Communication fault diagnosis method and device based on power system, equipment and medium

Abstract

The invention discloses a communication fault diagnosis method and device based on a power system, equipment and a medium. The method comprises the following steps: S1, collecting and outputting multi-source data; s2, preprocessing the multi-source data; s3, performing topological constraint attention mechanism weighting processing on the data matrix; s4, performing dynamic threshold calculation on the fusion feature matrix; s5, positioning the abnormal node list; s6, calculating to obtain a posterior probability; s7, matching the rule base to obtain a standardized disposal instruction; according to the method, the communication traffic is collected through the compressed sensing collection algorithm, the line impedance parameters are integrated into the observation matrix in combination with the topological parameters, key node data are collected preferentially, and the effective data acquisition rate is increased; the accuracy of switch false alarm identification is improved by outputting a time-space unified fusion feature matrix; the computing resources are focused to the abnormal nodes and neighborhoods thereof through the abnormal node mask matrix, so that the positioning precision of the aged equipment is improved, and the judgment omission rate caused by switch aging is prevented from being improved.

Description

Technical Field

[0001] This invention relates to the field of communication fault diagnosis technology for power systems, specifically to communication fault diagnosis methods, devices, equipment, and media based on power systems. Background Technology

[0002] Communication fault diagnosis in power systems is a crucial aspect of ensuring the safe and stable operation of power systems. As the "neural network" of the power system, the communication system is responsible for transmitting critical information such as control commands and monitoring data. Faults in this system can lead to serious consequences such as grid dispatch failures and equipment malfunctions. However, existing methods for diagnosing communication faults in power systems still have certain shortcomings.

[0003] Existing methods for diagnosing communication faults in power systems process electrical quantities, communication quantities, and environmental quantities independently, leading to data asynchrony in time and space. This results in a high rate of critical data loss when the signal-to-noise ratio drops sharply under severe weather conditions.

[0004] In existing power system communication fault diagnosis methods, the data fusion process and weighted average method ignore the physical correlation between features, and the identification rate of complex faults such as switch loose connections is insufficient.

[0005] In existing power system communication fault diagnosis methods, the standard GCN uses a fixed adjacency matrix, which cannot adapt to dynamic topology changes in the distribution network, and the missed detection rate increases due to switch aging.

[0006] Therefore, it is essential to design communication fault diagnosis methods, devices, equipment, and media based on power systems. Summary of the Invention

[0007] The purpose of this invention is to provide a communication fault diagnosis method, device, equipment, and medium based on power systems, in order to solve the problems in existing power system communication fault diagnosis methods, such as the independent processing of electrical quantities, communication quantities, and environmental quantities, which leads to data spatiotemporal asynchrony, a large loss rate of key data when the signal-to-noise ratio drops sharply under severe weather, the neglect of physical correlation between features in the data fusion process by the weighted average method, insufficient identification rate of complex faults such as switch loose connections, and the inability of the standard GCN to adapt to dynamic topology changes in the distribution network due to the use of a fixed adjacency matrix, as well as the high missed detection rate caused by switch aging.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a communication fault diagnosis method based on a power system, comprising:

[0009] S1. Acquire and output multi-source data, including electrical quantities, communication quantities, environmental quantities, and topology data, where the topology data includes switch status and impedance parameters;

[0010] S2. Preprocess the collected multi-source data and output a standardized data matrix;

[0011] S3. Perform topological constraint attention mechanism weighted processing on the data matrix to output a spatiotemporally unified fusion feature matrix;

[0012] S4. Perform dynamic threshold calculation on the fused feature matrix, and output a list of abnormal nodes for abnormal intervals that continuously exceed the threshold.

[0013] S5. Locate the abnormal node list and output the probability of the fault interval;

[0014] S6. Input the fault interval probability into the Bayesian network for calculation to obtain the posterior probability;

[0015] S7. Based on the posterior probability, the matching rule base is used to obtain standardized processing instructions.

[0016] As a further technical solution of the present invention, in step S1, the electrical quantities collected by the electricity meter include: voltage, current, and harmonics; the environmental quantities collected by the meteorological sensor network include: temperature and humidity, meteorological data, and equipment temperature; the switching status and impedance parameters of the topology data are collected through remote signaling; and the communication data is collected through a compressed sensing acquisition algorithm, including: signal-to-noise ratio, bit error rate, and message delay. The compressed sensing acquisition algorithm is as follows:

[0017]

[0018] Where φ is the constructed observation matrix, ψ is the wavelet basis matrix, and F50 is the power frequency notch filter. It is the row normalized vector of the impedance-weighted adjacency matrix A.

[0019] y = φ@x

[0020] Where y is the compressed sampled data and x is the input original communication signal.

[0021] As a further technical solution of the present invention, the impedance-weighted adjacency matrix A is obtained in the following manner:

[0022] Suppose the distribution network has N nodes, and define the switch state matrix:

[0023] S∈{0,1} N×N

[0024]

[0025] Impedance parameter matrix:

[0026] Z∈R N×N

[0027] Z ij =Equivalent impedance between nodes i and j (Ω)

[0028] Impedance-weighted adjacency matrix formula:

[0029]

[0030] Where ε = 10 -5 To prevent small constants from being divided by zero, when the switch is closed (S ij =1), the weight is inversely proportional to the impedance, when the switch is open (S ij =0), the weight is forced to zero.

[0031] As a further technical solution of the present invention, in step S2,

[0032] Electrical quantity processing: A standard 50Hz notch filter is applied to the original voltage and current waveforms to eliminate fundamental frequency interference. The effective values ​​(RMS) of voltage and current are calculated using a 200ms time window. Harmonic distortion rate (THD) parameters are extracted simultaneously. The output electrical quantity matrix is: N×3, with N nodes. Each row includes voltage RMS, current RMS, and THD.

[0033] Traffic processing: The inherent transmission delay of the device is subtracted from the measured delay to achieve time synchronization calibration of control commands. Hamming code forward error correction technology is used to automatically repair bit errors in communication messages, outputting a traffic matrix: N×3, with N nodes, each row including the calibrated delay τ. cor Signal-to-noise ratio (SNR) and Quality of Service (QoS);

[0034] Environmental quantity processing: Perform a 5-point moving average on the parameters with a sampling interval of 1 second to suppress measurement noise, map the original values ​​to the [0,1] interval, obtain the wind speed index and humidity index, calculate the comprehensive meteorological influence coefficient: meteorological coefficient = 60% wind speed index + 30% temperature index + 10% humidity index, output environmental quantity matrix: N×3, N nodes, each row includes wind speed index, temperature, and meteorological coefficient;

[0035] Topology data processing: Compare the voltage difference between adjacent nodes. If it exceeds 5% of the rated voltage, the switch is determined to be actually open. If the line power is less than the threshold, the switch is determined to be loosely connected. Based on the verified switch status and preset impedance parameters, construct an impedance-weighted adjacency matrix of N×N.

[0036] As a further technical solution of the present invention, in step S3, the data are first subjected to feature grouping and encoding:

[0037] Electrical characteristic coding:

[0038] E e =ReLU(W e ·[V rms U rms ,THD]+b e )

[0039] Among them W e V is the electrical feature weight matrix. rms I is the effective value of the voltage. rms Here, b represents the effective value of the current, THD represents the total harmonic distortion, and b represents the effective value of the current. e The electrical characteristic bias term is ReLU, which is the corrected linear unit.

[0040] Communication feature encoding:

[0041] E c =σ(W c ·[τ cor [,SNR,QoS]+b c )

[0042] Where, τ cor The transmission delay after calibration is given by SNR, which is the signal-to-noise ratio, QoS is the communication quality level, and σ is the Sigmoid function.

[0043] Environmental feature coding:

[0044] E env =tanh(W env ·[W norm ,T,W weather ]+b env )

[0045] Where, x env This is an environmental feature vector containing the normalized wind speed W. norm Temperature T, meteorological coefficient W weather b env It is the bias vector;

[0046] Then, a topological constraint attention mechanism is introduced to calculate the attention weights:

[0047]

[0048] Where, α i A represents the modal weighting coefficient. i is the node topology influence factor, and v is the learnable parameter vector;

[0049] Then, a weighted fusion of features is performed:

[0050] H = α e E e +α c E c +α env E env

[0051] Where H is the output fused feature matrix, and α vector is the feature weight distribution. envFor meteorological adaptive mechanisms:

[0052]

[0053] Among them, W weather The weather impact coefficient is 0.7, which is the threshold for severe weather.

[0054] As a further technical solution of the present invention, in step S4, the dynamic threshold is calculated as follows:

[0055] T i =μ i +k·σ·(1+αW)

[0056] Where, μ i Let σ be the historical feature mean of node i. i Let be the historical standard deviation of node i, k be the sensitivity coefficient, and α be the meteorological sensitivity factor;

[0057] Anomaly detection rules:

[0058]

[0059] Output the list of abnormal nodes Γ={i|Alert i =1}.

[0060] As a further technical solution of the present invention, in step S5, the abnormal node set Γ, the fused feature matrix H, the impedance-weighted adjacency matrix A are input, and the localization algorithm is as follows:

[0061] Construct the abnormal node mask matrix M:

[0062] M∈{0,1} N×N

[0063]

[0064] By focusing computational resources on the abnormal node and its neighborhood through the abnormal node mask matrix M, mask graph convolution calculation is performed:

[0065] H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) )

[0066] Where l is the number of graph convolutional layers, and D is the degree matrix D ii =∑ j A ij σ is the GELU activation function, W (l) These are trainable weights;

[0067] Loss function:

[0068]

[0069] Among them, L CE For cross-entropy loss, ||·|| Γ To mitigate topology reconstruction errors, feature consistency is constrained to prevent deviations in localization results.

[0070] Output the faulty device / line ID and the probability P(Interval) of the fault interval. k )distributed:

[0071]

[0072] Where, P(Interval) i Let N be the failure probability of node i. (i) Let be the neighboring points of node i.

[0073] As a further technical solution of the present invention, in step S6, the Bayesian network is:

[0074] Prior probability calculation:

[0075]

[0076] The lightning density is from the meteorological bureau's API, and 30 is the upper limit set for the operating years.

[0077] Evidence likelihood calculation:

[0078] P(Evidence|Cause)=P(R x |Cause)×P(Interval|Cause)

[0079] Where R x Evidence variables include electrical evidence, communication evidence, environmental evidence, and spatial evidence. Electrical evidence comes from sudden changes in voltage / current; communication evidence comes from bit error rate / message delay; environmental evidence comes from lightning / wind speed; and spatial evidence comes from the probability P(Interval) of the fault interval. k ), and perform Bayesian inference to obtain the posterior probability:

[0080] P(Cause|Evidence)∝P(Cause)×P(Evidence|Cause)×P(Interval|Cause)

[0081] Where P(Interval|Cause) is the weight of the fault interval.

[0082] As a further technical solution of the present invention, in step S7, the posterior probability analysis results in S6 are called from the rule base to match and the disposal action instruction is output.

[0083] As a further technical solution of the present invention, the rule base adopts a hierarchical decision logic design, wherein:

[0084] The L1 safety isolation layer is triggered when the fault probability is greater than 90% and the current surge is greater than 300%. The response is to immediately trip the circuit breaker remotely and block the reclosing mechanism to prevent equipment damage and the escalation of the accident.

[0085] The L2 emergency recovery layer is triggered when the probability of a lightning strike is greater than 60% and the wind speed is greater than level 8. The response is to activate drones for inspection, isolate the affected area, and then transfer power to the affected area. This layer is used for rapid power restoration in severe weather conditions.

[0086] L3 planned maintenance layer, the trigger condition is aging probability > 70% + load rate < 50%, the handling action is planned power outage replacement, load transfer, and solving the problems of insufficient switches and aging.

[0087] Secondly, a power system device is provided, comprising a data acquisition module, a data processing module, and an instruction module. The data acquisition module is capable of acquiring data from multiple sources. The data processing module is capable of processing the acquired data in steps S2 to S6, and the instruction module is capable of issuing processing instructions based on the data processing results.

[0088] Thirdly, an electronic device is provided, including a processor, a memory, and a computer program stored in the memory, wherein when the computer program is executed by the processor, it implements the aforementioned communication fault diagnosis method based on a power system.

[0089] Fourthly, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the aforementioned communication fault diagnosis method based on a power system.

[0090] Compared with existing technologies, the beneficial effects of this power system-based communication fault diagnosis method, device, equipment, and medium are:

[0091] By using compressed sensing acquisition algorithms to collect communication data and incorporating line impedance parameters into the observation matrix in conjunction with topology parameters, data from key nodes is collected first, thereby improving the effective data acquisition rate in scenarios such as heavy rain and channel degradation.

[0092] By applying a topological constraint attention mechanism to the data matrix for weighted processing, a spatiotemporally unified fusion feature matrix is ​​output, thereby improving the accuracy of switch false alarm identification.

[0093] By focusing computational resources on abnormal nodes and their neighborhoods through the abnormal node mask matrix and performing mask map convolution calculations, the positioning accuracy of aging equipment can be improved, and the missed detection rate can be avoided due to switch aging. Attached Figure Description

[0094] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0095] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0096] Please see the appendix Figure 1 An embodiment of the present invention provides a communication fault diagnosis method based on a power system, comprising:

[0097] S1. Acquire and output multi-source data, including electrical quantities, communication quantities, environmental quantities, and topology data, where the topology data includes switch status and impedance parameters;

[0098] Electrical quantities collected using electricity meters include: voltage, current, and harmonics; environmental quantities collected using a meteorological sensor network include: temperature and humidity, meteorological data, and equipment temperature. The switch status and impedance parameters of the topology data are acquired via remote signaling. Communication data is collected using a compressed sensing algorithm, including: signal-to-noise ratio, bit error rate, and message delay. The compressed sensing algorithm is as follows:

[0099]

[0100] Where φ is the constructed observation matrix, ψ is the wavelet basis matrix, and F50 is the power frequency notch filter. It is the row normalized vector of the impedance-weighted adjacency matrix A.

[0101] y = φ@x

[0102] Where y is the compressed sampled data and x is the input original communication signal;

[0103] The impedance-weighted adjacency matrix A is obtained in the following way:

[0104] Suppose the distribution network has N nodes, and define the switch state matrix:

[0105] S∈{0,1} N×N

[0106]

[0107] Impedance parameter matrix:

[0108] Z∈R N×N

[0109] Z ij =Equivalent impedance between nodes i and j (Ω)

[0110] Impedance-weighted adjacency matrix formula:

[0111]

[0112] Where ε = 10 -5 To prevent small constants from being divided by zero, when the switch is closed (S ij =1), the weight is inversely proportional to the impedance, when the switch is open (S ij =0), the weight is forced to zero;

[0113] S2. Preprocess the collected multi-source data and output a standardized data matrix;

[0114] Electrical quantity processing: A standard 50Hz notch filter is applied to the original voltage and current waveforms to eliminate fundamental frequency interference. The effective values ​​(RMS) of voltage and current are calculated using a 200ms time window. Harmonic distortion rate (THD) parameters are extracted simultaneously. The output electrical quantity matrix is: N×3, with N nodes. Each row includes voltage RMS, current RMS, and THD.

[0115] Traffic processing: The inherent transmission delay of the device is subtracted from the measured delay to achieve time synchronization calibration of control commands. Hamming code forward error correction technology is used to automatically repair bit errors in communication messages, outputting a traffic matrix: N×3, with N nodes, each row including the calibrated delay τ. cor Signal-to-noise ratio (SNR) and Quality of Service (QoS);

[0116] Environmental quantity processing: Perform a 5-point moving average on the parameters with a sampling interval of 1 second to suppress measurement noise, map the original values ​​to the [0,1] interval, obtain the wind speed index and humidity index, calculate the comprehensive meteorological influence coefficient: meteorological coefficient = 60% wind speed index + 30% temperature index + 10% humidity index, output environmental quantity matrix: N×3, N nodes, each row includes wind speed index, temperature, and meteorological coefficient;

[0117] Topology data processing: Compare the voltage difference between adjacent nodes. If it exceeds the rated voltage by 5%, it is determined that the switch is actually open. If the line power is less than the threshold, it is determined that the switch is loose. Based on the verified switch status and preset impedance parameters, construct an impedance-weighted adjacency matrix of N×N.

[0118] S3. Perform topological constraint attention mechanism weighted processing on the data matrix to output a spatiotemporally unified fusion feature matrix;

[0119] First, feature grouping and encoding are performed on each data point:

[0120] Electrical characteristic coding:

[0121] E e =ReLU(W e ·[V rms U rms ,THD]+b e )

[0122] Among them W e V is the electrical feature weight matrix. rms I is the effective value of the voltage. rms Here, b represents the effective value of the current, THD represents the total harmonic distortion, and b represents the effective value of the current. e The electrical characteristic bias term is ReLU, which is the corrected linear unit.

[0123] Communication feature encoding:

[0124] E c =σ(W c ·[τ cor [,SNR,QoS]+b c )

[0125] Where, τ cor The transmission delay after calibration is given by SNR, which is the signal-to-noise ratio, QoS is the communication quality level, and σ is the Sigmoid function.

[0126] Environmental feature coding:

[0127] E env =tanh(W env ·[W norm ,T,W weather ]+b env )

[0128] Where, x env This is an environmental feature vector containing the normalized wind speed W. norm Temperature T, meteorological coefficient W weather b env It is the bias vector;

[0129] Then, a topological constraint attention mechanism is introduced to calculate the attention weights:

[0130]

[0131] Where, α i A represents the modal weighting coefficient. i is the node topology influence factor, and v is the learnable parameter vector;

[0132] Then, a weighted fusion of features is performed:

[0133] H = α e E e +αc E c +α env E env

[0134] Where H is the output fused feature matrix, and α vector is the feature weight distribution. env For meteorological adaptive mechanisms:

[0135]

[0136] Among them, W weather This is the weather impact coefficient, with 0.7 being the severe weather threshold.

[0137] S4. Perform dynamic threshold calculation on the fused feature matrix, and output a list of abnormal nodes for abnormal intervals that continuously exceed the threshold.

[0138] The dynamic threshold is calculated as follows:

[0139] T i =μ i +k·σ·(1+αW)

[0140] Where, μ i Let σ be the historical feature mean of node i. i Let be the historical standard deviation of node i, k be the sensitivity coefficient, and α be the meteorological sensitivity factor;

[0141] Anomaly detection rules:

[0142]

[0143] Output the list of abnormal nodes Γ={i|Alert i =1};

[0144] S5. Locate the abnormal node list and output the probability of the fault interval;

[0145] Given the set of abnormal nodes Γ, the fused feature matrix H, and the impedance-weighted adjacency matrix A, the localization algorithm is as follows:

[0146] Construct the abnormal node mask matrix M:

[0147] M∈{0,1} N×N

[0148]

[0149] By focusing computational resources on the abnormal node and its neighborhood through the abnormal node mask matrix M, mask graph convolution calculation is performed:

[0150] H (l+1) =σ(D -1 / 2 AD -1 / 2 H(l) W (l) )

[0151] Where l is the number of graph convolutional layers, and D is the degree matrix D ii =∑ j A ij σ is the GELU activation function, W (l) These are trainable weights;

[0152] Loss function:

[0153]

[0154] Among them, L CE For cross-entropy loss, ||·|| Γ To mitigate topology reconstruction errors, feature consistency is constrained to prevent deviations in localization results.

[0155] Output the faulty device / line ID and the probability P(Interval) of the fault interval. k )distributed:

[0156]

[0157] Where, P(Interval) i Let N be the failure probability of node i. (i) Let be the neighboring points of node i;

[0158] S6. Input the fault interval probability into the Bayesian network for calculation to obtain the posterior probability;

[0159] Bayesian networks are:

[0160] Prior probability calculation:

[0161]

[0162] The lightning density is from the meteorological bureau's API, and 30 is the upper limit set for the operating years.

[0163] Evidence likelihood calculation:

[0164] P(Evidence|Cause)=P(R x |Cause)×P(Interval|Cause)

[0165] Where R x Evidence variables include electrical evidence, communication evidence, environmental evidence, and spatial evidence. Electrical evidence comes from sudden changes in voltage / current; communication evidence comes from bit error rate / message delay; environmental evidence comes from lightning / wind speed; and spatial evidence comes from the probability P(Interval) of the fault interval. k), and perform Bayesian inference to obtain the posterior probability:

[0166] P(Cause|Evidence)∝P(Cause)×P(Evidence|Cause)×P(Interval|Cause)

[0167] Where P(Interval|Cause) is the weight of the fault interval;

[0168] S7. Based on the posterior probability, the standardized processing instructions are obtained by matching the rule base.

[0169] Call the rule base to match the posterior probability analysis results in S6, and output the action command.

[0170] The rule base adopts a hierarchical decision logic design, in which:

[0171] The L1 safety isolation layer is triggered when the fault probability is greater than 90% and the current surge is greater than 300%. The response is to immediately trip the circuit breaker remotely and block the reclosing mechanism to prevent equipment damage and the escalation of the accident.

[0172] The L2 emergency recovery layer is triggered when the probability of a lightning strike is greater than 60% and the wind speed is greater than level 8. The response is to activate drones for inspection, isolate the affected area, and then transfer power to the affected area. This layer is used for rapid power restoration in severe weather conditions.

[0173] L3 planned maintenance layer, the trigger condition is aging probability > 70% + load rate < 50%, the handling action is planned power outage replacement, load transfer, and solving the problems of insufficient switches and aging.

[0174] One embodiment of the present invention provides a power system device, comprising a data acquisition module, a data processing module, and an instruction module. The data acquisition module is capable of acquiring data from multiple sources. The data processing module is capable of processing the acquired data in steps S2 to S6. The instruction module is capable of issuing processing instructions based on the data processing results.

[0175] One embodiment of the present invention is an electronic device comprising a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements a communication fault diagnosis method based on a power system.

[0176] One embodiment of the present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements a communication fault diagnosis method based on a power system.

[0177] In summary, the communication fault diagnosis method, device, equipment, and medium based on power systems provided by this invention collect communication data through compressed sensing algorithms and integrate line impedance parameters into the observation matrix by combining topology parameters. This prioritizes the collection of data from key nodes, improving the effective data acquisition rate in scenarios with channel degradation such as heavy rain. Furthermore, by introducing a topology-constrained attention mechanism to weight the data matrix and output a spatiotemporally unified fusion feature matrix, the accuracy of identifying false alarms from switches is improved. Finally, by focusing computational resources on abnormal nodes and their neighborhoods through an abnormal node mask matrix and performing mask map convolution calculations, the positioning accuracy of aging equipment can be improved, avoiding increased false alarm rates due to switch aging.

[0178] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A communication fault diagnosis method based on power systems, including: S1. Acquire and output multi-source data, including electrical quantities, communication quantities, environmental quantities, and topology data, where the topology data includes switch status and impedance parameters; S2. Preprocess the collected multi-source data and output a standardized data matrix; S3. Perform topological constraint attention mechanism weighted processing on the data matrix to output a spatiotemporally unified fusion feature matrix; S4. Perform dynamic threshold calculation on the fused feature matrix, and output a list of abnormal nodes for abnormal intervals that continuously exceed the threshold. S5. Locate the abnormal node list and output the probability of the fault interval; S6. Input the fault interval probability into the Bayesian network for calculation to obtain the posterior probability; S7. Based on the posterior probability, the matching rule base is used to obtain standardized processing instructions.

2. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S1, the electrical quantities collected using electricity meters include: voltage, current, and harmonics; the environmental quantities collected using a meteorological sensor network include: temperature and humidity, meteorological data, and equipment temperature; the switch status and impedance parameters of the topology data are collected via remote signaling; and communication data is collected using a compressed sensing acquisition algorithm, including: signal-to-noise ratio, bit error rate, and message delay. The compressed sensing acquisition algorithm is as follows: Where φ is the constructed observation matrix, ψ is the wavelet basis matrix, and F50 is the power frequency notch filter. , is the row normalized vector of the impedance-weighted adjacency matrix A. y = φ@x Where y is the compressed sampled data and x is the input original communication signal.

3. The communication fault diagnosis method based on power systems according to claim 2, characterized in that: The impedance-weighted adjacency matrix A is obtained in the following way: Suppose the distribution network has N nodes, and define the switch state matrix: S∈{0,1} N×N Impedance parameter matrix: Z∈R N×N Z ij =Equivalent impedance between nodes i and j (Ω) Impedance-weighted adjacency matrix formula: Where ε = 10 -5 To prevent small constants from being divided by zero, when the switch is closed (S ij =1), the weight is inversely proportional to the impedance, when the switch is open (S ij =0), the weight is forced to zero.

4. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S2 Electrical quantity processing: A standard 50Hz notch filter is applied to the original voltage and current waveforms to eliminate fundamental frequency interference. The effective values ​​(RMS) of voltage and current are calculated using a 200ms time window. Harmonic distortion rate (THD) parameters are extracted simultaneously. The output electrical quantity matrix is: N×3, with N nodes. Each row includes voltage RMS, current RMS, and THD. Traffic processing: The inherent transmission delay of the device is subtracted from the measured delay to achieve time synchronization calibration of control commands. Hamming code forward error correction technology is used to automatically repair bit errors in communication messages, outputting a traffic matrix: N×3, with N nodes, each row including the calibrated delay τ. cor Signal-to-noise ratio (SNR) and Quality of Service (QoS); Environmental quantity processing: Perform a 5-point moving average on the parameters with a sampling interval of 1 second to suppress measurement noise, map the original values ​​to the [0,1] interval, obtain the wind speed index and humidity index, calculate the comprehensive meteorological influence coefficient: meteorological coefficient = 60% wind speed index + 30% temperature index + 10% humidity index, output environmental quantity matrix: N×3, N nodes, each row includes wind speed index, temperature, and meteorological coefficient; Topology data processing: Compare the voltage difference between adjacent nodes. If it exceeds 5% of the rated voltage, the switch is determined to be actually open. If the line power is less than the threshold, the switch is determined to be loosely connected. Based on the verified switch status and preset impedance parameters, construct an impedance-weighted adjacency matrix of N×N.

5. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S3, the data are first grouped and encoded using features: Electrical characteristic coding: HAVE BEEN e =ReLU(W e ·[V rms ,U rms ,THD]+b e ) Among them W e V is the electrical feature weight matrix. rms I is the effective value of the voltage. rms Here, b represents the effective value of the current, THD represents the total harmonic distortion, and b represents the effective value of the current. e The electrical characteristic bias term is ReLU, which is the corrected linear unit. Communication feature encoding: E c =σ(W c ·[t cor ,SNR,QoS]+b c ) Where, τ cor The transmission delay after calibration is given by SNR, which is the signal-to-noise ratio, QoS is the communication quality level, and σ is the Sigmoid function. Environmental feature coding: E env =tanh(W env ·[W norm ,T,W weather ]+b env ) Where, x env This is an environmental feature vector containing the normalized wind speed W. norm Temperature T, meteorological coefficient W weather b env It is the bias vector; Then, a topological constraint attention mechanism is introduced to calculate the attention weights: Where, α i A represents the modal weighting coefficient. i is the node topology influence factor, and v is the learnable parameter vector; Then, a weighted fusion of features is performed: H=a e E e +a c E c +a env E env Where H is the output fused feature matrix, and α vector is the feature weight distribution. env For meteorological adaptive mechanisms: Among them, W weather The weather impact coefficient is 0.7, which is the threshold for severe weather.

6. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S4, the dynamic threshold is calculated as follows: T i =μ i +k·σ·(1+αW) Where, μ i Let σ be the historical feature mean of node i. i Let be the historical standard deviation of node i, k be the sensitivity coefficient, and α be the meteorological sensitivity factor; Anomaly detection rules: Output the list of abnormal nodes Γ={i|Alert i =1}.

7. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S5, the abnormal node set Γ, the fused feature matrix H, and the impedance-weighted adjacency matrix A are input, and the localization algorithm is as follows: Construct the abnormal node mask matrix M: M∈{0，1} N×N By focusing computational resources on the abnormal node and its neighborhood through the abnormal node mask matrix M, mask graph convolution calculation is performed: H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) ) Where l is the number of graph convolutional layers, and D is the degree matrix D ii =∑ j A ij σ is the GELU activation function, W (l) These are trainable weights; Loss function: Among them, L CE For cross-entropy loss, ||·|| Γ To mitigate topology reconstruction errors, feature consistency is constrained to prevent deviations in localization results. Output the faulty device / line ID and the probability P(Interval) of the fault interval. k )distributed: Where, P(Interval) i Let N be the failure probability of node i. (i) Let be the neighboring points of node i.

8. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S6, the Bayesian network is: Prior probability calculation: The lightning density is from the meteorological bureau's API, and 30 is the upper limit set for the operating years. Evidence likelihood calculation: P(Evidence|Cause)=P(R x |Cause)×P(Interval|Cause) Where R x Evidence variables include electrical evidence, communication evidence, environmental evidence, and spatial evidence. Electrical evidence comes from sudden changes in voltage / current; communication evidence comes from bit error rate / message delay; environmental evidence comes from lightning / wind speed; and spatial evidence comes from the probability P(Interval) of the fault interval. k ), and perform Bayesian inference to obtain the posterior probability: P(Cause|Evidence)∝P(Cause)×P(Evidence|Cause)×P(Interval|Cause) Where P(Interval|Cause) is the weight of the fault interval.

9. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: In step S7, the rule base is called to match the posterior probability analysis results in S6, and the action instruction is output.

10. The communication fault diagnosis method based on power systems according to claim 1, characterized in that: The rule base adopts a hierarchical decision logic design, wherein: The L1 safety isolation layer is triggered when the fault probability is greater than 90% and the current surge is greater than 300%. The response is to immediately trip the circuit breaker remotely and block the reclosing mechanism to prevent equipment damage and the escalation of the accident. The L2 emergency recovery layer is triggered when the probability of a lightning strike is greater than 60% and the wind speed is greater than level 8. The response is to activate drones for inspection, isolate the affected area, and then transfer power to the affected area. This layer is used for rapid power restoration in severe weather conditions. L3 planned maintenance layer, the trigger condition is aging probability > 70% + load rate < 50%, the handling action is planned power outage replacement, load transfer, and solving the problems of insufficient switches and aging.

11. A power system device, characterized in that, It includes a data acquisition module, a data processing module, and an instruction module. The data acquisition module is capable of acquiring data from multiple sources. The data processing module is capable of processing the acquired data in steps S2 to S6. The instruction module is capable of issuing processing instructions based on the data processing results.

12. An electronic device comprising a processor, a memory, and a computer program stored in the memory, characterized in that, When the computer program is executed by the processor, it implements the communication fault diagnosis method based on the power system as described in any one of claims 1-10.

13. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the communication fault diagnosis method based on the power system as described in any one of claims 1-10.

Images