

[0085] The reconstructed tensor is obtained by recombining the core tensor G and the factor matrices A, B and C of the Tucker decomposition.

[0086] The reconstructed tensor is obtained by low-rank approximation of the original tensor T, so some high-order information is lost, and this loss causes the reconstructed tensor to have differences with the original tensor T, so the following residual tensor can be calculated.

[0087] The calculation method of the residual tensor Residual is:

[0088]

[0089] In the formula, t ijk is an element of the original tensor T at time point i, data type j and collection point k, is the reconstructed tensor obtained by Tucker decomposition. is an element at the same position.

[0090] By subtracting the element in the corresponding position of the reconstructed tensor from each element t in the original tensor T ijk Subtracting the element in the corresponding position of the reconstructed tensor Obtaining the residual tensor Residual The residual tensor reflects the difference between the original data and the low-rank approximation, which may be caused by a fault. ijk The residual tensor reflects the difference between the original data and the low-rank approximation, which may be caused by a fault.

[0091] The method for calculating the norm of the residual tensor is:

[0092]

[0093] In the formula, ||Residual|| F is the norm of the residual tensor, Residual ijk is the element of the residual tensor Residual at time point i, data type j, and collection point k;

[0094] The norm result of the residual tensor is used to determine whether there is a fault. If ||Residual|| F exceeds a preset threshold ∈, it is considered that there is a fault. In this embodiment, if ∈ == 0.1, when ||Residual|| F > = 0.1, it is determined to be a fault.

[0095] This step extracts the fault features and detects the potential fault through Tucker decomposition and residual calculation, providing a basis for subsequent fault positioning and classification.

[0096] Further, based on the detection result, the specific position of the fault is determined, including:

[0097] Calculate the residual tensor norm of each time point and collection point combination, and determine the initial fault position by finding the time point and collection point combination with the largest norm;

[0098] Introducing the contribution degree of the time dimension and the collection point dimension, weighting the comprehensive residual norm, and determining the final fault position after weighting;

[0099] Specifically, the contribution degree of each factor vector is calculated:

[0100] Contribution degree calculation:

[0101]

[0102] Where, a r is the rth column of the factor matrix A, indicating the eigenvector of the time dimension, and i indicates the index of the time dimension.

[0103]

[0104] where b r is the rth column of factor matrix B, representing the eigenvector of data type dimension, j represents the index of data type dimension.

[0105]

[0106] where c r is the rth column of factor matrix C, representing the eigenvector of collection point dimension, k represents the index of collection point dimension.

[0107] The contribution degree plays a role in highlighting the more important factors in fault detection. If the contribution degree of a certain time point is high, even if the residual norm of this time point is not the largest, it may become the final determined fault location after weighting; similarly, if the contribution degree of a certain data type is high, even if its residual norm is not the largest, it may also become the final determined fault type after weighting.

[0108] In determining the fault location, the contribution degrees of time dimension and collection point dimension are introduced when calculating the comprehensive residual norm of each time point and collection point combination; in determining the fault type, the contribution degree of data type dimension is also introduced to weight the data type norm.

[0109] Specifically, determining the fault location comprises:

[0110] Determining the final fault location by calculating the residual tensor norm of each time point and collection point combination.

[0111] ①Calculating the residual tensor norm of each time point i and collection point k combination:

[0112]

[0113] where J represents the size of data type dimension, i.e. the number of data types.

[0114] The above formula calculates the square root of the sum of squares of residual tensor norms of all data types j for each time point i and collection point k, to obtain the comprehensive residual norm, which reflects the possibility of fault at a specific time point and collection point.

[0115] ②Determining the maximum norm to obtain the final fault location:

[0116]

[0117] The final fault location is determined by finding the combination of time point and collection point with the largest norm. This multi-dimensional comprehensive analysis method ensures the comprehensiveness and accuracy of fault detection, and provides more detailed and reliable guidance for subsequent self-healing control strategies.

[0118] ③Determine the final fault location after weighting:

[0119] By introducing the contribution of time dimension and collection point dimension, the comprehensive residual norm is weighted, which can more accurately reflect the importance of each time point and collection point in fault detection.

[0120] Weighted Combined Residual Norm(i,k)

[0121] = Combined Residual Norm(i,k) × (Contribution(a r )

[0122] × Contribution(c r ))

[0123] Wherein, Weighted Combined Residual Norm(i,k) is the weighted residual tensor norm of each time point and collection point combination. Combined Residual Norm(i,k) is the residual tensor norm of each time point and collection point combination, which reflects the possibility of fault at a specific time point and collection point. Contribution(a r ) is the contribution of factor vector a r of time dimension, which represents the importance of time dimension in fault detection. Contribution(c r ) is the contribution of factor vector c r of collection point dimension, which represents the importance of collection point dimension in fault detection.

[0124] The above formula weights the comprehensive residual norm by multiplying the contribution of time dimension and collection point dimension, which can highlight those time points and collection points that are more important in fault detection. Time points and collection points with high contribution will occupy a larger weight in the weighted norm, thus playing a greater role in determining the final fault location, improving the accuracy of fault location determination. For example, if the contribution of a certain time point is high, even if its residual norm is not the largest, it may become the final determined fault location after weighting.

[0125] Further, based on the detection result, the type of the fault is determined, comprising:

[0126] A data type dimension is determined, and by calculating a residual tensor norm of each data type dimension, a data type with the largest norm is determined as an initial fault type;

[0127] By introducing a contribution degree of the data type dimension, the data type norm is weighted, and a final fault type is output.

[0128] Specifically, ① a data type dimension is determined:

[0129]

[0130] Wherein, I represents the size of the time dimension, i.e. the number of time points. K represents the size of the collection point dimension, i.e. the number of collection points.

[0131] The above formula calculates the square root of the sum of the squares of the residual tensor norms of all time points i and collection points k for each data type j, to obtain a data type residual norm, so as to determine the data type with the largest norm as the fault type.

[0132] ② determine the maximum norm to obtain the final fault type:

[0133] Fault Type=argm j ax(Type Residual Norm(j))

[0134] By calculating the residual tensor norm of each data type, the data type with the largest norm is determined as the fault type.

[0135] In this embodiment, the specific fault types include:

[0136] Data transmission fault: if the residual tensor has the highest value in the data transmission rate dimension, it is determined as a data transmission fault.

[0137] Packet loss rate fault: if the residual tensor has the highest value in the packet loss rate dimension, it is determined as a packet loss rate fault.

[0138] Time synchronization fault: if the residual tensor has the highest value in the time synchronization error dimension, it is determined as a time synchronization fault.

[0139] Current fault: if the residual tensor has the highest value in the current dimension, it is determined as a current fault.

[0140] Voltage fault: if the residual tensor has the highest value in the voltage dimension, it is determined as a voltage fault.

[0141] Temperature fault: If the value of the residual tensor in the temperature dimension is the highest, it is determined as a temperature fault.

[0142] ③ Determine the final fault type after weighting:

[0143] By introducing the contribution of the data type dimension, the data type norm is weighted, which can more accurately reflect the importance of each data type in fault detection.

[0144] Weighted Combined Residual Norm(j)

[0145] = Combined Residual Norm(j) × Contribution(b r )

[0146] Where Weighted Combined Residual Norm(i,k) is the weighted residual tensor norm of each data type. Combined Residual Norm(j) is the residual tensor norm of each data type, reflecting the possibility of fault in a specific data type. Contribution(b r ) is the contribution of the factor vector b r of the data type dimension, indicating the importance of the data type dimension in fault detection.

[0147] The above formula weights the data type norm by multiplying the contribution of the data type dimension, which can highlight those more important data types in fault detection. Data types with high contribution will occupy a larger weight in the weighted norm, thus playing a greater role in the final determination of fault type, improving the accuracy of fault type determination. For example, if the contribution of a certain data type is high, even if its residual norm is not the largest, it may become the final determined fault type after weighting.

[0148] Further, the self-healing control strategy is:

[0149] Increase data redundancy, adjust time synchronization mechanism;

[0150] If the fault is still not solved after adjusting the time synchronization mechanism, switch to the backup merging unit, isolate the faulty component, and at the same time try to repair the faulty component to restore its function.

[0151] Specifically, after confirming the fault location and type, the self-healing control strategy is started, which realizes the automatic repair of the fault by adjusting the data transmission parameters, time synchronization mechanism, switching the backup merging unit, and fault isolation and repair, etc.

[0152] The method for increasing data redundancy is:

[0153] New Redundancy = Current Redundancy + δ

[0154] where δ is the increased redundancy. Increasing data redundancy can improve data reliability and integrity, reducing the risk of data loss.

[0155] In this embodiment, if the current redundancy is 1.25 and δ == 0.1, the new redundancy is 1.25 + 0.1 = 1.35.

[0156] After increasing the data redundancy, the system will proceed to the next step, which is to adjust the time synchronization mechanism to reduce time synchronization errors.

[0157] The method of adjusting the time synchronization mechanism is:

[0158] t new = t measured - α · Δt

[0159] where α is the adjustment coefficient and Δt is the time deviation.

[0160] In this embodiment, if the measured time t measured is 10.5 seconds, the reference time is 10.0 seconds, and the adjustment coefficient is 0.8, then Δt = 10.5 - 10 = 0.5 seconds, and the new time t new = 10.5 - 0.8 x 0.5 = 10.1 seconds. Adjusting the clock can ensure the accuracy of time synchronization and reduce failures caused by time errors.

[0161] Further, by taking appropriate measures, the automatic repair of the fault is realized, including:

[0162] Increasing the bandwidth of data transmission, optimizing the data transmission protocol, increasing the data retransmission mechanism, optimizing the network topology, using more accurate time synchronization protocol, checking the power supply line, increasing the current protection device, checking the power supply equipment, increasing the voltage stabilizer, checking the cooling system and increasing the environmental monitoring.

[0163] After adjusting the time synchronization mechanism, if the fault is still not solved, the system will proceed to the next step, which is to switch to the backup merging unit to ensure the continuity of data acquisition and transmission.

[0164] Specifically, after increasing the data redundancy and adjusting the time synchronization mechanism, if the fault still exists, the system will automatically switch to the backup merging unit. The specific method is as follows:

[0165] Detecting the fault: when the system detects that a merging unit has failed, it immediately starts the switching mechanism.

[0166] Switching process: The system automatically switches the data acquisition and transmission tasks to the backup merging unit, ensuring the continuity of data acquisition and transmission. This measure can avoid system downtime caused by single-point failure.

[0167] After switching to the backup merging unit, the system will further isolate and repair the fault to completely solve the problem. The specific methods include isolating the faulty component to ensure that it no longer affects the normal operation of the system, while trying to repair the faulty component to restore its function and ensure that the system can quickly return to normal state.

[0168] Additional measures for specific fault types:

[0169] After completing the above general self-healing measures, additional targeted measures are taken according to the fault type to further improve the reliability of the system. Specific measures include increasing data transmission bandwidth, optimizing data transmission protocol, increasing data retransmission mechanism, optimizing network topology, using more accurate time synchronization protocol, checking power supply line, increasing current protection device, checking power supply equipment, increasing voltage stabilizer, checking cooling system and increasing environmental monitoring, etc., as follows:

[0170] ① Data transmission failure: Increase data transmission bandwidth. Optimize data transmission protocol to reduce transmission delay and packet loss rate.

[0171] ② Packet loss failure: Increase data retransmission mechanism to ensure data integrity. Optimize network topology to reduce network congestion.

[0172] ③ Time synchronization failure: Use more accurate time synchronization protocol such as PTP (Precision Time Protocol). Regularly calibrate the clock to ensure the accuracy of time synchronization.

[0173] ④ Current failure: Check the power supply line to eliminate short circuit or open circuit problems. Increase current protection device to prevent overload.

[0174] ⑤ Voltage failure: Check the power supply equipment to ensure voltage stability. Increase voltage stabilizer to prevent voltage fluctuation.

[0175] ⑥ Temperature failure: Check the cooling system to ensure good heat dissipation of the equipment. Increase environmental monitoring to detect temperature abnormalities in a timely manner.

[0176] The self-healing control strategy in this embodiment is applicable to all types of faults mentioned above, and these measures include increasing data redundancy, adjusting time synchronization mechanism, switching to backup merging unit, and additional measures for specific fault types.

[0177] In this embodiment, if the system detects that the current data of the third acquisition point has abnormal fluctuations, and the time synchronization error is large. Through Tucker decomposition and residual calculation, it is determined that the third acquisition point has a fault, and the fault type is data transmission fault. The system first increases the redundancy of data transmission from 1.25 to 1.35 to reduce data loss. At the same time, adjust the clock, adjust the time deviation from 0.5 seconds to 0.1 seconds to ensure data synchronization. If these measures still cannot solve the problem, the system will switch to the standby MU. Finally, the system isolates the third acquisition point and tries to repair it.

[0178] In the above technical solutions, data collection and preprocessing provide a high-quality data basis for subsequent tensor decomposition and fault detection. Tensor decomposition and fault detection extract fault features through Tucker decomposition and determine whether there is a fault through residual tensor calculation, providing a basis for fault location and classification. Fault location and classification determine the specific location and type of the fault by analyzing the residual tensor, providing clear guidance for self-healing control strategies. The self-healing control strategy takes appropriate self-healing measures according to the fault location and type to ensure the stable operation of the system and the continuity of the data.

[0179] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for fault diagnosis and self-healing of a merging unit in a smart substation based on tensor decomposition, characterized in that, The application relates to a tensor-based fault detection method and device for smart substations. Real-time collection and processing of operation data of each unit of a real-time collection unit in a sensor network of a smart substation, organization of the processed data into a tensor form, and output of multi-dimensional data; Decomposition of the multi-dimensional data by using a tensor decomposition method, extraction of fault features, detection of the fault features, and obtaining of a detection result; The detection of the fault features comprises: Judging whether a fault exists by calculating a residual tensor norm between an original tensor and a reconstructed tensor, and judging that a fault exists if the residual tensor norm exceeds a preset threshold value; The reconstructed tensor is obtained by recombining a core tensor G and factor matrices A, B and C of Tucker decomposition; The method for calculating the residual tensor norm is: where ||Residual|| is the residual tensor norm F Residual ijk is the element of the residual tensor Residual at time point i, data type j, and acquisition point k; The method for calculating the residual tensor Residual is: where t ijk is an element of the original tensor T at time point i, data type j, and acquisition point k, is the reconstructed tensor obtained by Tucker decomposition is an element at the same position; Determining the position of the fault based on the detection result comprises: Calculating a residual tensor norm of each time point and collection point combination, finding a time point and collection point combination with the maximum norm, and determining an initial fault position; Introducing the contribution degrees of the time dimension and the collection point dimension, weighting the comprehensive residual norm, and determining a final fault position after weighting; Determining the fault position comprises: Calculating a residual tensor norm of each time point and collection point combination, and determining a final fault position; ①Calculating a residual tensor norm of each time point i and collection point k combination: Wherein, J represents the size of the data type dimension, that is, the number of data types; The above formula calculates the square root of the sum of squares of residual tensor norms of all data types j for each time point i and collection point k, obtains a comprehensive residual norm, and the norm value reflects the possibility of a fault at a specific time point and collection point; ②Determining the maximum norm to obtain a final fault position: Finding a time point and collection point combination with the maximum norm, and determining a final fault position; that is, the position of the fault is determined by comprehensively considering the time dimension and the collection point dimension, and the final fault position is determined by calculating a residual tensor norm of each time point and collection point combination and finding a combination with the maximum norm; ③Determining a final fault position after weighting: By introducing the contribution degrees of the time dimension and the collection point dimension, the comprehensive residual norm is weighted, and the importance of each time point and collection point in fault detection can be more accurately reflected: Weighted Combined Residual Norm(i,k) = Combined Residual Norm(i,k) x (Contribution(a r ) Contribution (c r )) wherein Weighted Combined Residual Norm(i, k) is the weighted residual tensor norm of each time point and acquisition point combination; Combined Residual Norm(i, k) is the residual tensor norm of each time point and acquisition point combination, reflecting the possibility of failure at a specific time point and acquisition point; Contribution(a r ) is the contribution degree of the factor vector a r of the time dimension, indicating the importance of the time dimension in failure detection; Contribution(c r ) is the contribution degree of the factor vector c r of the acquisition point dimension, indicating the importance of the acquisition point dimension in failure detection; Determining the type of the fault based on the detection result comprises: Determining a data type dimension, calculating a residual tensor norm of each data type dimension, determining a data type with the maximum norm as an initial fault type, and weighting the data type norm by introducing the contribution degree of the data type dimension, and outputting a final fault type; Based on the detection result, the position and type of the fault are determined, a self-recovery control strategy is started, and corresponding measures are taken to realize automatic repair of the fault. The processing comprises standardization processing of the operation data; wherein the operation data comprises current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature. 2.The tensor decomposition based intelligent substation merging unit fault diagnosis and self-healing method of claim 1, wherein, The method for organizing the processed data into a tensor form is: 3.The method of claim 2, wherein, ​ The current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature collected per second are organized into a three-dimensional tensor, specifically: T = (t ijk ) where T denotes a tensor, t ijk denotes an element in the tensor, i is the first dimension, j is the second dimension, and k is the third dimension. The dimensions in the three-dimensional tensor are: Time: representing the time point of data collection; Data type: including current, voltage, data transmission rate, packet loss rate, time synchronization error and temperature; Collection point: representing different data collection points. 4.The method of claim 1, wherein, The method for decomposing the multi-dimensional data by using the tensor decomposition method is: T≈G×1A×2B×3C In the formula, G is a core tensor, representing the low-rank structure of data; A is a factor matrix of the first dimension, representing the characteristics of the time dimension; B is a factor matrix of the second dimension, representing the characteristics of the data type; C is a factor matrix of the third dimension, representing the characteristics of the collection point; ×1 represents the product along the first dimension; ×2 represents the product along the second dimension; and ×3 represents the product along the third dimension.

5. The method of claim 1, wherein, The self-healing control strategy is: Increasing data redundancy and adjusting the time synchronization mechanism; If the fault is still not solved after adjusting the time synchronization mechanism, switching to a backup merging unit to isolate the faulty component and simultaneously attempting to repair the faulty component to restore the component function. 6.The method of claim 5, wherein, By taking corresponding measures, automatic repair of the fault is realized, including: Increasing the bandwidth of data transmission, optimizing the data transmission protocol, increasing the data retransmission mechanism, optimizing the network topology structure, using a more accurate time synchronization protocol, checking the power supply circuit, increasing the current protection device, checking the power supply equipment, increasing the voltage stabilizer, checking the cooling system and increasing the environmental monitoring.

Images