Multi-stage evaluation electric power system short circuit and open circuit fault diagnosis method

Abstract

A multi-stage evaluation electric power system short circuit and open circuit fault diagnosis method comprises the steps that voltage and current sensors and fault recording devices are installed at key nodes such as a main transformer and a distribution line, and three-phase voltage and current waveform data and zero-sequence components are collected in real time and transmitted to a fault diagnosis server; the server firstly performs preliminary judgment and alarm triggering on short circuit, open circuit and one-way grounding / electric leakage faults, then performs discrete wavelet decomposition and multi-scale time-frequency domain feature extraction on acquired waveforms by using a deep fault classification model based on wavelet analysis and neural network fusion, and realizes fine classification of the faults through a multi-layer residual neural network. And finally, establishing a fault diagnosis result self-adaptive evaluation and response mechanism, dynamically adjusting an alarm level and a switching scheme according to a fault type, realizing protection equipment tripping and fault isolation for a short-circuit fault, generating a standby line switching suggestion for an open-circuit fault, and effectively improving the accuracy and coping capacity of power system fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power system fault diagnosis technology, and in particular to a multi-stage assessment method for diagnosing short-circuit and open-circuit faults in power systems. Background Technology

[0002] In today's society, the stability and reliability of power supply are crucial to people's daily lives, industrial production, and various social activities. A power system failure can range from affecting localized power supply to causing widespread blackouts and severe economic losses. Therefore, accurate and timely diagnosis of power system failures has become a key research focus in the power sector.

[0003] Currently, various methods and approaches exist for power system fault diagnosis. Traditional fault diagnosis methods primarily rely on human experience and simple electrical measuring instruments. Technicians need to periodically inspect power equipment, judging whether there are faults by observing the equipment's appearance, listening to its operating sounds, and using simple instruments to measure parameters such as voltage and current. However, this approach has many limitations. On the one hand, the frequency of manual inspections is limited, making real-time monitoring impossible, which may lead to faults occurring during inspection intervals without being detected in time. On the other hand, relying on human experience to judge faults is highly subjective; different technicians may have different judgment standards, easily leading to misjudgments or missed diagnoses.

[0004] With technological advancements, automated fault diagnosis systems based on electrical parameter monitoring have gradually emerged. These systems collect basic parameters such as voltage and current by installing sensors on power lines and setting simple thresholds to determine whether a fault has occurred. For example, when the current exceeds a certain set value, an overload fault is suspected. However, such systems have significant drawbacks. First, the parameters they collect are relatively limited, relying on only a few electrical parameters, making it difficult to comprehensively and accurately reflect the complex operating state of the power system. For some highly concealed faults requiring multi-parameter comprehensive analysis, such as certain complex short-circuit faults and grounding faults, this single-parameter judgment method is prone to misjudgment. Second, their fault judgment logic is too simplistic, relying solely on threshold comparisons without fully considering various dynamic changes and interference factors during power system operation. This leads to frequent false fault alarms when the power system is under special operating conditions or subjected to external interference, affecting the accuracy and reliability of fault diagnosis.

[0005] Furthermore, some existing fault diagnosis technologies fall short when dealing with large-scale, complex power networks. As power networks continue to expand, the number of nodes and lines increases dramatically, significantly enhancing the likelihood and complexity of faults. Traditional fault diagnosis systems suffer from low computational efficiency when processing such massive and complex data, failing to quickly and accurately pinpoint the location and type of faults. This leads to prolonged fault handling time, further expanding the scope and extent of the fault's impact on the power system.

[0006] Meanwhile, existing fault diagnosis technologies often lack effective evaluation and adaptive response mechanisms for fault diagnosis results. After a fault is diagnosed, the handling strategy cannot be dynamically adjusted according to factors such as the severity and scope of the fault. Whether it is a minor or serious fault, a relatively fixed handling method is used, which fails to achieve optimized resource allocation and efficient fault handling.

[0007] In summary, existing power system fault diagnosis technologies are inadequate in terms of real-time performance, accuracy, handling of complexities, and intelligent fault response. There is an urgent need for a more comprehensive, efficient, and intelligent power parameter acquisition, real-time transmission, and fault diagnosis technology to meet the ever-increasing demand for stable operation of modern power systems. Summary of the Invention

[0008] To address the above problems, this invention proposes a multi-stage assessment method for diagnosing short-circuit and open-circuit faults in power systems, with the following specific steps:

[0009] Step 1: Construct a power parameter acquisition and real-time transmission system. Install voltage and current sensors at the main transformer, distribution lines and key nodes, and transmit the collected three-phase voltage and current waveform data and zero-sequence component to the fault diagnosis server in real time.

[0010] Step 2: Perform preliminary fault diagnosis and alarm triggering. First, perform short circuit fault, open circuit fault and one-way grounding / leakage fault diagnosis in the fault diagnosis server. If the fault is preliminarily diagnosed, record the fault timestamp and corresponding node, and proceed to step 3 for fault interpretation.

[0011] Step 3: Develop a deep fault classification model based on the fusion of wavelet analysis and neural networks. First, the acquired waveform is decomposed into discrete wavelet to extract multi-scale time-frequency domain features. The features at each scale are then concatenated and input into a multi-layer residual neural network ResNet to achieve refined classification of single-phase grounding, two-phase short circuit, three-phase short circuit and open circuit faults.

[0012] Step 4: Establish an adaptive evaluation and response mechanism for fault diagnosis results. Based on the fault type output by the model, dynamically adjust the alarm level and switching scheme. Trigger protection equipment tripping and fault isolation for short-circuit faults, and generate backup line switching suggestions for open-circuit faults.

[0013] As a further improvement to the present invention, step 2, which involves preliminary fault assessment and alarm triggering, can be represented as follows:

[0014] Step 2.1: The regional diagnostic server receives the three-phase voltage, three-phase current and zero-sequence current signals Va, Vb, Vc, Ia, Ib, Ic and I0 collected in Step 1.

[0015] Step 2.2, Short Circuit Fault Early Warning

[0016] Continuously monitor the current signal of any phase. When any value of Ia, Ib, or Ic exceeds the rated current I of the corresponding phase... rated When it is 1.05 times, that is When the "overcurrent warning" state is active, the voltage of the same phase is marked as "overcurrent warning". If the "overcurrent warning" state persists for at least 3 sampling periods, the voltage of the same phase... It suddenly dropped to near 0V, that is... At the same time, the zero-sequence current is lower than the threshold. If the above conditions are met simultaneously, it is determined to be the initiation of a short circuit fault; if the above conditions are met simultaneously, a short circuit fault warning is immediately triggered.

[0017] Step 2.3, Circuit Breaker Fault Early Warning

[0018] When the three-phase currents Ia, Ib, and Ic are simultaneously detected at 50 consecutive sampling points The system is marked as "current interruption"; during the "current interruption" state, the phase voltage of a certain section of the line is detected to be stable near the system's rated power supply voltage range. Furthermore, the fluctuation range is less than ±2%, and the zero-sequence current is below the threshold. When both conditions are met, the system determines it to be an open circuit fault and immediately triggers an open circuit warning.

[0019] Step 2.4, One-way grounding / leakage warning

[0020] when exceeding the threshold continuously If the warning lasts for 10ms, mark "zero sequence warning". Simultaneously, if any value of Ia, Ib, or Ic exceeds the corresponding phase rated current I... rated When the value is 1.05 times that of the ground fault, it is determined to be a "single-phase ground fault warning";

[0021] Step 2.5: After the above warning is triggered, proceed to the second stage of fault diagnosis in Step 3.

[0022] As a further improvement to the present invention, the deep fault classification model based on the fusion of wavelet analysis and neural network in step 3 can be represented as follows:

[0023] Step 3.1, Discrete Wavelet Decomposition

[0024] The original three-phase current, three-phase voltage, and zero-sequence current are combined to form a signal. Where n is the sampling point index, and K-level discrete wavelet decomposition is performed respectively to obtain the approximation coefficients of each level. With detail coefficient :

[0025]

[0026] in, J is 4. , and These are wavelet low-pass and high-pass decomposition filters, respectively. Let be the approximation coefficient of the j-th order, and k be the index of the coefficients after decomposition. Let j be the level of detail coefficient;

[0027] Step 3.2, Multi-scale time-frequency feature extraction

[0028] For the seven signals including three-phase voltage, three-phase current, and zero-sequence current, detail coefficients at each stage are calculated. With the last level of approximation coefficient Extract the following time-domain and frequency-domain indices:

[0029] Step 3.2.1 Time Domain Features

[0030] Mean:

[0031] variance:

[0032] Kuroshi:

[0033] in, For detail coefficients The mean of K is the coefficient, and K is the total number of points in the coefficients after decomposition. For detail coefficients variance For detail coefficients Peak value;

[0034] Step 3.2.2 Frequency Domain Features

[0035] Spectral Energy:

[0036]

[0037] in, For detail coefficients The spectral energy, w is the discrete frequency point index; FFT is the Fourier transform;

[0038] Spectral entropy:

[0039]

[0040] in, For detail coefficients Spectral entropy, Normalized spectral probability;

[0041] The five features extracted from each signal at each level of decomposition are used to form a feature vector x;

[0042] Step 3.3: Based on ResNet fault classification, construct a multi-layer residual network containing L residual blocks. The network input is 𝑥, and the output is a fault probability vector y for 5 types of faults: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal.

[0043] Residual block definition: The input of the l-th block is Output:

[0044]

[0045] in, Let be the output of the l-th block, and let be the residual function containing the weights. It consists of two fully connected layers, ReLU, and BN:

[0046]

[0047] in, , Here, represents the weight matrix of the convolutional layer, BN is the batch normalization operation, and ReLU is the rectified linear activation function.

[0048] The output of the last residual stage is used as the input to the output layer Softmax, and the final output is the probability of the fault type:

[0049]

[0050] in, This is the output of the last residual stage. GAP is the global average pooling operation, Dropout is the random deactivation operation, and Softmax is the softmax activation function. and For the weights and biases of the fully connected layer, and The weights and biases of the output layer are defined; ultimately, y corresponds to five types of fault probability distributions, including: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal.

[0051] This invention provides a multi-stage assessment method for diagnosing short-circuit and open-circuit faults in power systems, with beneficial effects. The technical advantages of this invention are as follows:

[0052] 1. This invention utilizes discrete wavelet decomposition to perform multi-scale analysis of acquired waveforms, extracting rich time-frequency domain features containing detailed information about the power system's operating status. These features are then input into a multilayer residual neural network (ResNet), leveraging the neural network's nonlinear fitting capability to achieve refined classification of single-phase grounding, two-phase short-circuit, three-phase short-circuit, and open-circuit faults. Compared to traditional, simple fault classification methods, this model can handle more complex fault features, significantly improving fault classification accuracy and accurately distinguishing between different types of faults.

[0053] 2. Based on the fault type output by the model, this invention enables the system to dynamically adjust alarm levels and switching schemes. For short-circuit faults, it rapidly triggers protection equipment tripping and fault isolation to prevent further fault propagation and ensure the normal operation of other parts of the power system. For open-circuit faults, it generates backup line switching suggestions to quickly restore power supply and reduce outage time and scope. This differentiated and intelligent handling strategy based on different fault types effectively improves the power system's ability to cope with faults, optimizes resource allocation, minimizes the impact of faults on the stability and reliability of power supply, and ensures the safe, stable, and efficient operation of the power system. Attached Figure Description

[0054] Figure 1 This is a flowchart of the present invention;

[0055] Figure 2 This is the structure diagram of the ResNet structure of the present invention. Detailed Implementation

[0056] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0057] This invention discloses a multi-stage assessment method for diagnosing short-circuit and open-circuit faults in power systems. Sensors are installed at key nodes such as main transformers and distribution lines to collect and transmit power parameters in real time. After initial fault assessment and alarm triggering, a fusion model of wavelet analysis and neural networks is used to finely classify the faults. Alarms and responses are then dynamically adjusted based on the diagnostic results, improving the fault diagnosis and response capabilities of the power system. The invention flowchart is shown below. Figure 1 As shown, the steps of the present invention will be described in detail below.

[0058] Step 1: Construct a power parameter acquisition and real-time transmission system. Install voltage and current sensors at the main transformer, distribution lines and key nodes, and transmit the collected three-phase voltage and current waveform data and zero-sequence component to the fault diagnosis server in real time.

[0059] Voltage and current sensors and fault recording devices are installed on the high and low voltage sides of the transformer, overhead and cable sections, ring main units, and load centers, respectively. The collected three-phase voltage and current waveform data and zero-sequence components are transmitted to the fault diagnosis server in real time.

[0060] Step 2: Perform preliminary fault diagnosis and alarm triggering. First, perform short circuit fault, open circuit fault and one-way grounding / leakage fault diagnosis in the fault diagnosis server. If the fault is preliminarily diagnosed, record the fault timestamp and corresponding node, and proceed to step 3 for fault interpretation.

[0061] Step 2.1: The regional diagnostic server receives the three-phase voltage, three-phase current and zero-sequence current signals Va, Vb, Vc, Ia, Ib, Ic and I0 collected in Step 1.

[0062] Step 2.2, Short Circuit Fault Early Warning

[0063] Continuously monitor the current signal of any phase. When any value of Ia, Ib, or Ic exceeds the rated current I of the corresponding phase... rated When it is 1.05 times, that is When the "overcurrent warning" state is active, the voltage of the same phase is marked as "overcurrent warning". If the "overcurrent warning" state persists for at least 3 sampling periods, the voltage of the same phase... It suddenly dropped to near 0V, that is... At the same time, the zero-sequence current is lower than the threshold. If the above conditions are met simultaneously, it is determined to be the initiation of a short circuit fault; if the above conditions are met simultaneously, a short circuit fault warning is immediately triggered.

[0064] Step 2.3, Circuit Breaker Fault Early Warning

[0065] When the three-phase currents Ia, Ib, and Ic are simultaneously detected at 50 consecutive sampling points The system is marked as "current interruption"; during the "current interruption" state, the phase voltage of a certain section of the line is detected to be stable near the system's rated power supply voltage range. Furthermore, the fluctuation range is less than ±2%, and the zero-sequence current is below the threshold. When both conditions are met, the system determines that it is an open circuit fault and immediately triggers an open circuit warning.

[0066] Step 2.4, One-way grounding / leakage warning

[0067] when exceeding the threshold continuously If the warning lasts for 10ms, mark "zero sequence warning". Simultaneously, if any value of Ia, Ib, or Ic exceeds the corresponding phase rated current I... rated When the value is 1.05 times that of the ground fault, it is determined to be a "single-phase ground fault warning".

[0068] Step 2.5: After the above warning is triggered, proceed to the second stage of fault diagnosis in Step 3.

[0069] Step 3: Develop a deep fault classification model based on the fusion of wavelet analysis and neural networks. First, perform discrete wavelet decomposition on the acquired waveforms to extract multi-scale time-frequency domain features. Then, concatenate the features at each scale and input them into a multi-layer residual neural network (ResNet). The ResNet structure diagram is shown below. Figure 2 As shown, this enables a refined classification of single-phase grounding, two-phase short circuit, three-phase short circuit, and open circuit faults;

[0070] Step 3.1, Discrete Wavelet Decomposition

[0071] The original three-phase current, three-phase voltage, and zero-sequence current are combined to form a signal. Where n is the sampling point index, and K-level discrete wavelet decomposition is performed respectively to obtain the approximation coefficients of each level. With detail coefficient :

[0072]

[0073] in, J is 4. , and These are wavelet low-pass and high-pass decomposition filters, respectively. Let be the approximation coefficient of the j-th order, and k be the index of the coefficients after decomposition. Let be the detail coefficient of level j.

[0074] Step 3.2, Multi-scale time-frequency feature extraction

[0075] For the seven signals including three-phase voltage, three-phase current, and zero-sequence current, detail coefficients at each stage are calculated. With the last level of approximation coefficient Extract the following time-domain and frequency-domain indices:

[0076] Step 3.2.1 Time Domain Features

[0077] Mean:

[0078] variance:

[0079] Kuroshi:

[0080] in, For detail coefficients The mean of K is the coefficient, and K is the total number of points in the coefficients after decomposition. For detail coefficients variance For detail coefficients The peak value.

[0081] Step 3.2.2 Frequency Domain Features

[0082] Spectral Energy:

[0083]

[0084] in, For detail coefficients The spectral energy is given by w, where w is the discrete frequency point index. FFT stands for Fourier Transform.

[0085] Spectral entropy:

[0086]

[0087] in, For detail coefficients Spectral entropy, This represents the normalized spectral probability.

[0088] The five features extracted from each signal at each decomposition level are used to form a feature vector x.

[0089] Step 3.3: Based on ResNet fault classification, construct a multi-layer residual network containing L residual blocks. The network input is 𝑥, and the output is a fault probability vector y for 5 types: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal.

[0090] Residual block definition: The input of the l-th block is Output:

[0091]

[0092] in, Let be the output of the l-th block, and let be the residual function containing the weights. It consists of two fully connected layers, ReLU, and BN:

[0093]

[0094] in, , is the weight matrix of the convolutional layer, BN is the batch normalization operation, and ReLU is the linear rectified activation function.

[0095] The output of the last residual stage is used as the input to the output layer Softmax, and the final output is the probability of the fault type:

[0096]

[0097] in, This is the output of the last residual stage. GAP is the global average pooling operation, Dropout is the random deactivation operation, and Softmax is the softmax activation function. and For the weights and biases of the fully connected layer, and The weights and biases of the output layer are defined; ultimately, y corresponds to five types of fault probability distributions, including: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal.

[0098] Step 4: Establish an adaptive evaluation and response mechanism for fault diagnosis results. Based on the fault type output by the model, dynamically adjust the alarm level and switching scheme; trigger protection equipment tripping and fault isolation for short-circuit faults, and generate backup line switching suggestions for open-circuit faults.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A multi-stage assessment method for diagnosing short-circuit and open-circuit faults in power systems, comprising the following specific steps, characterized in that: Step 1: Construct a power parameter acquisition and real-time transmission system. Install voltage and current sensors at the main transformer, distribution lines and key nodes, and transmit the collected three-phase voltage and current waveform data and zero-sequence component to the fault diagnosis server in real time. Step 2: Perform preliminary fault diagnosis and alarm triggering. First, perform short circuit fault, open circuit fault and one-way grounding / leakage fault diagnosis in the fault diagnosis server. If the fault is preliminarily diagnosed, record the fault timestamp and corresponding node, and proceed to step 3 for fault interpretation. Step 3: Develop a deep fault classification model based on the fusion of wavelet analysis and neural networks. First, the acquired waveform is decomposed into discrete wavelet to extract multi-scale time-frequency domain features. The features at each scale are then concatenated and input into a multi-layer residual neural network ResNet to achieve refined classification of single-phase grounding, two-phase short circuit, three-phase short circuit and open circuit faults. Step 4: Establish an adaptive evaluation and response mechanism for fault diagnosis results. Based on the fault type output by the model, dynamically adjust the alarm level and switching scheme. Trigger protection equipment tripping and fault isolation for short-circuit faults, and generate backup line switching suggestions for open-circuit faults.

2. The method for diagnosing short-circuit and open-circuit faults in a power system based on multi-stage evaluation according to claim 1, characterized in that: Step 2, which involves preliminary fault diagnosis and alarm triggering, can be represented as follows: Step 2.1: The regional diagnostic server receives the three-phase voltage, three-phase current and zero-sequence current signals Va, Vb, Vc, Ia, Ib, Ic and I0 collected in Step 1. Step 2.2, Short Circuit Fault Early Warning Continuously monitor the current signal of any phase. When any value of Ia, Ib, or Ic exceeds the rated current I of the corresponding phase... rated When it is 1.05 times, that is When the "overcurrent warning" state is active, the voltage of the same phase is marked as "overcurrent warning". If the "overcurrent warning" state persists for at least 3 sampling periods, the voltage of the same phase... It suddenly dropped to near 0V, that is... At the same time, the zero-sequence current is lower than the threshold. If the above conditions are met simultaneously, it is determined to be the initiation of a short circuit fault; if the above conditions are met simultaneously, a short circuit fault warning is immediately triggered. Step 2.3, Circuit Breaker Fault Early Warning When the three-phase currents Ia, Ib, and Ic are simultaneously detected at 50 consecutive sampling points Mark the "current interruption" status; During the "current interruption" state, the phase voltage of a certain section of the line is detected to be stable near the system's rated power supply voltage range, i.e. Furthermore, the fluctuation range is less than ±2%, and the zero-sequence current is below the threshold. ; When both conditions are met, the system determines that there is a circuit breaker fault and immediately triggers a circuit breaker warning. Step 2.4, One-way grounding / leakage warning when exceeding the threshold continuously If the warning lasts for 10ms, mark "zero sequence warning". Simultaneously, if any value of Ia, Ib, or Ic exceeds the corresponding phase rated current I... rated When the value is 1.05 times that of the ground fault, it is determined to be a "single-phase ground fault warning"; Step 2.5: After the above warning is triggered, proceed to the second stage of fault diagnosis in Step 3.

3. The method for diagnosing short-circuit and open-circuit faults in a power system based on multi-stage assessment as described in claim 1, characterized in that: The deep fault classification model based on the fusion of wavelet analysis and neural network in step 3 can be represented as follows: Step 3.1, Discrete Wavelet Decomposition The original three-phase current, three-phase voltage, and zero-sequence current are combined to form a signal. Where n is the sampling point index, and K-level discrete wavelet decomposition is performed respectively to obtain the approximation coefficients of each level. With detail coefficient : ； in, J is 4. , and These are wavelet low-pass and high-pass decomposition filters, respectively. Let be the approximation coefficient of the j-th order, and k be the index of the coefficients after decomposition. Let j be the level of detail coefficient; Step 3.2, Multi-scale time-frequency feature extraction For the seven signals including three-phase voltage, three-phase current, and zero-sequence current, detail coefficients at each stage are calculated. With the last level of approximation coefficient Extract the following time-domain and frequency-domain indices: Step 3.2.1 Time Domain Features Mean: ; variance: ; Kurtosis: ; in, For detail coefficients The mean of the coefficients, where K is the total number of coefficients after decomposition. For detail coefficients variance For detail coefficients Peak value; Step 3.2.2 Frequency Domain Features Spectral Energy: ； in, For detail coefficients The spectral energy, w is the discrete frequency point index; FFT is the Fourier transform; Spectral entropy: ； in, For detail coefficients Spectral entropy, Normalized spectral probability; The five features extracted from each signal at each level of decomposition are used to form a feature vector x; Step 3.3: Based on ResNet fault classification, construct a multi-layer residual network containing L residual blocks. The network input is 𝑥, and the output is a fault probability vector y for 5 types of faults: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal. Residual block definition: The input of the l-th block is Output: ； in, Let be the output of the l-th block, and let be the residual function containing the weights. It consists of two fully connected layers, ReLU, and BN: ； in, , Here, represents the weight matrix of the convolutional layer, BN is the batch normalization operation, and ReLU is the rectified linear activation function. The output of the last residual stage is used as the input to the output layer Softmax, and the final output is the probability of the fault type: ； in, This is the output of the last residual stage. GAP is the global average pooling operation, Dropout is the random deactivation operation, and Softmax is the softmax activation function. and For the weights and biases of the fully connected layer, and The weights and biases of the output layer are defined; ultimately, y corresponds to five types of fault probability distributions, including: single-phase grounding, two-phase short circuit, three-phase short circuit, open circuit, and normal.

Images