Fuse slow fusing fault detection method based on phase voltage correlation characteristics

Abstract

The invention provides a fuse slow fusing fault detection method based on phase voltage correlation characteristics. The method comprises the following steps: acquiring three-phase phase voltages of a plurality of nodes at the upstream and the middle and downstream of a line, performing normalization and sliding window processing on the acquired phase voltages, and inputting the phase voltages into a neural network model for training; the fuse anomaly detection method based on the neural network is constructed according to the characteristic that phase voltage may slightly drop when the fuse is slowly fused and by combining priori knowledge of voltage regular change of each node of a line, and a phase voltage reconstruction value is updated in real time by mining nonlinear characteristics among voltage data of each node through the neural network. And finally, according to whether the root-mean-square error between the actual value and the reconstruction value of the phase voltage exceeds an adaptive threshold value, identifying that the fuse is abnormal. Example results show that the method can effectively identify the slow fusing process of the fuse, and a foundation is laid for reliable operation of a protection system and improvement of operation and maintenance efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of power equipment status, and in particular to a method for detecting slow-blow faults of fuses based on phase voltage correlation characteristics. Background Art

[0002] As an overload current and short-circuit current protection element with a simple structure, fast and reliable action, the fuse is usually installed on the primary side of the TV. The stable and reliable operation of the fuse is the key basis for accurate measurement, protection and control of the power grid system. In recent years, extreme weather has occurred frequently around the world, and fuses are facing a more severe operating environment, and there are more and more abnormal fusing cases. Fuse abnormalities directly affect the metering accuracy of the transformer, causing increased line losses and affecting the economic benefits of the power grid. In severe cases, it may even cause major safety accidents such as circuit protection failure, TV equipment damage, and power outages. Since fuse failures usually occur internally, it is difficult to determine the status and performance of the fuse. In actual engineering inspections, a lot of manpower is often required for regular inspections.

[0003] Early studies on fuse anomalies were mostly centered around the physical characteristics of the power grid system during operation. By improving the tolerance of the fuse equipment during system operation, the entire power system can be protected. For example, when the power system operating state changes suddenly, ferromagnetic resonance may occur, causing damage to the voltage transformer fuse. Therefore, a series of detuning measures are proposed to ensure the stable operation of the power system. For the ultra-low frequency transient impact current caused by system short circuit, the fuse blows, and the VT or 4VT wiring method with good excitation characteristics can be selected. The instantaneous surge current on the primary side is the direct cause of the damage to the high-voltage fuse. Increasing the rated capacity of the fuse effectively improves the fuse's tolerance to the surge current. When the ambient temperature rises, the power load increases, the quality of the fuse is not up to standard, and the fuse fuse blows abnormally and frequently. This method, combined with the equipment operating environment, can protect the power system by improving the tolerance of the fuse equipment when the system state changes suddenly. Due to differences in the actual operating environment, capacity, and access equipment, in practice, only targeted improvements are made to a single region. However, slow blowing caused by objective factors such as the equipment operating environment and aging still requires a lot of manpower and material resources from operation and maintenance personnel. At present, there is no suitable slow blowing detection method for fuses. Generally, on-site inspection is carried out only after the fuse develops from slow blowing to complete blowing, which reduces operation and maintenance efficiency and increases the probability of failure.

[0004] In summary, at this stage, there is an urgent need to propose a detection method suitable for slow-blow fuse faults, which can detect fuse abnormalities, improve operation and maintenance efficiency, and reduce the risk of power grid failures. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting a slow-blowing fuse fault to solve the above shortcomings of the prior art. When a fuse blows slowly, the voltage change shows a slight drop characteristic, which is easily confused with the voltage drop caused by line overload or three-phase load imbalance. Therefore, the traditional TV disconnection detection method that sets a voltage threshold is difficult to detect the slow-blowing fuse.

[0006] The method of the present invention is a TV fuse slow-blowing detection method based on the GRU attention mechanism. When the system is operating normally, the phase voltage of each node fluctuates within the normal range, and the phase voltage has a similar change trend; when the unbalanced load is switched, the phase voltage suddenly changes at the same time, and the difference between the two voltage values ​​increases, and then the phase voltage still has a similar change trend; when the TV fuse blows slowly, the local resistance increases the voltage division, and the phase voltage shows a slight drop feature, while the phase voltage of the non-faulty node still maintains a similar change trend. Combined with the characteristics that the same-phase voltage changes at different nodes of the distribution line are similar and the voltage shows a slight drop when the fuse blows slowly, a fuse slow-blowing detection method based on the GRU attention mechanism is proposed, and the improved cyclic network structure is used to mine the linear relationship between the phase voltages of each node to realize the identification of abnormal fuses.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for detecting a slow-blow fuse fault based on phase voltage correlation characteristics comprises the following steps:

[0009] Step 1) Data collection: collecting three-phase voltages of multiple nodes upstream and mid-downstream of the line;

[0010] Step 2) Data processing: normalizing and sliding window processing the collected phase voltages;

[0011] Step 3) Model training: The processed data set is input into the neural network model for training. The model is iteratively optimized by minimizing the MSE. When the MSE value tends to be stable, the output is converted into a reconstructed value through a denormalization operation, and the root mean square error L between the reconstructed value and the actual value is calculated. RMSE ;

[0012] Step 4) Fault judgment: Adopt the adaptive threshold η to judge the fuse abnormality. RMSE When the root mean square error value L is greater than η, an abnormal warning is issued. RMSE It is normal when it is less than or equal to η.

[0013] Preferably, the phase voltage of the upstream node in step 1) refers to the phase voltage measured by the 220kV / 110kV / 35kV / 10kV bus TV of the substation, and the phase voltage of the mid- and downstream nodes is obtained by multiplying the actual phase voltage of the 220kV / 110kV / 35kV / 10kV bus outgoing line to the user by the transformation ratio, or is the bus phase voltage of the 220kV / 110kV / 35kV / 10kV bus outgoing line to the next-level substation 220kV / 110kV / 35kV / 10kV corresponding voltage level.

[0014] Preferably, the phase voltage data in step 1) is collected continuously or periodically.

[0015] Preferably, the normalization of the phase voltage in step 2) refers to the formula Normalize, where: x * is the normalized value; x is the original input data; x min is the minimum value of the original input set; x max is the maximum value of the original input set.

[0016] Preferably, the neural network model in step 3) is a neural network prediction algorithm based on a recurrent neural network and its variants. Further, the recurrent neural network and its variants include a GRU neural network or a LSTM neural network.

[0017] Preferably, the root mean square error value L in step 3) is RMSE The calculation method is Where: m is the number of samples; x i is the actual value; i is the reconstructed value.

[0018] Preferably, the adaptive threshold calculation method in step 4) is η=3σ+μ, where μ is L RMSE The mean of RMSE The standard deviation is calculated as

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: when the system is operating normally, the phase voltage of each node fluctuates within the normal range, and the phase voltage has a similar change trend; when the unbalanced load is switched, the phase voltage suddenly changes at the same time, the difference between the two voltage values ​​increases, and then the phase voltage still has a similar change trend; when the TV fuse blows slowly, the local resistance increases the voltage division, and the phase voltage shows a slight drop feature, while the phase voltage of the non-faulty node still maintains a similar change trend. Combined with the characteristics that the same-phase voltage changes at different nodes of the distribution line are similar and the voltage shows a slight drop when the fuse blows slowly, a fuse slow-blowing detection method based on the GRU attention mechanism is proposed, and the improved cyclic network structure is used to mine the linear relationship between the phase voltages of each node to realize the identification of abnormal fuses. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for detecting a slow-blow fault of a fuse based on phase voltage correlation characteristics of the present invention;

[0021] Figure 2 It is a schematic diagram of the sliding window processing of the data set;

[0022] Figure 3 This is the normal A-phase voltage waveform diagram of the upstream, midstream and downstream of the line;

[0023] Figure 4 It is the result obtained by diagnosing a normally operating circuit using the method of the present invention;

[0024] Figure 5 It is the voltage waveform of phase B of the fuse abnormal circuit;

[0025] Figure 6 It is the result obtained by diagnosing abnormally running lines using the method of the present invention. DETAILED DESCRIPTION

[0026] The present invention proposes a method for detecting slow-blowing faults of fuses based on phase voltage correlation characteristics. The method utilizes the characteristics that the fluctuation laws of single-phase voltages at each node in the same line are similar, and proposes a method for detecting slow-blowing faults of fuses based on a neural network. The method of the present invention collects three-phase phase voltages of multiple nodes in the upstream, midstream and downstream of the line, normalizes the collected phase voltages, performs sliding window processing, and then inputs them into a neural network model for training; according to the characteristics that the phase voltage may have a slight drop when the fuse is slowly blown, combined with the prior knowledge of the voltage law changes at each node of the line, a method for detecting fuse anomalies based on a neural network is constructed, and the nonlinear characteristics between the voltage data of each node are mined by the neural network to update the phase voltage reconstruction value in real time. Finally, the fuse anomaly is identified based on whether the root mean square error between the actual value and the reconstructed value of the phase voltage exceeds the adaptive threshold. The example results show that the proposed method can effectively identify the slow-blowing process of the fuse, laying a foundation for the reliable operation of the protection system and improving the operation and maintenance efficiency.

[0027] In an exemplary embodiment, if Figure 1 As shown in the flow chart, a method for detecting a slow-blow fuse fault based on phase voltage correlation characteristics of the present invention comprises the following steps:

[0028] Step 1) Data collection: Collect the three-phase voltages of multiple nodes in the upstream, midstream and downstream of the line.

[0029] The above-mentioned upstream node voltage refers to the phase voltage measured by the 220kV / 110kV / 35kV / 10kV bus TV of the substation. The mid- and downstream node phase voltage is obtained by multiplying the actual phase voltage of the user in the mid- and downstream positions of the line by the transformation ratio, or the phase voltage measured by the bus TV of the mid- and downstream substation; voltage data collection can be continuous or periodic. There can be one or more mid- and downstream nodes, for example, it can be measured at the end user or at the lower substation. In the case of multiple substations, it can be measured at each substation, or one or several substations can be selected for measurement, or it can be measured at multiple substations and users.

[0030] Step 2) Data processing: normalize and perform sliding window processing on the collected phase voltages.

[0031] When considering the neural network for multivariate time series prediction, the value of the variable will seriously affect the training of the network weight parameters. Therefore, the sample data is first normalized, and the calculation formula is as follows:

[0032] x * = x-x min

[0033] x max -x min

[0034] Where: x * is the normalized value; x is the original input data; x min is the minimum value of the original input set; x max is the maximum value of the original input set.

[0035] The normalized multivariate sequence is processed by sliding window. Considering the voltage drop caused by the fuse anomaly, which is often in the local point, the GRU detection model based on the attention mechanism performs anomaly detection based on the sampling points of the previous 4 days. The acquisition frequency is 5 minutes, the window is 12 sampling points (the time dimension corresponds to 1 hour), and the step size is 1 sampling point. Figure 2 shown.

[0036] Step 3) Model training: Input the normalized and sliding window processed data set into the GRU attention mechanism model, and iteratively optimize the model by minimizing the MSE (Mean Squared Error). When the MSE value tends to be stable, the output is converted into a reconstructed value through a denormalization operation.

[0037] The input value of this model is the single-phase voltage data of the middle and downstream nodes, which is used to train and reconstruct the single-phase voltage of the target port. The target value is the single-phase voltage upstream of the port to be detected. It is reconstructed as the target of the model and is only used in the training phase, not as an input feature of the model. The preprocessed data set is input into the GRU attention mechanism model, and the model is iteratively optimized by minimizing the MSE. When the MSE value tends to be stable, the output is converted into a reconstructed value through a denormalization operation.

[0038] Step 4) RMSE (Root Mean Squared Error) is selected as the evaluation index, and an adaptive threshold is used to judge the fuse abnormality. When the RMSE value of the reconstructed value and the actual value exceeds the adaptive threshold, an abnormal warning is issued.

[0039] The nonlinear characteristics of single-phase voltages at each node of the line are learned through neural networks, and there is still a certain fluctuation between the reconstructed value and the actual value. In order to further distinguish the transient voltage drop caused by single-phase overload of the line, large-scale grid connection of new energy and voltage transformer fusing, RMSE, which is more sensitive to deviation data, is selected as the evaluation index, and an adaptive threshold is used to judge the fuse anomaly. When the reconstructed value and the actual value exceed the threshold, an abnormal warning is issued. The calculation rules are as follows:

[0040]

[0041] η=3σ+μ

[0042]

[0043] Where: m is the number of samples; x i is the actual value; i is the reconstruction value; μ is the mean, σ is the standard deviation; η is the adaptive threshold; L RMSE is the root mean square error value.

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicating the orientation or positional relationship are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0046] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] The present invention provides a method for detecting a slow-blow fault of a fuse based on phase voltage correlation characteristics, comprising the following steps:

[0048] Step 1) Data collection: Collect the three-phase voltages of each node in the upstream, midstream and downstream of the line;

[0049] Step 2) Data processing: normalizing and sliding window processing the collected phase voltages;

[0050] Step 3) Model training: Input the preprocessed data set into the GRU attention mechanism model, iteratively optimize the model by minimizing the MSE, and when the MSE value tends to be stable, convert the output into a reconstructed value through a denormalization operation;

[0051] Step 4) Root Mean Squared Error (RMSE) is selected as the evaluation index, and an adaptive threshold is used to judge the fuse abnormality. When the RMSE value between the reconstructed value and the actual value exceeds the adaptive threshold, an abnormal warning is issued.

[0052] Example 1

[0053] Flowchart as Figure 1 As shown in Figure 2, in order to verify the reconstruction performance of the model, the three-phase voltage data of each node in the 35KV normal line is first collected, as shown in Figure 2. Figure 3 As shown, the process of the method of the present invention is explained by taking the fault detection of phase A high-voltage fuse as an example.

[0054] (1) GRU algorithm detection during normal operation of the fuse

[0055] The sample data set contains the A-phase voltage data of each metering port in the upstream, midstream and downstream, which are recorded as A1, A2 and A3 respectively. A1 is the A-phase voltage measured by the PT of the 35kV I bus section in a 110kV substation, A2 is the A-phase voltage measured on the 35kV Ludong 365 outgoing line switch on the 35kV I bus section, and A3 is the 10KVA phase voltage measured from the 35kV Ludong 365 outgoing line through the step-down transformer to the 10kV line pole switch multiplied by the step-down transformer ratio (35kV / 10kV). The acquisition frequency is 5 minutes and the time dimension is 4 days. The A1, A2 and A3 phase voltage waveforms are as follows Figure 3 As shown. Then, the sample data set is preprocessed and then input into the GRU attention mechanism model. The point to be detected is the upstream A phase port. The phase voltage data features between A2 and A3 are learned and refined. When the MSE between the reconstructed value based on the mid- and downstream phase voltage features and the actual value tends to be stable, it indicates that the model has completed learning and outputs the single-phase voltage reconstruction value of the port to be detected. The result is shown in Figure 4 The upper curve in the figure is the A1 phase voltage, and the lower bar is the RMSE value between the reconstructed value and the actual value (for the convenience of display, only the last 1000 sampling points are shown in the figure).

[0056] Depend on Figure 3 , 4 Analysis shows that:

[0057] During normal operation, the line is affected by the size of the connected load, and the same-phase voltage is always in a dynamic change process, but the overall change trend of the A1, A2, and A3 phase voltages is still similar.

[0058] The RMSE between the reconstructed value and the actual value is always less than the threshold (η=0.0399 calculated by the formula η=3σ+μ), and is judged to be normal and consistent with the actual situation.

[0059] The A1, A2, and A3 phase voltage data are input into the GRU attention mechanism model, and the RMSE average is 0.0107, which is relatively small overall. This shows that the model can learn the phase voltage fluctuation characteristics at different nodes and can accurately reconstruct the single-phase voltage of the target node. It has a good learning ability for the operating characteristics of the voltage at each node when the line is operating normally.

[0060] (2) GRU algorithm detection during abnormal operation of the fuse

[0061] In order to verify the effectiveness of the above model when the fuse is abnormal, the three-phase voltage data of the upstream, midstream and downstream of the 35KV bus TV is collected as a simulation example when the B-phase fuse is slowly blown and eventually causes a power outage of the line. The operation data of the first four days when the voltage starts to drop slightly are taken for detection. The data set contains the B-phase voltage data of each metering port in the upstream, midstream and downstream, which are recorded as B1, B2 and B3 respectively. The acquisition frequency is 5 minutes. The B-phase voltage waveform is as follows: Figure 5 As shown, the red box shows a slight voltage drop caused by the start of the abnormal slow fuse failure. The pre-processed B phase voltage data is input into the model of the present invention, and the reconstruction result is as follows: Figure 6 As shown (for the convenience of showing the effect, only the last 1000 sampling points are shown in the figure).

[0062] Depend on Figure 5 , 6 Analysis shows that:

[0063] Before the fuse is abnormal, the phase voltages of lines B1, B2, and B3 change regularly. The phase voltage fluctuates with the connected load and remains stable for a long time. During the normal operation of the corresponding line, the average value of RMSE is 0.0109. The model reconstruction performance is good, indicating that the model has extracted the nonlinear characteristics between the phase voltages under this line.

[0064] The RMSE between 829 and 1000 is higher than the threshold (calculated by the formula η = 0.1204), which is judged as a fuse abnormality, and the actual corresponding fuse is slow to blow. The slow-blowing resistance of the fuse increases the voltage divider, resulting in a decrease in the measurement accuracy of the voltage transformer on the B1 side, and the phase voltage value gradually deviates from the actual value over time. At this time, the B2 and B3 phase voltages are the true values ​​of the line and are not affected by the fuse. The voltage fluctuations of B2 and B3 maintain the previous rules, while the voltage fluctuation curve of the B1 phase voltage is significantly abnormal from the previous rules due to the slow blowing of the fuse, resulting in the model being unable to effectively reconstruct the phase voltage value at this location, and is therefore identified as abnormal.

[0065] The model can effectively extract the line voltage characteristics and fit the reconstruction value based on the line phase voltage characteristics in real time. The short-term drop in phase voltage caused by the connection of new energy to the grid, excessive single-phase load, etc. will not cause a decrease in model reconstruction performance or misjudgment.

[0066] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0067] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified and replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for detecting slow-blow fuse faults based on phase voltage correlation characteristics, characterized in that: The method comprises the following steps: Step 1) Data collection: collecting three-phase voltages of multiple nodes upstream and mid-downstream of the line; Step 2) Data processing: normalizing and sliding window processing the collected phase voltages; Step 3) Model training: The processed data set is input into the neural network model for training. The model is iteratively optimized by minimizing the MSE. When the MSE value tends to be stable, the output is converted into a reconstructed value through a denormalization operation, and the root mean square error L between the reconstructed value and the actual value is calculated. RMSE ; Step 4) Fault judgment: Adopt the adaptive threshold η to judge the fuse abnormality. RMSE When the root mean square error value L is greater than η, an abnormal warning is issued. RMSE It is normal when it is less than or equal to η.

2. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: The phase voltage of the upstream node in step 1) refers to the phase voltage measured by the 220kV / 110kV / 35kV / 10kV bus TV of the substation, and the phase voltage of the mid- and downstream nodes is obtained by multiplying the actual phase voltage of the 220kV / 110kV / 35kV / 10kV bus outgoing line to the user by the transformation ratio, or the bus phase voltage of the 220kV / 110kV / 35kV / 10kV bus outgoing line to the next-level substation 220kV / 110kV / 35kV / 10kV corresponding voltage level.

3. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: The phase voltage data in step 1) is collected continuously or periodically.

4. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: In step 2), the normalization of the phase voltage is based on the formula Normalize, where: x * is the normalized value; x is the original input data; x min is the minimum value of the original input set; x max is the maximum value of the original input set.

5. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: The neural network model in step 3) is a neural network prediction algorithm based on recurrent neural network and its variants.

6. A method for detecting slow-blow fuse faults based on phase voltage correlation characteristics as described in claim 5, characterized in that: The recurrent neural network and its variants include a GRU neural network or a LSTM neural network.

7. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: Step 3) The root mean square error value L RMSE The calculation method is Where: m is the number of samples; x i is the actual value; i is the reconstructed value.

8. The method for detecting slow-blow fuse faults based on phase voltage correlation characteristics according to claim 1, characterized in that: The adaptive threshold calculation method in step 4) is η=3σ+μ, where μ is L RMSE The mean of RMSE The standard deviation of

Images