Direct current arc fault location method based on redundant antenna array and ellipse algorithm

Abstract

The application proposes a direct current arc fault positioning method based on a redundant antenna array and an elliptical algorithm, which first filters the electromagnetic radiation signal EMR collected by the antenna during arc burning, calculates the root mean square value as the received signal strength indicator RSSI, and constructs a neural network model, which predicts the distance from the fault point to the receiving point based on the RSSI value and the irradiance; secondly, a redundant antenna array is constructed to analyze the influence of different antenna numbers and layout modes on the positioning accuracy; thirdly, after determining the optimal antenna layout, the three strongest signals are selected, the antenna coordinates and the predicted distance of the fault point are input into the improved elliptical positioning method based on the three-edge positioning method, and the arc fault position is obtained; finally, the density-based noise clustering method DBSCAN is used to fuse multiple measurement results, eliminate interference and determine the final fault coordinates; the application has high positioning accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of arc fault detection and positioning of a power distribution network, and in particular to a direct-current arc fault positioning method based on a redundant antenna array and an elliptical algorithm. BACKGROUND

[0002] Photovoltaic power generation realizes the conversion of solar energy into electric energy, and relieves the dependence of electric energy supply on conventional energy generation. A photovoltaic power generation system comprises a photovoltaic array, a current combiner box and a photovoltaic inverter and the like, and there are many connections and joints on the direct-current side of the array, which are prone to insulation damage and poor contact in an outdoor environment, and are extremely prone to cause direct-current arc faults, among which series arc faults are particularly prominent. Since the direct-current arc does not have a natural zero point, it is difficult to extinguish once it occurs, and is prone to cause electrical fire accidents. Traditional protection cannot find the position of the arc, and therefore, it is necessary to not only monitor whether an arc fault occurs in the photovoltaic system in real time, but also to position the fault.

[0003] Most of the existing researches mainly focus on the detection of direct-current arc faults, and through extracting the time domain, frequency domain or time-frequency domain features of the voltage and current during the arc fault, artificial intelligence, threshold method and the like are applied to realize the identification and classification of the features. During the formation of the fault arc, high-frequency conduction current and electromagnetic radiation as well as arc sound, arc light and other macro physical characteristics are presented, and through collecting, counting and calculating the same, the position of the arc can be accurately positioned. Compared with other physical characteristics, the arc EMR signal with a higher frequency has stronger anti-interference ability, and a suitable measurement antenna can achieve sufficient sensitivity. SUMMARY

[0004] The application provides a direct-current arc fault positioning method based on a redundant antenna array and an elliptical algorithm, and the method has high positioning accuracy.

[0005] The application adopts the following technical scheme.

[0006] The direct-current arc fault positioning method based on the redundant antenna array and the elliptical algorithm first performs filtering processing on electromagnetic radiation signals EMR collected by the antenna during arc combustion, calculates a root mean square value as a received signal strength indicator RSSI, and constructs a neural network model, the model predicts the distance from the fault point to the receiving point according to the RSSI value and the irradiance; secondly, a redundant antenna array is constructed, and the influence of different numbers and layout modes of the antennas on the positioning accuracy is analyzed; thirdly, after the optimal antenna layout is determined, three antennas with the strongest signals are selected, and the antenna coordinates and the predicted distance of the fault point are input into the elliptical positioning method improved on the basis of the trilateration method, so that the arc fault position is obtained; finally, a density-based noise clustering method DBSCAN is used to fuse multiple measurement results, and the final fault coordinates are determined after interference is excluded.

[0007] The method comprises the following steps when used for direct current arc fault positioning of a photovoltaic system:

[0008] Step S01: using a single antenna, collecting arc EMR signals at different distances and different irradiances at a preset sampling rate, and obtaining RSSI values after filtering processing;

[0009] Step S02: using RSSI and irradiance to train a neural network model to predict output distance;

[0010] Step S03: establishing a Cartesian coordinate system with the photovoltaic array as a plane, determining the number and layout mode of the redundant antenna array, and placing the antenna;

[0011] Step S04: converting the arc EMR signals collected by the antenna array into RSSI, obtaining the average value after multiple time windows, and selecting the antenna according to the size order of RSSI;

[0012] Step S05: inputting the RSSI of the three antennas and the irradiance at the corresponding moment into the trained neural network model to obtain the predicted distance of the arc to the antenna array, and using an elliptical positioning algorithm to obtain the arc fault point position;

[0013] Step S06: using DBSCAN algorithm for cluster analysis to further reduce the error and obtain the final coordinates of the arc fault point position, and completing the arc fault positioning.

[0014] The construction method of the neural network model is: connecting an arc generator between photovoltaic modules, sampling when the arc is stable, and sampling at a default sampling frequency of 200 kHz; gradually increasing the distance between the antenna and the arc generator at a default step of 0.5 meters (not more than the size length of the photovoltaic panel, default 0.5 meters), and collecting a large amount of arc EMR signal data at different irradiances and distances with one antenna, which is used as a training set for constructing a neural network model;

[0015] When constructing the test set of the neural network model, during the data collection process, the antenna is placed according to the preset antenna layout, and the EMR signals at different irradiances and distances are collected in the same way as described above with a non-fixed step, and the distance range is the same as the test set.

[0016] In step S01, the antenna collects signals from the arc generating device connected between the photovoltaic modules.

[0017] The arc length and current of the photovoltaic direct current can both affect the arc EMR intensity, but the current has a greater impact, and the impact of the arc length is ignored in step S01, and the arc length of the arc generating device is used to approximate the arc length of the photovoltaic direct current, and the gap length between the electrodes is randomly set;

[0018] The electric arc generating device is used for generating a direct current arc, and the direct current arc forms different current signal waveforms and EMR model waveforms in normal, starting combustion, stable combustion and extinguishing four stages, the current signal waveforms have different characteristics in each stage, and the arc EMR signal waveform changes as follows: when the arc starts to burn, the EMR signal amplitude rapidly increases, when the arc is extinguished, the EMR signal rapidly decreases to the normal state, and there is almost no transition state.

[0019] The method is used for quickly finding a fault point after an arc fault occurs in a photovoltaic array, and in step S01, the acquisition time and mode of the arc EMR signal are selected as the waveform in the stable combustion stage after the fault occurs, that is, sampling when the arc of the arc generating device is stable.

[0020] In step S03, the setting method of the redundant antenna array is as follows: arranging n antennas in the photovoltaic array area, each antenna has a corresponding coordinate, and sequentially numbering them as S1, S2, …, Sn. n Wherein n≥4. Select the first three antennas with the maximum RSSI: RSSI i ,RSSI j ,RSSI k ,i,j,k∈n, so that RSSI i ≥RSSI j ≥RSSI k ≥RSSI m Wherein, m≠i, j, k, m∈n.

[0021] In step S05, the implementation steps of the ellipse algorithm include:

[0022] Step A1, arranging three antennas A, B, C and an arc fault point T, and inputting RSSI into a neural network to fit the distance, so as to determine the known information: the coordinates of the three antennas and the distances from the arc fault point to each antenna; step A2, taking antennas A, B and C as centers and the distances from the antennas to the arc fault as radii, respectively making three circles, and using the following formula to represent them. Wherein ε n represents the ellipse made (the circle is a special ellipse), x n represents the center, Q n represents the radius, n=A, B, C; step A3, sequentially finding the union set of each two of the three circles, and then finding the smallest outer ellipse of the three union sets, forming three ellipses, and using the following formula to represent them, wherein ∪ and represent that two sets are first operated, and then belong to a new set together;

[0023]

[0024] Step A4, find the largest interior ellipse of the intersection of the three ellipses in step A3, and express it by the following formula.

[0025]

[0026] Step A5, finally, let the ellipse center coordinate ε μ (x μ ,Q μ ) be the arc fault location.

[0027] In step S06, the input parameters of the DBSCAN algorithm are the neighborhood radius Eps and the threshold value MinPts of the number of data objects in the neighborhood; the output is the density connected cluster. The number of sample points in the neighborhood radius R is greater than or equal to MinPts, which is the core point; the point that is not a core point but is in the neighborhood of a core point is a boundary point; neither a core point nor a boundary point is a noise point;

[0028] The processing flow of the DBSCAN algorithm for clustering analysis is as follows:

[0029] Step B1, arbitrarily select a point p, which is neither assigned to a class nor specified as a peripheral point, and calculate whether the selected data object point is a core point for parameters Eps and MinPts; if so, find all density-reachable data object points from p to form a cluster;

[0030] Step B2, if the selected data object point p is an edge point, select another data object point;

[0031] Step B3, repeat steps B1 and B2 until all points have the characteristics of being in a cluster or are noise points, and the characteristics of the point in the cluster are core points or edge points;

[0032] Step B4, select the clustering cluster with more sample points, abandon the cluster with less sample points and noise points, and refer to the K-means algorithm to calculate the mean of each sample point as the clustering center.

[0033] In step S01, the method for obtaining RSSI is: collecting EMR signals during stable combustion of the arc at a lower sampling rate (for example, 200 kHz), and randomly cutting a predetermined length waveform (the length is recommended to be greater than or equal to 5 ms) from the EMR signals; each waveform is first removed from the pulse interference point by a filtering method, and then the root mean square value is calculated as RSSI.

[0034] Compared with the existing technical solutions, the present application has the following advantages:

[0035] 1. The ellipse algorithm replaces the traditional positioning method "three-edge positioning method", solves the problem of multiple solutions and no solution of the three-edge positioning method, and obtains an approximate solution with smaller error.

[0036] 2. Construct a redundant antenna array, analyze the impact of the number and layout of antennas in the array on the calculation results and ease of use, obtain the optimal antenna layout scheme, and reduce the impact of fading.

[0037] 3. A density-based noisy clustering method is used to fuse multiple measurement results, eliminating the interference of noise signals and unstable arcing data, thereby further improving the positioning accuracy. Attached Figure Description

[0038] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0039] Appendix Figure 1 This is a schematic diagram of the arc EMR signal spectrum before and after the occurrence of the arc fault, as proposed in this invention.

[0040] Appendix Figure 2 This is a schematic diagram illustrating the trilateration method proposed in this invention.

[0041] Appendix Figure 3 This is a schematic diagram of the positioning model proposed in this invention;

[0042] Appendix Figure 4 This is a schematic diagram of the antenna layout proposed in this invention;

[0043] Appendix Figure 5 This is a schematic diagram showing the location of the arc fault occurrence point (arc generator) proposed in this invention;

[0044] Appendix Figure 6 This is a schematic diagram illustrating the possible sampling stages of the arc EMR signal proposed in this invention;

[0045] Appendix Figure 7 This is a schematic diagram illustrating the results of using the DBSCAN clustering algorithm proposed in this invention;

[0046] Appendix Figure 8 This is a schematic diagram illustrating the principle of the present invention. Detailed Implementation

[0047] As shown in the figure, the direct current arc fault positioning method based on redundant antenna array and ellipse algorithm first filters the electromagnetic radiation signal EMR collected by the antenna during arc burning, calculates the root mean square value as the received signal strength indicator RSSI, and constructs a neural network model, which predicts the distance from the fault point to the receiving point based on the RSSI value and the irradiance; secondly, a redundant antenna array is constructed to analyze the influence of different number and layout of antennas on the positioning accuracy; thirdly, after determining the optimal antenna layout, the three antennas with the strongest signal are selected, and the antenna coordinates and predicted distance of the fault point are input into the improved ellipse positioning method based on the trilateration method to obtain the arc fault position; finally, the density-based noise clustering method DBSCAN is used to fuse multiple measurement results, eliminate interference and determine the final fault coordinates.

[0048] The method is used for direct current arc fault positioning of a photovoltaic system, and comprises the following steps:

[0049] Step S01: using a single antenna, collecting arc EMR signals at different distances and different irradiance at a preset sampling rate, and obtaining RSSI values after filtering processing;

[0050] Step S02: using RSSI and irradiance to train a neural network model to predict the output distance;

[0051] Step S03: establishing a Cartesian coordinate system with the photovoltaic array as a plane, determining the number and layout of the antennas of the redundant antenna array, and placing the antennas;

[0052] Step S04: converting the arc EMR signals collected by the antenna array into RSSI, obtaining the average value after multiple time windows, and selecting the antennas according to the size order of RSSI;

[0053] Step S05: inputting the RSSI of the three antennas and the corresponding irradiance into the trained neural network model to obtain the predicted distance from the arc to the antenna array, and using the ellipse positioning algorithm to obtain the arc fault point position;

[0054] Step S06: using the DBSCAN algorithm for clustering analysis to further reduce the error and obtain the final coordinates of the arc fault point position, and completing the arc fault positioning.

[0055] The construction method of the neural network model is: connecting the arc generator between the photovoltaic components, sampling when the arc is stable, sampling at a default sampling frequency of 200 kHz; gradually increasing the distance between the antenna and the arc generator at a default step of 0.5 meters, collecting a large amount of arc EMR signal data at different irradiance and distances with one antenna, which is used as the training set for constructing the neural network model;

[0056] In the test set of the neural network model, in the data acquisition process, the antenna is placed according to the preset antenna layout, and the EMR signals under different irradiance and distance are collected by the above method with a non-fixed step length, and the distance range is the same as that of the test set.

[0057] In step S01, the antenna collects signals of the arc generating device connected to the photovoltaic module.

[0058] The arc length and current of the photovoltaic DC of the photovoltaic system can affect the arc EMR intensity, but the current has a greater impact. In step S01, the influence of the arc length is ignored, and the arc length of the arc generating device is used to approximate the arc length of the photovoltaic DC. The gap length between the electrodes is randomly set.

[0059] The arc generating device is used to generate a DC arc, and the DC arc forms different current signal waveforms and EMR model waveforms in the normal, starting combustion, stable combustion and extinguishing four stages. The current signal waveform has different characteristics in each stage, and the arc EMR signal waveform changes as follows: when the arc starts to burn, the EMR signal amplitude increases rapidly, and when the arc is extinguished, the EMR signal rapidly decreases to the normal state, and there is almost no transition state.

[0060] The method is used to quickly find the fault point after the arc fault occurs in the photovoltaic array. In step S01, the acquisition time and method of the arc EMR signal are selected to select the waveform in the stable combustion stage after the fault occurs as the positioning signal, that is, the arc of the arc generating device is sampled when the arc is stable.

[0061] In step S03, the setting method of the redundant antenna array is as follows: n antennas are arranged in the photovoltaic array area, each antenna has a corresponding coordinate, and the order number is S1, S2, …, Sn. n Wherein, n≥4. The first three antennas with the maximum RSSI are selected: RSSI i ,RSSI j ,RSSI k ,i,j,k∈n, such that RSSI i ≥RSSI j ≥RSSI k ≥RSSI m Wherein, m≠i, j, k, m∈n.

[0062] In step S05, the implementation steps of the ellipse algorithm include:

[0063] Step A1, arrange three antennas A, B, C and arc fault point T, input RSSI into neural network to fit distance, and known information can be determined as: coordinates of three antennas and distances from arc fault point to each antenna; Step A2, take antennas A, B, C as centers and distances from arc fault to each antenna as radii, respectively draw three circles, and the circles are expressed by the following formula. Wherein, ε n represents an ellipse (a circle is a special ellipse), x n represents the center, Q n represents the radius, n=A, B, C;

[0064] Step A3, sequentially find the union of each two of the three circles, and then find the minimum outer ellipse of the three unions to form three ellipses, which are expressed by the following formula, wherein, ∪ and represents that two sets are first operated and then belong to a new set together;

[0065]

[0066] Step A4, find the inner ellipse with the largest area in the intersection of the three ellipses in step A3, and the inner ellipse is expressed by the following formula.

[0067]

[0068] Step A5, finally, let the center coordinates of the ellipse ε μ (x μ , Q μ ) be the arc fault position.

[0069] In step S06, the input parameters of the DBSCAN algorithm are neighborhood radius Eps and threshold MinPts of the number of data objects in the neighborhood; and the output is density connected clusters. The number of sample points in the neighborhood radius R is greater than or equal to MinPts, which is a core point; a point that is not a core point but is in the neighborhood of a core point is a boundary point; neither a core point nor a boundary point is a noise point;

[0070] The processing flow of the DBSCAN algorithm for clustering analysis is as follows:

[0071] Step B1, arbitrarily select a point p, which is neither assigned to a class nor specified as a peripheral point, and calculate whether the selected data object point is a core point for parameters Eps and MinPts; if yes, find all density-reachable data object points from p to form a cluster;

[0072] Step B2, if the selected data object point p is an edge point, select another data object point;

[0073] Step B3, repeat step B1, step B2 step, until all points have the characteristics of the cluster or noise points, points in the cluster characteristics of the core point or edge point;

[0074] Step B4, select the cluster with more sample points, abandon the cluster with less sample points and noise points, refer to the K-means algorithm, calculate the mean of each sample point, and take it as the clustering center.

[0075] In step S01, the method for obtaining RSSI is: collecting EMR signals during stable combustion of the arc at a low sampling rate (for example, 200 kHz), and randomly cutting a plurality of waveforms of a certain length (the length is recommended to be greater than or equal to 5 ms) from the EMR signals; each waveform is first removed from the pulse interference point by using a filtering method, and then the root mean square value is calculated as the RSSI.

[0076] Embodiment 1:

[0077] In this example, arc EMR signals are collected under normal operating conditions and arc fault conditions, and the sampling frequency is 1 GHz. Then, the frequency domain part of the signal is observed by using fast Fourier transform, as shown in Figure 1 .

[0078] The arc frequency domain is mainly concentrated in 3-500 MHz. The filter frequency band of the selected band-pass filter is set within this interval, and the time domain signal in the frequency band is used as the data source for subsequent positioning. The photovoltaic direct current arc is modeled, and it is deduced that both the arc length and the current can affect the arc EMR intensity, but the current has a greater impact. Without considering the influence of the arc length, the arc length of the arc generator is used to approximate the arc length, and the electrode gap length is randomly set.

[0079] The method for obtaining RSSI is: collecting EMR signals during stable combustion of the arc at a sampling rate of 200 kHz, and randomly cutting a plurality of waveforms of 5 ms from the EMR signals. Each waveform is first removed from the pulse interference point by using a median filtering method, and then the root mean square value is calculated as the RSSI.

[0080] The traditional trilateration method is based on the distance measurement algorithm to obtain the distance between the unknown node and the three reference nodes, and then solve the positioning node coordinates according to the equation established by the distance formula. The coordinates of the three reference nodes, i.e. three antennas, are set as A(x1, y1), B(x2, y2) and C(x3, y3), and the coordinates of the unknown node, i.e. the arc fault point, are set as T(x o ,y o). If an arc fault occurs, the three antennas each receive the EMR signal emitted by the arc burning, and the results fitted by the ANN can obtain the distances from the arc fault point to the antennas A, B, and C, respectively, d1, d2, and d3. Knowing the positions of the three antennas and the distances from the arc fault point to the three antennas, the coordinates of the arc fault can be calculated. The ideal trilateration fitting results d1, d2, and d3 have no error, and the correct coordinates can be uniquely solved. Taking A, B, and C as the centers and d1, d2, and d3 as the radii, circles are drawn, and the three circles intersect at a point. The complex environment will affect the transmission effect of the electromagnetic wave, resulting in measurement errors. If the measured value is too large, there may be intersections between the circles, resulting in multiple solutions of the equation set; if the measured value is too small, there is no intersection between the circles, which will result in no solution of the above equation set. The four cases are shown in FIG. 8. Figure 2

[0081] Figure 3 The positioning model is shown, and the arc fault point emits EMR signals to the surrounding. There are obstacles between antennas S2 and S7 and the arc, and the signal is mainly affected by shadow fading, with a decrease in amplitude; there are obstacles between antenna S4 and the arc, and there may be multiple propagation paths, and the signal is mainly affected by multipath effect; antenna S6 is far away from the arc, and the signal is mainly affected by path loss. After comparison, S1, S3, and S5 are the three antennas with the strongest signals, and the proportion of direct signal received is large. The final positioning result is the arc occurrence point calculated by the algorithm shown in the figure, which may have a positional deviation from the actual arc occurrence point.

[0082] In addition, the positioning performance will also be affected by the antenna array arrangement. The symmetrical layout has relatively excellent performance and applicability, and is widely used in engineering. The arrangement of the redundant antenna array is discussed. Figure 4 Several commonly used arrangements are shown, and the naming rule is as follows: the front number represents the number of antennas, and the back letters R, D, and H represent rectangular, diamond, and hexagonal structures, respectively. For example, 4R represents a 4-antenna rectangular structure layout. The more the number of antennas, the more complex the calculation, and the time overhead will increase exponentially, but the accuracy improvement is not significant.

[0083] In this example, the layout of the six antennas can meet the requirements of actual application.

[0084] Example 2

[0085] ​In the present embodiment, the power of the photovoltaic module MPPT state is 270W, the voltage is 31.3V, and the current is 8.63A. The inverter model is SG6KTL-MT, and the starting voltage is 250V. The antenna used is an omnidirectional wideband antenna, which has a working frequency of 20-500MHz, a gain of-5-2dB, a noise figure of ≤7dB, an output impedance of 50Ω, an output standing wave of ≤2, and a length of 33cm. The amplifier used is a low-band noise radio frequency amplifier, which has a working frequency of 50MHz-6GHz and a gain of 20dB. The filter has a working frequency of 200-500MHz. The oscilloscope used for signal acquisition has a model of DSOX2024A and a bandwidth of 2GHz. In all experiments, the antenna and the AFG are located in the same plane.

[0086] The arc generator is connected between the photovoltaic modules, and the sample is taken when the arc is stable. The sampling frequency of the oscilloscope is set to 200kHz, and the irradiance range is 200-1100W / m2. The distance between the antenna and the arc generator is gradually increased by 0.5m steps. A large amount of arc EMR signal data under different irradiance and distance is collected by one antenna, and an ANN training set is formed.

[0087] The farthest distance from the antenna to the arc fault source of the photovoltaic array is 8.942m, and the farthest effective distance of the antenna to receive the EMR signal under low irradiance in the open area is 11.389m. Therefore, the range of the training sample is set to 0-9m. For the test set, the antenna is placed at the position of the antenna layout in the figure, and the EMR signal under different irradiance and distance is collected with a fixed step. The distance range is the same as that of the test set.

[0088] The arc fault occurrence point is set as shown in Figure 5 The 5R layout method is discussed and analyzed. First, the three strongest signals in the redundant antenna array are determined, and then the elliptical positioning algorithm is used to obtain the positioning result. The actual coordinates of the arc are represented by ×, and the predicted coordinates obtained by the algorithm are marked by circles. After the photovoltaic array is connected to the grid, the arc is simulated, and the data is collected when the arc is stable. The average value is obtained by taking 10 time windows of 5ms within 2s, and the experiment is repeated 10 times under different irradiance. The experimental results are shown in Figure 6 In most experimental scenarios, the three antennas receiving the strongest EMR signal are also the closest to the arc. However, due to the prediction error of ANN, shadow fading, multipath effect, and site error, the 5R layout selects 2 groups of antennas at position ⑥.

[0089] The sampling stages of the EMR signal with probability occurrence are analyzed: the first stage (I) contains data of two stages of arc starting combustion and arc stable combustion; the second stage (II) contains data of the arc stable combustion stage; the third stage (III) contains data of two stages of arc stable combustion and arc extinguishing; and the fourth stage (IV) contains data of three stages of arc starting combustion, arc stable combustion and arc extinguishing, as shown in FIG. 4. Figure 6 For example, the position ④ contains two noise points, which means that in the 10 positioning experiments, 8 times are normal positioning, and 2 times are affected by noise and deviate from the fault points. After clustering, as shown in FIG. 5, the clustering algorithm can better identify the core store and the noise point: there are two data blocks in the figure, and two clustering center points are obtained after clustering. The center point of the largest clustering cluster is taken as the final positioning result, and the positioning error at this time is 0.468 m, which is reduced by 0.521 m compared with before clustering. Figure 7

[0090] The above only describes the preferred embodiments of the present application, and any changes made according to the technical solutions of the present application, as long as the generated function does not exceed the scope of the technical solutions of the present application, should be within the scope of the present application.​

Claims

1. A DC arc fault location method based on a redundant antenna array and an ellipse algorithm, characterized in that: The method first filters the electromagnetic radiation signal EMR collected by the antenna when the arc is burning, calculates the root mean square value as the received signal strength indicator RSSI, and constructs a neural network model, which predicts the distance from the fault point to the receiving point based on the RSSI value and the irradiance; secondly, a redundant antenna array is constructed, and the influence of different number and layout of antennas on the positioning accuracy is analyzed; Thirdly, after determining the optimal antenna layout, the three strongest signals are selected, the antenna coordinates and the predicted distance of the fault point are input into the improved elliptical positioning method based on the three-edge positioning method, and the arc fault position is obtained; finally, the density-based noise clustering method DBSCAN is used to fuse multiple measurement results, eliminate interference and determine the final fault coordinates; The method is used for direct current arc fault positioning of a photovoltaic system, and includes the following steps: Step S01: using a single antenna, collecting arc EMR signals at different distances and different irradiance at a preset sampling rate, and obtaining RSSI values after filtering; Step S02: using RSSI and irradiance to train a neural network model to predict the output distance; Step S03: establishing a Cartesian coordinate system with the photovoltaic array as the plane, determining the number and layout of the redundant antenna array, and placing the antennas; Step S04: converting the arc EMR signals collected by the antenna array into RSSI, obtaining the average value after multiple time windows, and selecting the antennas in order of RSSI size; Step S05: inputting the RSSI of the three antennas and the corresponding irradiance into the trained neural network model to obtain the predicted distance from the arc to the antenna array, and using the elliptical positioning algorithm to obtain the arc fault point position; step S06: using the DBSCAN algorithm for clustering analysis to further reduce the error and obtain the final coordinates of the arc fault point position, and completing the arc fault positioning.

2. The method of claim 1, wherein the method is based on a redundant antenna array and an ellipse algorithm. The construction method of the neural network model is: connecting the arc generator between the photovoltaic components, sampling when the arc is stable, and sampling at a default sampling frequency; gradually increasing the distance between the antenna and the arc generator at a default step, and collecting a large amount of arc EMR signal data at different irradiance and distances with one antenna for training set of the neural network model; When constructing the test set of the neural network model, during the data collection process, the antenna is placed according to the preset antenna layout, and the EMR signals at different irradiance and distances are collected in the same way as above with a non-fixed step, and the distance range is the same as the test set.

3. The method of claim 1, wherein the method is based on a redundant antenna array and an ellipse algorithm. In step S01, the antenna collects signals from the arc generating device connected between the photovoltaic components.

4. The method of claim 3, wherein: The arc length and current of the photovoltaic direct current can affect the arc EMR intensity, but the current has a greater impact, and the influence of the arc length is ignored in step S01, and the arc length of the arc generating device is used to approximate the arc length of the photovoltaic direct current, and the gap length between the electrodes is randomly set; The electric arc generating device is used for generating a direct current arc, and the direct current arc forms different current signal waveforms and EMR model waveforms in normal, starting combustion, stable combustion and extinguishing four stages, the current signal waveforms have different characteristics in each stage, and the arc EMR signal waveform changes as follows: when the arc starts to burn, the EMR signal amplitude rapidly increases, when the arc is extinguished, the EMR signal rapidly decreases to the normal state, and there is almost no transition state.

5. The method of claim 3, wherein: the method further comprises: determining a first ellipse based on the first set of antenna signals; determining a second ellipse based on the second set of antenna signals; and determining a third ellipse based on the third set of antenna signals. The method is used for quickly finding a fault point after an electric arc fault occurs in a photovoltaic array, and in step S01, the acquisition time and method of the arc EMR signal are used to select a waveform in the stable combustion stage after the fault occurs as a positioning signal, that is, sampling when the arc of the arc generating device is stably combusted.

6. The method of claim 1, wherein: The setting method of the redundant antenna array in the step S03 is: arranging n antennas in the photovoltaic array region, each antenna has corresponding coordinates, and the order number is: S1, S2, …, S n wherein n≥4; Select the top three antennas with the largest RSSI: RSSI i RSSI j RSSI k i,j,k∈n, such that RSSI i ≥RSSI j ≥RSSI k ≥RSSI m where m≠i,j,k, m∈n.

7. The method of claim 1, wherein: the method further comprises: determining a first ellipse based on the first set of antenna signals; determining a second ellipse based on the second set of antenna signals; and determining a third ellipse based on the third set of antenna signals. In step S05, the implementation steps of the ellipse algorithm specifically include: Step A1, arranging three antennas A, B and C and an arc fault point T, and after the RSSI is input into the neural network to fit the distance, the known information is that the coordinates of the three antennas and the distances from the arc fault point to each antenna are determined; Step A2, with the antenna A, B, C as the center, the respective distance of the antenna and the arc fault as the radius, three circles are made respectively, which are expressed by the following formula; ε n (x n -x n )2+ (x o -x o )2=1 n T Q n -1 (x o -x n )≤1}; wherein, ε n represents the ellipse made, x n represents the center, Q n represents the radius, n=A, B, C; x o is the horizontal coordinate of the arc fault point;​ Step A3, find the union of each two of the three circles, then find the minimum circumscribed ellipse of the three unions, form three ellipses, which are expressed by the following formula, where ∪ and denote the set operation of the two sets first, and then belong to the new set together; Step A4, the largest internal ellipse of the intersection of the three ellipses in step A3 is obtained, and is expressed by the following formula: Step A5, Finally, let the ellipse center coordinate ε μ (x μ , Q μ ) be the arc fault location.

8. The method of claim 1, wherein the method is based on a redundant antenna array and an ellipse algorithm. In step S06, the input parameters of the DBSCAN algorithm are the neighborhood radius Eps and the threshold value MinPts of the number of data objects in the neighborhood, the output density connected cluster, the number of sample points in the neighborhood radius R is greater than or equal to MinPts, and the point is a core point; the point which is not a core point but in the neighborhood of a certain core point is a boundary point; neither a core point nor a boundary point is a noise point; The processing flow of the DBSCAN algorithm for clustering analysis is as follows: Step B1, an arbitrary point p is selected, the point is neither assigned to a class nor specified as a peripheral point, and whether the selected data object point is a core point is calculated for the parameters Eps and MinPts; if yes, all the data object points density reachable from p are found to form a cluster; Step B2, if the selected data object point p is an edge point, another data object point is selected; Step B3, steps B1 and B2 are repeated until all points have the characteristics of being in a cluster or being a noise point, and the characteristics of the point in the cluster are a core point or an edge point; Step B4, a cluster with more sample points is selected, and a cluster with less sample points and noise points are abandoned, and the mean value of each sample point is calculated as a clustering center according to the K-means algorithm.

9. The method of claim 1, wherein: the method further comprises: determining a plurality of ellipse parameters for each of the plurality of antennas; and determining a plurality of ellipse parameters for each of the plurality of antennas. In step S01, the method for acquiring RSSI is as follows: collecting the EMR signal during the stable combustion of the arc at a lower sampling rate, and randomly cutting a predetermined length of waveform from the EMR signal; each waveform is first removed from the pulse interference point by filtering, and then the root mean square value is calculated as RSSI.

Images