Power distribution network fault distance measurement method and system based on wide-area synchronous measurement data

Abstract

The invention discloses a power distribution network fault distance measurement method and system based on wide-area synchronous measurement data, and the method comprises the steps: sequentially carrying out the wavelet packet decomposition and mask operation of a voltage phasor in a set time period before a fault moment when a fault occurs in a power distribution network, and obtaining a pre-training data set; adjusting a pair of mask variables to control the width of a selected region in the pre-training data set; performing traveling wave recognition on the input data under the sliding window by using a pre-trained traveling wave recognition model, and calculating the traveling wave recognition accuracy under the sliding window; the maximum traveling wave recognition accuracy under the sliding window is used as an objective function, and under the constraint condition that the sum of a pair of mask variables is not larger than the input time sequence data length of the traveling wave recognition model, the objective function is iteratively solved to obtain the maximum traveling wave recognition accuracy under the sliding window and the corresponding mask variable optimal value; therefore, the timestamp and the traveling wave transmission speed corresponding to the selected area in the input data are determined, the fault distance measurement result is determined, and the fault distance measurement accuracy of the power distribution network is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power distribution, and in particular relates to a distribution network fault distance measurement method and system based on wide-area synchronous measurement data. Background Art

[0002] With the rapid development of modern power systems, the scale and complexity of distribution networks are increasing. Rapid and accurate fault location is crucial to ensure the stable operation of power systems and reduce power outages. With the widespread application of wide area measurement systems (WAMS) and synchronized phasor measurement units (PMU) technology, high-frequency data collection can be performed on distribution networks to improve the accuracy of fault location in distribution networks.

[0003] In the prior art, the distribution network fault distance measurement method adopts the traveling wave method to perform traveling wave detection. However, since the acquisition frequency of wide-area synchronous measurement data is too high, it will lead to the generation of a large amount of data. Although it can improve the response strength of the traveling wave data, it will also bring a large amount of interference data. Therefore, how to quickly complete the distribution network fault distance measurement based on wide-area synchronous measurement data has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides a distribution network fault ranging method and system based on wide-area synchronous measurement data, firstly pre-training a traveling wave recognition model, and adding mask variables to the pre-trained traveling wave recognition model to improve the accuracy of the pre-trained traveling wave recognition model in traveling wave recognition, and by iteratively optimizing the mask variables, reducing the computing power requirement brought about by the improvement of the accuracy in traveling wave recognition, and after determining the best recognition result of the traveling wave, reducing the error in distribution network fault detection using traveling waves through mask variables, thereby improving the accuracy of distribution network fault ranging.

[0005] The present invention adopts the following technical solution.

[0006] The present invention proposes a distribution network fault location method based on wide-area synchronous measurement data, and deploys PMU in the distribution network, including:

[0007] Collect the voltage and current phasors measured by all PMUs at each sampling timestamp;

[0008] A traveling wave recognition model is established based on the CNN-LSTM network architecture. When it is determined that a fault occurs in the distribution network, the voltage phasor in the set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, wherein a pair of mask variables are adjusted in the mask operation to control the width of the selected area in the pre-training data set. The traveling wave recognition model is pre-trained using the pre-training data set.

[0009] The pre-trained traveling wave recognition model is used to perform traveling wave recognition on the input data under the sliding window, and the traveling wave recognition accuracy under the sliding window is calculated; the maximum traveling wave recognition accuracy under the sliding window is taken as the objective function, and under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, the objective function is iteratively solved to obtain the maximum traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable;

[0010] According to a pair of optimal values ​​of mask variables corresponding to the maximum value of the maximum traveling wave recognition accuracy under all sliding windows, the selected area in the input data is determined, and the fault location result is determined according to the timestamp and traveling wave transmission speed corresponding to the selected area in the input data.

[0011] Preferably, when the amplitude of the voltage phasor at sampling timestamp t is not within a set voltage range, or the amplitude of the current phasor at any sampling timestamp t is not within a set current range, it is determined that a fault occurs in the distribution network at sampling timestamp t.

[0012] Preferably, the set voltage range is [60, 260]V, and the set current range is [10, 200]A.

[0013] Preferably, when it is determined that a fault occurs in the distribution network, wavelet packet decomposition and mask operation are performed on the voltage phasors in a set time period before the fault moment in sequence to obtain a pre-training data set, including:

[0014] The wide-area synchronous measurement data between sampling timestamp tn and sampling timestamp t are obtained to construct a voltage phasor segment sequence CA and a current phasor segment sequence CB for traveling wave detection; wherein n is the ratio of the abnormal line length of the distribution network to the traveling wave transmission speed;

[0015] The wavelet packet algorithm is used to decompose the voltage phasor segment sequence CA to obtain the decomposed components, and all the decomposed components are combined into a voltage two-dimensional matrix DA.

[0016] When performing a mask operation on the voltage two-dimensional matrix DA, a pair of mask variables is adjusted to control the width of the selected region in the voltage two-dimensional matrix DA, thereby forming a pre-training data set.

[0017] Preferably, in the mask operation, a pair of mask variables are adjusted to control the width of the selected region in the pre-training dataset, including:

[0018] Use a binary matrix of the same size as the voltage two-dimensional matrix DA to perform a mask operation on the voltage two-dimensional matrix DA; in the binary matrix, the elements at positions corresponding to the regions to be selected in the voltage two-dimensional matrix DA are 1, and the elements at other positions are 0;

[0019] The binary matrix used for mask operation corresponds to the mask layer image. A pair of mask variables are introduced based on the mask layer image, including: the Euclidean distance value X1 from the area with a value of 1 in the mask layer image to the left boundary of the mask layer image, and the Euclidean distance value X2 from the area with a value of 1 in the mask layer image to the right boundary of the mask layer image.

[0020] Preferably, pre-training the traveling wave recognition model using a pre-training data set includes:

[0021] Set the input time series data length of the traveling wave recognition model;

[0022] Label each input data, including: when the input data has a traveling wave, the label value is 1, and when the input data does not have a traveling wave, the label value is 0;

[0023] The cross entropy loss function is used as the loss function of the traveling wave recognition model. The traveling wave recognition model is pre-trained using the unique hot encoding corresponding to the pre-training data set. The sigmoid function is used to obtain the output distribution of the traveling wave recognition model, including: the probability Y_1 that the input data belongs to the traveling wave and the probability Y_2 that the input data does not belong to the traveling wave.

[0024] During pre-training, the initial values ​​of X1 and X2 are set in the range of [0, 3].

[0025] Preferably, using a pre-trained traveling wave recognition model to perform traveling wave recognition on input data under the sliding window, and calculating the traveling wave recognition accuracy under the sliding window, including:

[0026] The wide area synchronous measurement data between sampling timestamp tn and sampling timestamp t is used as the input data of the pre-trained traveling wave recognition model; the sliding window width is the input time series data length of the traveling wave recognition model;

[0027] The input data is identified by using a sliding window. The output data of the traveling wave identification model includes: the probability that the input data under the sliding window belongs to the traveling wave, and the probability that the input data under the sliding window does not belong to the traveling wave;

[0028] The output data of the traveling wave recognition model is used to calculate the traveling wave recognition accuracy, satisfying: CY = e (Y_1-Y_2) , where CY is the traveling wave recognition accuracy.

[0029] Preferably, the maximum accuracy of traveling wave recognition under the sliding window is taken as the objective function, and under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, the objective function is iteratively solved to obtain the maximum accuracy of traveling wave recognition under the sliding window and the corresponding optimal value of the mask variable, including:

[0030] The objective function is solved iteratively using the swarm intelligence algorithm, and {X1,X2,CY} is obtained in each iteration;

[0031] The iteration stop value is calculated as follows:

[0032]

[0033] Where W is the iteration stop value, {CY} is the index sequence consisting of the traveling wave recognition accuracy index obtained by multiple iterations, max({CY}) is the maximum value in the index sequence, min({CY}) is the minimum value in the index sequence, mean({CY}) is the average value of the index sequence, S_1 is the randomness index of a variable sequence {X1} obtained by multiple iterations, and S_2 is the randomness index of another variable sequence {X2} obtained by multiple iterations;

[0034] The conditions for stopping the iterative solution include: the iterative stop values ​​of 10 consecutive iterative solutions are all less than the set threshold, or the set number of iterations is reached.

[0035] Preferably, the threshold value is set to 0.5 and the number of iterations is set to 50 times.

[0036] The present invention also proposes a distribution network fault location system based on wide-area synchronous measurement data, comprising:

[0037] The data acquisition module is used to collect the voltage phasors and current phasors measured by all PMUs at each sampling time stamp;

[0038] The traveling wave recognition model pre-training module is used to establish a traveling wave recognition model based on the CNN-LSTM network architecture. When it is determined that a fault occurs in the distribution network, the voltage phasor in the set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, wherein the mask operation controls the width of the selected area in the pre-training data set by adjusting a pair of mask variables. The traveling wave recognition model is pre-trained using the pre-training data set.

[0039] The traveling wave recognition model recognition module is used to use the pre-trained traveling wave recognition model to perform traveling wave recognition on the input data under the sliding window, and calculate the traveling wave recognition accuracy under the sliding window; taking the maximum traveling wave recognition accuracy under the sliding window as the objective function, under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, iteratively solve the objective function to obtain the maximum value of the traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable; according to the optimal value of a pair of mask variables corresponding to the maximum value of the maximum value of the traveling wave recognition accuracy under all sliding windows, determine the selected area in the input data, and determine the fault distance measurement result according to the timestamp corresponding to the selected area in the input data and the traveling wave transmission speed.

[0040] The present invention is also a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0041] The present invention is also a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0042] The beneficial effect of the present invention is that, compared with the prior art, at least the method proposed by the present invention includes: installing a PMU device on the distribution network, collecting wide-area synchronous measurement data through the PMU device; decomposing the collected data according to the wide-area synchronous measurement data collected by the PMU device, and pre-training the traveling wave recognition network according to the decomposition result of the collected data to obtain a pre-trained traveling wave recognition model. Mask variables are set for the pre-trained traveling wave recognition model to obtain the traveling wave recognition result after adding the mask variables, and the optimization function of the mask variables is constructed according to the traveling wave recognition result, and the optimization function of the mask variables is solved, wherein in the solution process, the distribution of the mask variables is analyzed to achieve the purpose of adaptive iteration stop, thereby completing the traveling wave recognition. According to the traveling wave recognition result, the distribution network fault distance measurement based on the wide-area synchronous measurement data is completed. The present invention reduces the computing power demand brought by the improvement of the accuracy rate in traveling wave recognition, and after determining the best recognition result of the traveling wave, the error in the distribution network fault detection using the traveling wave is further reduced through the mask layer, thereby improving the accuracy of the distribution network fault distance measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the distribution network fault location method based on wide-area synchronous measurement data proposed by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0045] The present invention proposes a distribution network fault location method based on wide-area synchronous measurement data. Figure 1 As shown, including:

[0046] Step 1: Collect the voltage phasors and current phasors measured by all PMUs at each sampling timestamp.

[0047] Specifically, PMU equipment is deployed at key nodes of the distribution network, where the key nodes include but are not limited to: substations, line branch points; to ensure that voltage and current phasors and their timestamp information can be collected in real time.

[0048] In the embodiment, the PMU device samples the voltage signal and the current signal at a set sampling frequency to obtain the amplitude and phase of the voltage phasor and the current phasor, and synchronizes the data through the GPS clock to obtain the sampling timestamp; wherein the set sampling frequency is 100 times per second. The collected synchronized phasor data is transmitted to the data center or control center through the communication network for subsequent preprocessing and analysis, wherein the communication network includes but is not limited to: optical fiber network, wireless communication network.

[0049] Among them, the voltage phase measured by the i-th PMU device at the sampling time stamp t is A i，t , the collected current phase is B i，t , and then the voltage phasor sequence A and current phasor sequence B measured by all PMU devices at each sampling timestamp can be obtained as wide-area synchronous measurement data.

[0050] Step 2, establishing a traveling wave recognition model based on the CNN-LSTM network architecture; when it is determined that a fault occurs in the distribution network, the voltage phasors in the set time period before the fault are successively subjected to wavelet packet decomposition and mask operation to obtain a pre-training data set, wherein a pair of mask variables are adjusted in the mask operation to control the width of the selected area in the pre-training data set; the traveling wave recognition model is pre-trained using the pre-training data set.

[0051] Specifically, step 2 includes:

[0052] Step 2.1, when the amplitude of the voltage phasor at sampling timestamp t is not within the set voltage range, or the amplitude of the current phasor at any sampling timestamp t is not within the set current range, it is determined that a fault occurs in the distribution network at sampling timestamp t, and the wide-area synchronous measurement data between sampling timestamp tn and sampling timestamp t are obtained to construct a voltage phasor segment sequence CA and a current phasor segment sequence CB for traveling wave detection; wherein n is the ratio of the abnormal line length of the distribution network to the traveling wave transmission speed;

[0053] In the embodiment, the voltage range is set to [60, 260] V, and the current range is set to [10, 200] A;

[0054] Since traveling waves are reflected in both voltage and current data, the present invention proposes to first analyze a single voltage or current data, and then comprehensively obtain the final traveling wave detection result. When performing traveling wave detection, it is to determine that an abnormality has occurred through voltage or current values, and then perform traveling wave detection. The reason why traveling wave detection is not performed in real time is that the amount of collected data is too large and the system cannot process it in real time. Then, when a fault occurs, the distance to the fault distance measurement point is inferred based on the time when the fault occurred and the transmission speed of the traveling wave. Due to various factors such as impedance changes, problems such as traveling wave reflection may occur, which are not considered in the present invention. Only the head wave is identified for fault distance measurement through traveling wave detection.

[0055] Step 2.2, using the wavelet packet algorithm to perform wavelet packet decomposition on the voltage phasor segment sequence CA to obtain each decomposition component, and all the decomposition components form a voltage two-dimensional matrix DA; when performing a mask operation on the voltage two-dimensional matrix DA, a pair of mask variables are adjusted to control the width of the selected area in the voltage two-dimensional matrix DA, thereby forming a pre-training data set; wherein, the binary matrix used for the mask operation corresponds to the mask layer image, and a pair of mask variables are introduced based on the mask layer image, including: the Euclidean distance value X1 from the area with a value of 1 in the mask layer image to the left boundary of the mask layer image, and the Euclidean distance value X2 from the area with a value of 1 in the mask layer image to the right boundary of the mask layer image;

[0056] When the present invention uses the voltage two-dimensional matrix DA to train the CNN-LSTM (convolutional neural network-long short-term memory) network, it is necessary to superimpose a mask layer on the voltage two-dimensional matrix DA, wherein the mask layer is superimposed because the CNN-LSTM network can only use the time series length data corresponding to the fixed window as input, resulting in that when the CNN-LSTM network is used for traveling wave detection, if the input time series length data of the CNN-LSTM network is too long, there will be more interference factors during the traveling wave detection, resulting in missed detection of the traveling wave detection; if the input time series length data of the CNN-LSTM network is too short, the traveling wave detection is separated during the traveling wave detection, resulting in poor detection effect of the CNN-LSTM network; and because there are different types of lines in the distribution network, and different faults and environmental factors will cause different response strengths and durations of the traveling waves, and thus the fixed input time series data length of the CNN-LSTM network cannot be used to solve the problem.

[0057] Specifically, a binary matrix of the same size as the voltage two-dimensional matrix DA is used to perform a mask operation on the voltage two-dimensional matrix DA; in the binary matrix, the elements at positions corresponding to the regions to be selected in the voltage two-dimensional matrix DA are 1, and the elements at other positions are 0;

[0058] When the voltage two-dimensional matrix DA is used for CNN-LSTM network training, the present invention superimposes a mask layer. In the mask layer, only the data in the middle of the voltage two-dimensional matrix DA is retained after the mask operation is performed on the voltage two-dimensional matrix DA, and the data from the middle of the voltage two-dimensional matrix DA to the boundary of the mask layer are all masked contents. When retaining data, the value of the mask layer is 1, and when masking data, the value of the mask layer is 0, so that the mask layer is a binary matrix with a middle value of 1 and values ​​of 0 on both sides.

[0059] The binary matrix used for mask operation corresponds to the mask layer image; since the input time series data length of the CNN-LSTM network is fixed, a pair of mask variables are introduced based on the mask layer image to control the width of the selected area obtained after the voltage two-dimensional matrix DA is subjected to mask operation, wherein one mask variable is the Euclidean distance value X1 from the area with a value of 1 in the mask layer image to the left boundary of the mask layer image, and the other mask variable is the Euclidean distance value X2 from the area with a value of 1 in the mask layer image to the right boundary of the mask layer image;

[0060] When pre-training the traveling wave recognition model, the initial value setting range of X1 and X2 is [0, 3]; because this step is only to obtain a pre-trained traveling wave recognition model, where 0 is a hyperparameter and can be adjusted by the implementer, but it is not recommended to set it to a too large value. The initial value of X1 and X2 is 1 or 2 or 3, but it is not recommended to be 7 or 8 or 9, etc.

[0061] Step 2.3, pre-training the traveling wave recognition model using the pre-training data set;

[0062] Since the voltage two-dimensional matrix DA is masked, the input time series data length of the CNN-LSTM network can be as large as possible when the CNN-LSTM network is trained. The input time series data length of the traveling wave recognition model in the embodiment is set to 30, which can be adjusted by the implementer according to the specific implementation scenario;

[0063] Since the traveling wave recognition model is to realize traveling wave recognition instead of prediction and regression tasks, it is necessary to label each input data, including: when the input data contains traveling waves, the label value is 1, and when the input data does not contain traveling waves, the label value is 0; the cross entropy loss function is used as the loss function of the traveling wave recognition model, the traveling wave recognition model is pre-trained using the unique hot encoding corresponding to the pre-training data set, and the sigmoid function is used to obtain the output distribution of the traveling wave recognition model, including: the probability that the input data belongs to the traveling wave and the probability that the input data does not belong to the traveling wave; among them, the sigmoid function can be replaced by other normalization functions by the implementer according to the specific implementation scenario.

[0064] Step 3, use the pre-trained traveling wave recognition model to perform traveling wave recognition on the input data under the sliding window, and calculate the traveling wave recognition accuracy under the sliding window; take the maximum traveling wave recognition accuracy under the sliding window as the objective function, and iteratively solve the objective function under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model to obtain the maximum traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable.

[0065] Specifically, step 3 includes:

[0066] Step 3.1, the wide area synchronous measurement data between sampling timestamp tn and sampling timestamp t is used as the input data of the pre-trained traveling wave recognition model; the sliding window width is the input time series data length of the traveling wave recognition model;

[0067] In the embodiment, since the traveling wave recognition is performed from the time when the fault occurs, the input data is obtained by using a sliding window in reverse order of time. Starting from the data at the sampling timestamp t when the fault occurs, the sliding window indexes the data at m sampling timestamps forward as the input data T_1 under the sliding window of the pre-trained traveling wave recognition model; wherein, the value of m is 30, which is the same as the length of the input time series data of the traveling wave recognition model;

[0068] Step 3.2, using a sliding window to perform traveling wave recognition on the input data, the output data of the traveling wave recognition model includes: the probability that the input data under the sliding window belongs to the traveling wave, and the probability that the input data under the sliding window does not belong to the traveling wave;

[0069] When using the input data T_1 for traveling wave recognition, when different X1 and X2 are used for the input data T_1, the output results of the sigmoid function of the traveling wave recognition model for the two label values ​​will change, resulting in the possible change of the recognition results output by the pre-trained traveling wave recognition model; the output results of the sigmoid function of the pre-trained traveling wave recognition model for the two label values ​​are Y_1 and Y_2 respectively, where Y_1 represents the probability that the input data belongs to the traveling wave, and Y_2 represents the probability that the input data does not belong to the traveling wave;

[0070] Step 3.3, using the output data of the traveling wave recognition model to calculate the traveling wave recognition accuracy; taking the maximum traveling wave recognition accuracy under the sliding window as the objective function, under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, iteratively solve the objective function to obtain the maximum value of the traveling wave recognition accuracy under the sliding window and the corresponding optimal value of a pair of mask variables;

[0071] The traveling wave recognition accuracy CY of the traveling wave recognition model satisfies: CY = e (Y_1-Y_2),The exp function is used to prevent the situation where Y_2 is zero. The larger the output CY value is, the higher the recognition accuracy of the traveling wave is;

[0072] Therefore, when X1+X2<m is satisfied, the objective function is to maximize the traveling wave recognition accuracy index CY, and a swarm intelligence algorithm is used to iteratively solve the objective function to determine X1 and X2; the swarm intelligence algorithm includes but is not limited to: genetic algorithm, ant colony algorithm, SSA algorithm.

[0073] When the swarm intelligence algorithm is used to iteratively solve the objective function, {X1,X2,CY} will be obtained for each iterative solution; the iteration stop value is calculated using the following relationship:

[0074]

[0075] Where W is the iteration stop value, {CY} is the index sequence consisting of the traveling wave recognition accuracy index obtained by multiple iterations, max({CY}) is the maximum value in the index sequence, min({CY}) is the minimum value in the index sequence, mean({CY}) is the average value of the index sequence, S_1 is the randomness index of a variable sequence {X1} obtained by multiple iterations, and S_2 is the randomness index of another variable sequence {X2} obtained by multiple iterations;

[0076] The conditions for stopping the iterative solution include: the iterative stop values ​​of 10 consecutive iterative solutions are all less than the set threshold, and the threshold value set in the embodiment is 0.5, or the set number of iterations is reached, and the number of iterations set in the embodiment is 50 times. The number of iterations can be adjusted by the implementer according to the specific implementation scenario.

[0077] When the difference between max({CY}) and min({CY}) is smaller than mean({CY}), it means that the traveling wave recognition accuracy index CY does not change significantly, and it is meaningless to continue iterating. This means that there is no traveling wave feature in the current T_1, and X1 and X2 are difficult to converge. If the iteration is continued, the computing power will be excessively consumed;

[0078] When S_1 is larger, it means that the randomness of a variable sequence {X1} is stronger, and there is no better optimal solution that makes becomes larger, which makes it difficult for X1 to converge in the iterative solution; when S_2 is larger, it means that the randomness of another variable sequence {X2} is stronger, and there is no better optimal solution that makes becomes larger, making it difficult for X2 to converge in the iterative solution; therefore, we obtain max(S_1,S_2) to represent the lower limit of randomness, and use e (-max(S_1,S_2)) The function performs negative correlation mapping. If the lower limit of randomness is higher, it means that there is indeed no traveling wave feature in the input data T_1, then the traveling wave identification of T_1 is stopped and the iterative solution is no longer sought.

[0079] Specifically, the calculation process of S_1 and S_2 includes: since the entropy value can represent the randomness of things, the information entropy value corresponding to the position statistics of X1 and X2 in each iteration is calculated as S_1 and S_2; wherein, in order to avoid being too strict on the data when performing statistics, the embodiment calculates the information entropy value corresponding to the domain range statistics of X1 and X2 in each iteration as S_1 and S_2, where the domain range is [-3,3].

[0080] Since X1 and X2 can be used to determine that there is no traveling wave change in the input data in the current window, when the next window data is selected in reverse order, the corresponding positions of X1 and X2 in this iterative solution process are used as the starting point of the window for the next traveling wave detection, which can further reduce the amount of troubleshooting and improve the efficiency of fault location.

[0081] Step 4: Determine the selected area in the input data according to a pair of optimal values ​​of mask variables corresponding to the maximum value of the maximum traveling wave recognition accuracy under all sliding windows, and determine the fault distance measurement result according to the timestamp and traveling wave transmission speed corresponding to the selected area in the input data.

[0082] In an embodiment, the maximum CY value corresponding to the best X1 and X2 during each sliding window check is obtained, a sequence consisting of multiple CY maximum values ​​is obtained, and the optimal values ​​of X1 and X2 corresponding to the maximum values ​​in the sequence are obtained to determine the selected area in the input data, and the fault ranging result is determined based on the timestamp and traveling wave transmission speed corresponding to the selected area in the input data, wherein the determined selected area in the input data further narrows the distribution network fault location range and improves the fault ranging accuracy.

[0083] The present invention also proposes a distribution network fault location system based on wide-area synchronous measurement data, comprising:

[0084] The data acquisition module is used to collect the voltage phasors and current phasors measured by all PMUs at each sampling time stamp;

[0085] The traveling wave recognition model pre-training module is used to establish a traveling wave recognition model based on the CNN-LSTM network architecture. When it is determined that a fault occurs in the distribution network, the voltage phasor in the set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, wherein the mask operation controls the width of the selected area in the pre-training data set by adjusting a pair of mask variables. The traveling wave recognition model is pre-trained using the pre-training data set.

[0086] The traveling wave recognition model recognition module is used to use the pre-trained traveling wave recognition model to perform traveling wave recognition on the input data under the sliding window, and calculate the traveling wave recognition accuracy under the sliding window; taking the maximum traveling wave recognition accuracy under the sliding window as the objective function, under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, iteratively solve the objective function to obtain the maximum value of the traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable; according to the optimal value of a pair of mask variables corresponding to the maximum value of the maximum value of the traveling wave recognition accuracy under all sliding windows, determine the selected area in the input data, and determine the fault distance measurement result according to the timestamp corresponding to the selected area in the input data and the traveling wave transmission speed.

[0087] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0088] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0089] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0090] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A distribution network fault location method based on wide area synchronous measurement data, wherein PMU is deployed in the distribution network, characterized in that: include: Collect the voltage and current phasors measured by all PMUs at each sampling timestamp; Establish a traveling wave recognition model based on the CNN-LSTM network architecture; When it is determined that a fault occurs in the distribution network, the voltage phasor in a set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, wherein a pair of mask variables are adjusted in the mask operation to control the width of the selected area in the pre-training data set; the traveling wave recognition model is pre-trained using the pre-training data set; The pre-trained traveling wave recognition model is used to perform traveling wave recognition on the input data under the sliding window, and the traveling wave recognition accuracy under the sliding window is calculated; the maximum traveling wave recognition accuracy under the sliding window is taken as the objective function, and under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, the objective function is iteratively solved to obtain the maximum traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable; According to a pair of optimal values ​​of mask variables corresponding to the maximum value of the maximum traveling wave recognition accuracy under all sliding windows, the selected area in the input data is determined, and the fault location result is determined according to the timestamp and traveling wave transmission speed corresponding to the selected area in the input data.

2. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 1, characterized in that: When the amplitude of the voltage phasor at sampling time stamp t is not within the set voltage range, or the amplitude of the current phasor at any sampling time stamp t is not within the set current range, it is determined that a fault occurs in the distribution network at sampling time stamp t.

3. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 2, characterized in that: The set voltage range is [60, 260]V, and the set current range is [10, 200]A.

4. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 1, characterized in that: When it is determined that a fault occurs in the distribution network, the voltage phasor in the set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, including: The wide-area synchronous measurement data between sampling timestamp tn and sampling timestamp t are obtained to construct a voltage phasor segment sequence CA and a current phasor segment sequence CB for traveling wave detection; wherein n is the ratio of the abnormal line length of the distribution network to the traveling wave transmission speed; The wavelet packet algorithm is used to decompose the voltage phasor segment sequence CA to obtain the decomposed components, and all the decomposed components are combined into a voltage two-dimensional matrix DA. When performing a mask operation on the voltage two-dimensional matrix DA, a pair of mask variables is adjusted to control the width of the selected region in the voltage two-dimensional matrix DA, thereby forming a pre-training data set.

5. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 4, characterized in that: The mask operation controls the width of the selected area in the pre-training dataset by adjusting a pair of mask variables, including: Use a binary matrix of the same size as the voltage two-dimensional matrix DA to perform a mask operation on the voltage two-dimensional matrix DA; in the binary matrix, the elements at positions corresponding to the regions to be selected in the voltage two-dimensional matrix DA are 1, and the elements at other positions are 0; The binary matrix used for mask operation corresponds to the mask layer image. A pair of mask variables are introduced based on the mask layer image, including: the Euclidean distance value X1 from the area with a value of 1 in the mask layer image to the left boundary of the mask layer image, and the Euclidean distance value X2 from the area with a value of 1 in the mask layer image to the right boundary of the mask layer image.

6. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 5, characterized in that: Use the pre-training data set to pre-train the traveling wave recognition model, including: Set the input time series data length of the traveling wave recognition model; Label each input data, including: when the input data has a traveling wave, the label value is 1, and when the input data does not have a traveling wave, the label value is 0; The cross entropy loss function is used as the loss function of the traveling wave recognition model. The traveling wave recognition model is pre-trained using the unique hot encoding corresponding to the pre-training data set. The sigmoid function is used to obtain the output distribution of the traveling wave recognition model, including: the probability Y_1 that the input data belongs to the traveling wave and the probability Y_2 that the input data does not belong to the traveling wave. During pre-training, the initial values ​​of X1 and X2 are set in the range of [0, 3].

7. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 6, characterized in that: Use the pre-trained traveling wave recognition model to perform traveling wave recognition on the input data under the sliding window, and calculate the traveling wave recognition accuracy under the sliding window, including: The wide area synchronous measurement data between sampling timestamp tn and sampling timestamp t is used as the input data of the pre-trained traveling wave recognition model; the sliding window width is the input time series data length of the traveling wave recognition model; The input data is identified by using a sliding window. The output data of the traveling wave identification model includes: the probability that the input data under the sliding window belongs to the traveling wave, and the probability that the input data under the sliding window does not belong to the traveling wave; The output data of the traveling wave recognition model is used to calculate the traveling wave recognition accuracy, satisfying: CY = e (Y_1-Y_2) , where CY is the traveling wave recognition accuracy.

8. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 7, characterized in that: Taking the maximum accuracy of traveling wave recognition under the sliding window as the objective function, under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, the objective function is iteratively solved to obtain the maximum accuracy of traveling wave recognition under the sliding window and the corresponding optimal value of the mask variable, including: The objective function is solved iteratively using the swarm intelligence algorithm, and {X1,X2,CY} is obtained in each iteration; The iteration stop value is calculated as follows: Where W is the iteration stop value, {CY} is the index sequence consisting of the traveling wave recognition accuracy index obtained by multiple iterations, max({CY}) is the maximum value in the index sequence, min({CY}) is the minimum value in the index sequence, mean({CY}) is the average value of the index sequence, S_1 is the randomness index of a variable sequence {X1} obtained by multiple iterations, and S_2 is the randomness index of another variable sequence {X2} obtained by multiple iterations; The conditions for stopping the iterative solution include: the iterative stop values ​​of 10 consecutive iterative solutions are all less than the set threshold, or the set number of iterations is reached.

9. The method for fault location in a distribution network based on wide-area synchronous measurement data according to claim 8, characterized in that: The threshold value is set to 0.5 and the number of iterations is set to 50.

10. A distribution network fault location system based on wide area synchronous measurement data, characterized in that: include: The data acquisition module is used to collect the voltage phasors and current phasors measured by all PMUs at each sampling time stamp; The traveling wave recognition model pre-training module is used to establish a traveling wave recognition model based on the CNN-LSTM network architecture. When it is determined that a fault occurs in the distribution network, the voltage phasor in the set time period before the fault is subjected to wavelet packet decomposition and mask operation in sequence to obtain a pre-training data set, wherein the mask operation controls the width of the selected area in the pre-training data set by adjusting a pair of mask variables. The traveling wave recognition model is pre-trained using the pre-training data set. The traveling wave recognition model recognition module is used to use the pre-trained traveling wave recognition model to perform traveling wave recognition on the input data under the sliding window, and calculate the traveling wave recognition accuracy under the sliding window; taking the maximum traveling wave recognition accuracy under the sliding window as the objective function, under the constraint that the sum of a pair of mask variables is not greater than the input time series data length of the traveling wave recognition model, iteratively solve the objective function to obtain the maximum value of the traveling wave recognition accuracy under the sliding window and the corresponding optimal value of the mask variable; according to the optimal value of a pair of mask variables corresponding to the maximum value of the maximum value of the traveling wave recognition accuracy under all sliding windows, determine the selected area in the input data, and determine the fault distance measurement result according to the timestamp corresponding to the selected area in the input data and the traveling wave transmission speed.

11. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images