A Fault Location Method for Distribution Lines Based on Spatial Data Fusion

Abstract

The present invention provides a method for fault location of distribution lines based on spatial data fusion. The present invention first performs simulation on the entire distribution network to be detected, then conducts equidistant fault tests on all lines, then converts and fuses the voltage and current signals collected at the fault points of the entire distribution network, and then performs simulation training by combining the method of convolutional neural network (CNN). Finally, the faults in the actual distribution network can be accurately located.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault location, and particularly to a method for locating distribution line faults based on spatial data fusion. Background Art

[0002] When a fault occurs in a distribution network line, if the fault cannot be accurately judged and located and eliminated in time, it may cause a power outage accident and further affect production and life.

[0003] The faults in the distribution network can be divided into two aspects: fault type determination and identification, and location. In the research of fault identification, Hu et al. [1] proved the effectiveness of combining wavelet singularity, and used transient voltage entropy and support vector machine (SVM) to identify fault types. Guo et al. [2] extracted the distribution parameters of the singular spectrum of the eigenvector from the image and its calculation, and used these eigenvectors as the input of the multi-level support vector machine to achieve fault identification. However, when a high-resistance grounding fault occurs, its singularity spectrum characteristics do not show obvious changes, resulting in difficult fault identification. Sun et al. [3] used empirical mode decomposition (EMD) to construct eigenvectors, and a fuzzy neural network (FNN) based on binary ant colony algorithm (BACA) can perform fault identification. However, EMD is prone to mode mixing problems, and there are problems of low efficiency and local optimum in the ant colony algorithm. Zhang et al. [4] combined wavelet transform with adaptive wavelet transform to identify system faults based on a network-based fuzzy inference system. The method used is the transient characteristics for distinguishing fault types. However, it is not applicable to noise interference and asynchronous interference sampling, which will obtain incorrect fault identifiers. In [5], a classification decision is established to identify faults and classify the fault types through the amplitude and phase difference of the three-phase current sequence components. However, it cannot effectively locate the fault types in the case of unbalanced loads. Gopakumar et al. [6] used fast Fourier transform (FFT) to extract the eigenvectors of the fault signal, and support vector machine can be used to distinguish the fault types of the feeder. It may get incorrect results, that is, there is a leakage problem in the FFT spectrum.

[0004] Among the faults in the distribution network, single-phase grounding faults account for more than 80% of all faults. The research results of fault location include: the fault location method for the distribution network based on the active component of zero-sequence current. The disadvantage of this method is that the active component in the system is weak, which is prone to misjudgment. Based on the traveling wave principle, various methods combined with traveling waves for line selection and location are proposed, which is the most common method at present. However, the traveling wave method has the problem of difficult determination of the traveling wave head, and the traveling wave method requires the moment when the reflected wave head appears to be obvious in order to accurately locate. When it is not obvious, the positioning accuracy is significantly affected. The ranging method based on the transient signal comparison method. The disadvantage of this type of method is that the duration of the transient signal is short, and it requires the equipment to be able to quickly and accurately extract the fault transient information. The artificial intelligence method combining neural network and other algorithms. Although this type of method can use advanced intelligent algorithms to solve the problem of distribution network fault location, this type of method requires a large number of samples to be collected in advance to train the neural network and has high requirements for the sample data.

[0005] In the above methods, most of them, the extraction of the fault feature vector is carried out after activities under artificial conditions, and then the identification of the feeder fault type is realized through the corresponding pattern recognition model. These traditional methods may have a great impact. The extraction process of the fault feature vector is relatively complex and requires robust research experience. There are also obvious inaccuracies in fault location, especially when the neutral point is not grounded and a high-resistance single-phase grounding fault occurs, the fault location accuracy is relatively low.

[0006] [1]Hu,W.,Li,Y.,Cao,Y.J.,et al.:'Fault identification based on LOF andSVM for smart distribution network',Electr.Power Autom.Eq.,2016,36,(6),pp.7-12

[0007] [2]Guo,M.F.,You,L.X.,Hong,C.,et al.:'Identification method ofdistribution network faults based on singular value of LCD-Hilbert spectrumsand multilevel SVM',High Volt.Eng.,2107,43,(4),pp.1239-1247

[0008] [3]Sun, P., Cao, Y.C., Liu, Y., et al.: 'Fault classification technique for power distribution network using binary ant colony algorithm and fuzzy neural network', High Volt. Eng., 2016, 41, (7), pp.2063-2072

[0009] [4]Zhang, J., He, Z.Y., Lin, S., et al.: 'An ANFIS-based fault classification approach in power distribution system', Int. J. Elec. Power, 2013, 49, pp.243-252

[0010] [5]Das, B.: 'Fuzzy logic-based fault-type identification in unbalanced radial power distribution system', IEEE Trans. Power Deliv., 2006, 21, (1), pp.278-285

[0011] [6]Gopakumar, P., Reddy, M.J.B., Mohanta, D.K.: 'Adaptive fault identification and classification methodology for smart power grids using synchronous phasor angle measurements', IET Gener. Transm. Distrib., 2015, 9, (2), pp.133-145 Summary of the Invention

[0012] The object of the present invention is to provide a fault location method for distribution lines based on spatial data fusion, so as to solve the foregoing problems existing in the prior art.

[0013] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0014] A fault location method for distribution lines based on spatial data fusion, characterized by comprising the following steps:

[0015] S1. Establish a simulation signal acquisition system consistent with the structure of the power distribution network to be detected, including a topology map and a fault indicator layout map consistent with the structure of the power distribution network to be detected. Set multiple fault points on the topology map, and use fault indicators to obtain multiple sampled voltage and current signal samples X=(x1, x2, …, x n ) corresponding to different fault types at each fault point;

[0016] S2. Normalize all the sampled voltage and current signal samples X=(x1, x2, …, x n ) collected by all fault indicators at the time of fault occurrence at each fault point to obtain the normalized That is, The elements of are

[0017] where,

[0018] S3. Use the Gram angular field matrix to perform time-space transformation on the normalized sampled voltage and current signal samples , that is, convert the time waveform of the signal into polar coordinate form, so that the time signal is converted into the form of a spatial image, thereby reflecting the amplitude and frequency information of the signal in the image space at the same time;

[0019] S4. Perform Gram transformation on the voltage and current signal samples collected by the fault indicators at all sampling points when a fault occurs, respectively, to obtain several single-point image samples. Use the multi-point spatial information fusion method to fuse the information of all single-point image samples to obtain the fused sample data of this fault point;

[0020] S5. Repeat the process in steps S2 - S4 to obtain the fused sample data of multiple sampling points when each fault point set on the topology map has a fault, and integrate them into a sample data set;

[0021] S6. Use a convolutional neural network to train the fused sample data set until the network converges. At this time, establish a state feature library, and the state feature library includes the fault point location and the fault type;

[0022] S7. Using the established state feature library as a reference object for fault identification, input the current and voltage signals after a fault into the neural network, and the fault point and fault type can be quickly located according to the reference data information in the state feature library.

[0023] Preferably, the fault points in step S1 are set at equal intervals on the power grid line, and the interval between each two fault points is 100 - 200 meters.

[0024] Preferably, there are three connection methods between the neutral point of the distribution network and the ground in step S1, namely: small resistance grounding, arc suppression coil grounding, and suspension. Therefore, it is necessary to collect fault point signals under different grounding methods.

[0025] Preferably, in step S3, the Gram angle field matrix is used to perform time-space transformation on the normalized voltage and current signal samples The Gram matrix form is

[0026]

[0027] where The calculation formula for converting the one-dimensional time waveform into a two-dimensional image form:

[0028]

[0029] That is, the pixel point

[0030]

[0031] Preferably, when performing multi-point spatial information fusion in step S4, for the fault indicator that can collect the over-limit information of the zero-sequence current, that is, the zero-sequence current component of the faulty line increases significantly, the proportion of such samples is increased during fusion.

[0032] Preferably, a small-signal data non-linear transformation is performed on such samples, and the formula x i = 5lg(y i ) is constructed to obtain the non-linear sample X; the normalized samples of the current time waveforms at each point collected are fused into a new sample image, and its formula is:

[0033]

[0034] where r, k, n, and l respectively represent the subscript values of the elements in the pixel matrix, M represents the total number of sampling points, and N represents the number of simulated faults.

[0035] Preferably, r, k, n, and l are determined according to the actual image size, so that the synthesized image sample is close to the structure of a square matrix, and the spatial occupancy of the picture is relatively small and reasonable.

[0036] Preferably, if the number of pixels in the image of a single sample is different, in order to form a matrix structure, the sample with fewer image pixels can be filled with 0 elements.

[0037] The beneficial effects of the present invention are:

[0038] The present invention provides a distribution line fault location method based on spatial data fusion. The present invention first simulates the entire distribution network to be detected, then performs equidistant fault tests on all lines, then converts and fuses the voltage and current signals collected from the fault points of the entire distribution network, and then combines the convolutional neural network (CNN) method for simulation training, and finally can accurately locate the faults of the distribution network in reality. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of the distribution line fault location method based on spatial data fusion provided in Example 1;

[0040] Figure 2 is a schematic diagram of a distribution network topology provided in Example 2;

[0041] Figure 3 is a schematic diagram of signal waveforms collected by 8 fault indicators in Example 2;

[0042] Figure 4 is a single-point sample after signal transformation using 8 fault indicators collected in Example 2;

[0043] Figure 5 It is the image after multi-point sample fusion in Example 2. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] Example 1

[0046] This embodiment provides a distribution line fault location method based on spatial data fusion, which is characterized by comprising the following steps:

[0047] S1, establish a simulation signal acquisition system consistent with the structure of the distribution network to be tested, including a topology map and a fault indicator layout map consistent with the structure of the distribution network to be tested, set multiple fault points on the topology map, and use the fault indicator to obtain multiple sampling point voltage and current signal samples corresponding to different fault types of each fault point X = (x1, x2, ..., x n );

[0048] S2, obtain different voltage and current signal samples of each fault point X=(x1,x2,…,x n ) is normalized to obtain the normalized Right now The elements of

[0049] Among them,

[0050] S3. Use the Gram angular field matrix to perform time-space transformation on the normalized voltage and current signal samples That is, convert the time waveform of the signal into polar coordinate form, convert the time signal into the form of a spatial image, so as to simultaneously reflect the amplitude and frequency information of the signal in the image space;

[0051] S4. For the voltage and current signal samples collected by the fault indicators at all sampling points when a fault occurs, perform Gram transformation on each of them to obtain several single-point image samples, and use the multi-point spatial information fusion method to fuse the information of all single-point image samples to obtain the fused sample data of the fault point;

[0052] S5. Repeat the process in steps S2 - S4 to obtain the fused sample data of multiple sampling points when each fault point set on the topology map has a fault, and integrate them into a sample data set;

[0053] S6. Use a convolutional neural network to train the fused sample data set until the network converges. At this time, establish a state feature library, and the state feature library includes the fault point location and the fault type;

[0054] S7. Using the established state feature library as a reference object for fault identification, input the current and voltage signals after a fault into the neural network, and the fault point and fault type can be quickly located according to the reference data information in the state feature library.

[0055] In this embodiment, the fault points in step S1 are set at equal intervals on the power grid line, and the interval between each two fault points is 100 - 200 meters. There are three connection methods between the neutral point of the distribution network and the ground in step S1, namely: small resistance grounding, arc suppression coil grounding, and suspension. Therefore, it is necessary to collect fault point signals under different grounding methods.

[0056] In step S3 of this embodiment, using the Gram angular field matrix to perform time-space transformation on the normalized voltage and current signal samples The Gram matrix form is

[0057]

[0058] Among them, The calculation formula for converting the one-dimensional time waveform into a two-dimensional image form:

[0059]

[0060] That is, the pixel point

[0061]

[0062] When performing multi-point spatial information fusion in step S4 of this embodiment, for the fault indicator that can collect the information of zero-sequence current exceeding the limit, that is, the zero-sequence current component of the faulty line increases significantly, the proportion of such samples is increased during fusion.

[0063] In this embodiment, a non-linear transformation of small-signal data is performed on such samples, and the formula x i = 5lg(y i ) is constructed to obtain the non-linear sample X; the normalized samples of the current time waveforms at each point collected are fused into a new sample image, and its formula is:

[0064]

[0065] where r, k, n, and l respectively represent the subscript values of the elements in the pixel matrix, M represents the total number of sampling points, and N represents the number of simulated faults.

[0066] r, k, n, and l are determined according to the actual size of the image, so that the synthesized image sample is close to the structure of a square matrix, and it is more reasonable that the space occupied by the picture is smaller. If the number of pixels in the image of a single sample is different, in order to form a matrix structure, the sample with fewer image pixels can be filled with 0 elements.

[0067] Embodiment 2

[0068] In this embodiment, a distribution network topology structure is provided. When there are multiple distribution lines, it is necessary to set the fault positions for each line (such as L1-L4) respectively to obtain fault signal samples, and encode them separately according to different lines. In this way, the encoding can correspond to the location where the fault occurs. As Figure 2 shown, F1, F2, F2-21, F2-22, F2-31, F2-32, F3, and F4 respectively represent the fault indicators at each point (branch line) on the distribution line, and are used to collect the three-phase current values on the line; the figure respectively shows that grounding faults have occurred at points S1 (1.25 km) on line L2 (2 km in length), S2 (0.6 km) on line L2-1 (3 km in length), S3 (0.9 km) on line L2-2 (2 km in length), and S4 (1.8 km) on line L4 (3 km in length), etc. Only four fault point positions are given in this embodiment. In fact, fault points should be set every 100-200 meters. When a fault occurs at each position, the voltage and current values on the fault indicators F1, F2, F2-21, F2-22, F2-31, F2-32, F3, and F4 will change accordingly, so as to obtain different sample data.

[0069] According to the previous introduction, since the entire distribution network line can be regarded as an active network, the voltage and current information collected at each location represents the response result of that point to the voltage source. When a ground fault occurs at a certain point in the network, it is equivalent to changing the network structure, which will inevitably cause changes in the responses at all points in the network. Therefore, collecting the voltage and current information at each point in the distribution network can reflect the state of the network, and thus these information can be used to identify and locate faults.

[0070] STEP1: Use devices such as fault indicators to collect the voltage and current signal samples at each point on the distribution network line. The voltage and current signal waveforms under normal and fault conditions need to be collected respectively. Assume it is X=(x1, x2, …, x n ), n = 8, where the zero-sequence current signals of eight sampling points (fault indicators) are as Figure 3 shown.

[0071] STEP2: Normalize the sample X=(x1, x2, …, x n ), n = 8 to obtain the normalized That is, the elements of are

[0072] where,

[0073] STEP3: Use the Gram angular field matrix to perform time-space transformation on the collected voltage and current signals. That is, convert the time waveform of the signal into polar coordinate form, so that the time signal is converted into the form of a spatial image, thereby simultaneously reflecting the amplitude and frequency information of the signal in the image space. The single-point samples collected by eight fault indicators are as Figure 4 shown.

[0074] Among them, the form of the Gram matrix used is

[0075]

[0076] where, The calculation formula for converting the one-dimensional time waveform into a two-dimensional image form:

[0077] That is, the pixel point

[0078]

[0079] STEP4: Perform multi-point spatial information fusion. The voltage and current signal sample images collected at each point are subjected to Gram transformation to obtain single-point image samples, and information fusion is performed on them. The fault indicator can collect the information of zero-sequence current over-limit, that is, the zero-sequence current component of the faulty line increases significantly. When fusing, the proportion of such samples is increased.

[0080] Perform non-linear transformation on the small-signal data of such samples, and construct the formula x i = 5lg(y i ), to obtain the non-linear sample X. Using non-linear transformation can enhance the role of small signals and improve the correct discrimination rate.

[0081] The normalized samples of the current time waveforms collected at each point are fused into a new sample image, and its formula is:

[0082]

[0083] In the case, 8 fault indicators are selected, namely: F1, F2, F 2-21 , F 2-22 , F 2-31 , F 2-32 , F3, F4 are information sampling points. The image size of a single sampling point in a single fault sample is 32×32. The information collected at multiple points is fused together in 2 rows and 4 columns to form an image with a size of 64×128, as Figure 5 shown.

[0084] STEP5: Use the convolutional neural network (CNN) to train the samples to make the network converge and establish a state feature library.

[0085] The parameters of the CNN used in this embodiment are: the kernel function is 5×5; there are 3 convolutional layers C; 3 pooling layers S; and the output of the fully connected layer.

[0086] STEP6: Use the trained and converged CNN network for fault identification. When the distribution network line is normal or a fault occurs, the CNN network gives a number, and the line position corresponding to the number is given. In the fault state, the result of the fault point position can be given.

[0087] By adopting the above technical solutions disclosed in the present invention, the following beneficial effects are obtained:

[0088] The present invention provides a method for fault location of distribution lines based on spatial data fusion. First, the entire distribution network to be detected is simulated. Then, equidistant fault tests are carried out on all lines. Next, the voltage and current signals collected at the fault points of the entire distribution network are converted and information-fused, and then simulated training is carried out by combining the method of convolutional neural network (CNN). Finally, the faults in the actual distribution network can be accurately located.

[0089] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A distribution line fault location method based on spatial data fusion, characterized in that, Including the following steps: S1. Establish a simulation signal acquisition system consistent with the structure of the distribution network to be detected, including a topology map and a fault indicator layout map consistent with the structure of the distribution network to be detected. Set multiple fault points on the topology map, and use fault indicators to obtain multiple sampled voltage and current signal samples X=(x1, x2, …, x n ) corresponding to different fault types at each fault point; S2. For each fault point, normalize all the sampled voltage and current signals X = (x1, x2, …, x n ) obtained at the fault occurrence time of the fault indicator to obtain the normalized That is The elements of are Among them, S3. Use the Gram angular field matrix to process the normalized voltage and current signal samples for time-space transformation, that is, convert the time waveform of the signal into the polar coordinate form, so that the time signal is converted into the form of a spatial image, thereby simultaneously reflecting the amplitude and frequency information of the signal in the image space; S4. For the voltage and current signal samples collected by the fault indicators at all sampling points during a fault occurrence, perform Gram transformation on each of them to obtain several single-point image samples, and use the multi-point spatial information fusion method to fuse the information of all single-point image samples to obtain the fused sample data of the fault point; S5. Repeat the process in steps S2 - S4 to obtain the fused sample data of multiple sampling points when a fault occurs at each fault point set on the topology map, and integrate them into a sample data set; S6. Use a convolutional neural network to train the fused sample data set until the network converges. At this time, establish a state feature library, and the state feature library includes the fault point location and the fault type; S7. Using the established state feature library as the reference object for fault identification, input the current and voltage signals after a fault into the neural network, and the fault point and fault type can be quickly located according to the reference data information in the state feature library. The fault point setting in step S1 is to set fault points equidistantly on the power grid line, with a distance of 100-200 meters between each fault point; the time-space transformation of the normalized voltage and current signal samples using the Gram angular field matrix in step S3 is performed, and the form of the Gram angular field matrix is Among them, Calculation formula for converting one-dimensional time waveform into two-dimensional image form: That is, pixel points When performing multi-point spatial information fusion in step S4, for the fault indicator that can collect the information of zero-sequence current over-limit, that is, the zero-sequence current component of the faulty line increases significantly, the proportion of such samples is increased during fusion; perform non-linear transformation of small-signal data on such samples, and construct the formula x i = 5lg(y i ), to obtain the non-linear sample X; the normalized samples of the current time waveforms at each point collected are fused into a new sample image, and its formula is: Among them, r, k, n, and l respectively represent the subscript values of the elements in the pixel matrix, M represents the total number of sampling points, and N represents the number of simulated faults; r, k, n, and l are determined according to the actual image size, so that the synthesized image sample is close to the structure of a square matrix, and the spatial occupancy of the picture is relatively small and reasonable.

Images