A method and system for identifying power transmission line faults based on clustering and ensemble learning

Abstract

The present invention proposes a transmission line fault identification method and system based on clustering and ensemble learning, comprising: using mean shift clustering to segment the power frequency voltage data of the processed transmission line according to the time sequence, and dividing it into three sections: A, B, and C, wherein section A is a normal waveform, section B is a fault or disturbance waveform, and section C is a tripping or disturbance waveform; using a cycle waveform of section A to perform sliding window difference with section C, and then summing the values ​​of the window to obtain a new waveform N, wherein the amplitude of waveform N represents the difference between the waveforms in sections A and C; taking the minimum value k of each cycle of waveform N, wherein k represents the moment with the minimum difference found in a certain cycle during the process of sliding window difference of A, that is, the waveform most similar to a certain section in section C at this moment; inputting the obtained minimum value of each cycle of waveform N as a set of features into a trained random forest model, and performing prediction labeling based on the trained random forest model to realize transmission line fault identification.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed power transmission line fault diagnosis, and in particular relates to a power transmission line fault identification method and system based on clustering and ensemble learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Transmission lines are crucial to the stable operation of power systems. However, due to their long distances and complex environments, various faults may occur, such as lightning strikes, tree obstacles, cranes touching wires, and bird damage, leading to line tripping events.

[0004] In order to monitor these line tripping events, the distributed device will collect specific current and voltage waveforms generated by the tripping event. However, since the distributed device collects waveforms according to a specific threshold method, when the line is disturbed, the distributed device will also collect specific current and voltage waveforms. When a fault or disturbance occurs, the amplitude of the voltage waveform will change significantly, especially the voltage waveform before and after the trip. As for the current waveform, when the transmission line trips, the current will suddenly increase and last for at least 2 cycles. After the trip ends, the current will gradually decrease and return to a stable state. Therefore, by observing the changing characteristics of the voltage and current waveforms, it can be determined whether it is a fault waveform. The current waveform recognition algorithms are mainly global threshold method, wavelet transform method, Fourier transform method, target detection method, etc.

[0005] The problems with the above approach are:

[0006] 1. The global threshold method is not compatible with lines of different voltage levels. Affected by factors such as the installation location and voltage level, the waveform collected by each distributed device is unique. Different lines need to set empirical thresholds separately, and the false alarm rate is high.

[0007] 2. It is impossible to find or determine the moment when the current waveform suddenly changes and whether the waveform at that moment is a stable mutant rather than a mutant caused by a pulse.

[0008] A mutation is a momentary, sudden change in a waveform that is manifested as a significant increase or decrease in the amplitude or frequency of the waveform over a short period of time. A mutation can last for several cycles or occur in just one cycle. It usually contrasts with the rest of the waveform, has a large difference in amplitude or frequency, and is visible as a distinct feature or transition point in the waveform.

[0009] 3. It is impossible to obtain timing information from the voltage waveform, and the waveform changes can be captured by using the information of different characteristic segments of the waveform.

[0010] 4. Due to the wide variety of changes in fault feature segment length, amplitude, and waveform frequency, the accuracy of target detection is low, which can easily lead to misjudgment.

[0011] In addition, although there are related fault identification methods using random forests in the prior art, the random forests used for fault detection are based on the voltage and current numerical points of the three phases a, b, and c as features, which classify the instantaneous state classification and cannot classify continuous waveforms; it is impossible to achieve the recognition effect of arbitrary waveforms at different voltage levels and different installation positions. Summary of the invention

[0012] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a transmission line fault identification method based on clustering and ensemble learning, which has good recognition effect for arbitrary waveforms of different voltage levels and different installation positions, and does not need to set a threshold to achieve adaptive recognition.

[0013] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0014] In a first aspect, a method for identifying power transmission line faults based on clustering and ensemble learning is disclosed, comprising:

[0015] The processed power frequency voltage data of the transmission line is segmented according to the time sequence into three sections: A, B, and C. Section A is the normal waveform, section B is the fault or disturbance waveform, and section C is the tripping or post-disturbance waveform.

[0016] Use a periodic waveform of segment A to perform sliding window subtraction with segment C, then sum the values ​​of the window and combine them to get a new waveform N. The amplitude of waveform N represents the difference between the waveforms in segment A and segment C.

[0017] Take the minimum value k of each cycle of waveform N, where k represents the minimum difference found in a certain cycle during the difference process of sliding window A. It is considered that the waveform at this moment is most similar to a certain segment in segment A and segment C;

[0018] The minimum value of each cycle of the obtained waveform N is input into the trained random forest model as a set of features, and the predicted labels are used based on the trained random forest model to realize the transmission line fault identification.

[0019] As a further technical solution, before the power frequency voltage data is segmented according to the time sequence, the power frequency voltage of the transmission line is first obtained; and the power frequency voltage data is normalized to the maximum and minimum values.

[0020] As a further technical solution, it also includes obtaining power frequency current data of the transmission line, and extracting the maximum value of each cycle of the power frequency current data as a characteristic value;

[0021] Data classification is performed on the characteristic values ​​extracted from the power frequency current data to check whether the power frequency current is segmented and whether the mutation segment is greater than or equal to the set period.

[0022] As a further technical solution, based on the time synchronization of the power frequency voltage and power frequency current signals, the power frequency current is divided into cycles and classified according to the peak height:

[0023] If the power frequency current has only one category, the first two cycles of the power frequency voltage are considered to be segment A, followed by the next three cycles as segment B, and the sixth cycle and thereafter as segment C;

[0024] If there are two types of power frequency current, arranged in time sequence, the first type of power frequency voltage is segment A, and the second type is segment C;

[0025] If the power frequency current is of 3 categories and the second category is greater than 2 cycles, the power frequency voltage is divided into three segments A, B, and C according to the timing sequence;

[0026] If the power frequency current is greater than Category 3, it is necessary to determine whether each one is greater than 2 cycles, otherwise exclude this category, and then divide the power frequency voltage into three sections: A, B, and C according to the timing.

[0027] As a further technical solution, when classifying the characteristic values ​​extracted from the power frequency current data, an improved mean shift clustering algorithm is used to automatically classify the current peak values ​​of each cycle into corresponding categories.

[0028] As a further technical solution, when using the improved mean shift clustering algorithm, the local density of each cycle peak of the current signal is estimated, the density difference between each class is evaluated, and then a reasonable classification bandwidth is given, and then the eigenvalues ​​extracted from the power frequency current data are classified.

[0029] As a further technical solution, when training the random forest model, a one-dimensional array composed of minimum values ​​is used as the feature of the waveform, and the training set and labels are screened and integrated, and sent to the random forest algorithm for training. The hyperparameter search is used to find the optimal parameters, and the accuracy above the set value is obtained on the validation set, and the model parameters are saved.

[0030] In a second aspect, a transmission line fault identification system based on clustering and ensemble learning is disclosed, comprising:

[0031] The voltage segmentation module is configured to: segment the processed power frequency voltage data of the transmission line according to the time sequence into three sections: A, B, and C, where section A is a normal waveform, section B is a fault or disturbance waveform, and section C is a tripping or post-disturbance waveform;

[0032] The sliding window difference module is configured to: use a periodic waveform of segment A to perform sliding window difference with segment C, and then sum the values ​​of the window to obtain a new waveform N, where the amplitude of waveform N represents the difference between the waveforms in segment A and segment C;

[0033] Take the minimum value k of each cycle of waveform N, where k represents the moment with the smallest difference found in a certain cycle during the difference of sliding window A, that is, the waveform at this moment is most similar to a certain segment in segment C;

[0034] The fault identification module is configured to: input the obtained minimum value of each cycle of the waveform N as a set of features into the trained random forest model, and predict the labels based on the trained random forest model to realize the transmission line fault identification.

[0035] One or more of the above technical solutions have the following beneficial effects:

[0036] The method proposed in the technical solution of the present invention has a good recognition effect for arbitrary waveforms of different voltage levels and different installation positions, and there is no need to set a threshold to achieve adaptive recognition.

[0037] The method proposed in the technical solution of the present invention utilizes the time series characteristics of the waveform, clearly depicts the waveform characteristics when there is a fault or disturbance, and uses the information of different moments of a waveform for classification. When segments A and C are similar, it indicates that a disturbance has occurred. When segments A and C differ greatly, it indicates that a fault has occurred in the line. It reduces the misjudgment of faults and greatly improves the recognition rate of faults. It has a more scientific explanation from the perspective of the power system.

[0038] The method algorithm proposed in the technical solution of the present invention is not affected by the sudden waveform of the sudden section (section B), eliminates the influence of disturbances such as shock waves and instantaneous discharges, greatly reduces the misjudgment rate of fault judgment, and improves the accuracy of fault diagnosis.

[0039] When there is no obvious change in the voltage waveform B segment, that is, the voltage has no mutant characteristics, the recognition algorithm is still applicable, and there is no need to consider the diversity of mutants. Compared with the target recognition algorithm, it has a wider application range and is more accurate.

[0040] The relevant data processed in the technical solution of the present invention is a time-continuous voltage waveform. The object is to classify a waveform, which contains timing information, and the waveform change information will also be reflected on the waveform. Through the feature processing method of this embodiment, random forest can process higher-dimensional features.

[0041] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0043] Figure 1 The present invention is a flowchart of a method for identifying power transmission line faults based on clustering and ensemble learning. DETAILED DESCRIPTION

[0044] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0045] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0046] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0047] When identifying transmission line faults, we expect a machine learning algorithm that is stable and performs well across the board. However, the actual situation is not always ideal. Sometimes, we may obtain multiple models that perform well in some aspects but have certain preferences (weakly supervised models).

[0048] In order to solve this problem, an ensemble learning method is used, that is, these weakly supervised models are combined to obtain a better and more comprehensive strong-supervised model. The basic idea of ​​ensemble learning is that even if one of the weak classifiers makes a wrong prediction, the other weak classifiers can correct the error, thereby improving the overall performance. Random forest is an algorithm that integrates multiple trees through the bagging idea of ​​ensemble learning, and its basic unit is the decision tree.

[0049] Mean shift is a density-based non-parametric clustering algorithm. The algorithm idea is to assume that data sets of different clusters conform to different probability density distributions, and find the fastest direction of density increase for any sample point (the meaning of the fastest direction is MeanShift). The area with high sample density corresponds to the maximum value of the distribution. These sample points will eventually converge to the local density maximum, and points that converge to the same local maximum are considered to be members of the same cluster.

[0050] Embodiment 1

[0051] This embodiment discloses a method for identifying power transmission line faults based on clustering and ensemble learning, including:

[0052] Step 1: Use a distributed traveling wave monitoring device to obtain the current and voltage signals of the fault line.

[0053] In this step, the current and voltage signals are obtained at the same time, and in the fault identification process, the current characteristics are first used to segment according to the time sequence, and the voltage is segmented with the help of the current segmentation result. The purpose of this is that when a fault occurs, the sudden change of current is more obvious and easy to segment. However, the current judgment of faults is complicated and diverse, and when judging faults by voltage, it is generally considered that there is no difference between segment A and segment C and it is considered to be non-fault, otherwise it is a fault, so both are used simultaneously to realize fault identification.

[0054] Step 2: Perform feature processing on the current (voltage) signal, extract the peak value of each cycle of the current to form a one-dimensional feature, and normalize the voltage using the maximum and minimum values.

[0055] Step 3: Use the improved mean shift clustering to infer the current characteristics, obtain the category corresponding to each cycle in the time series, and arrange the categories in time series to obtain the segmentation of each category in time series;

[0056] Since the traditional mean shift algorithm usually adopts a fixed bandwidth parameter, it is impossible to achieve accurate data classification. Therefore, this embodiment introduces an adaptive bandwidth selection mechanism to dynamically adjust the bandwidth parameter according to the density and distribution of the data.

[0057] Specifically, the local density of each cycle peak of the current signal is estimated, and the density difference between each category is evaluated to provide a reasonable classification bandwidth, and then the current features in step 2 are classified.

[0058] Specifically, for a given n-dimensional space R n The m sample points X in i , i = 1...m, for one of the samples X, its mean shift vector is The sample X refers to the maximum value of the current in each cycle A max , K represents the waveform A max The number of h Refers to a spherical area with a radius of h, defined as S h (X) = {y|(yx)(yx) T ≤h 2}.

[0059] Before clustering, the local density estimation is used to adjust the bandwidth according to the size of the estimation result, that is, the size of h, which represents the maximum current A in each cycle of the current waveform. max range, making the classification results more reasonable.

[0060] After classification, check whether the current is segmented and whether the mutation segment (second segment) is greater than or equal to 2 cycles.

[0061] When determining whether it is segmented, by checking the results of mean shift clustering, the result will give the classification label of the input, that is, the maximum current of each cycle. If there is only one category, it means that the entire waveform is relatively stable; if there are multiple categories, such as greater than or equal to 2 categories, the adjacent points between different categories are the dividing points of different segments, and each category must be greater than or equal to 2 cycles.

[0062] It should be noted that in high-voltage power grids, the fault clearing time is equal to the sum of the protection device and circuit breaker action time. The general fast protection action time is greater than 40ms (Power System Relay Protection Second Edition P10). Since the frequency of the current is 50Hz, the time for each cycle is 20ms, and 40ms is approximately two cycles. Therefore, the mutation segment is greater than or equal to 2 cycles as a condition to screen whether the current is truly segmented, so as to eliminate the influence of the sudden interference of a single cycle on the segmentation judgment.

[0063] Since the voltage and current signals are time synchronized, the voltage is segmented according to the classification results of the current;

[0064] When a fault or disturbance occurs, according to the rules of the sampling data, the current is prone to mutations and is more obvious, and the change in voltage is not easy to judge. Therefore, segmenting the voltage according to the results of current segmentation is more conducive to the accuracy and efficiency of segmentation. Then, the fault is judged according to the voltage segmentation results.

[0065] If the current has only one type, the first two cycles of the voltage are considered to be segment A, followed by the next three cycles as segment B, and the sixth cycle and thereafter as segment C;

[0066] If there are two types of current, arranged in time sequence, the first type is segment A and the second type is segment C.

[0067] If the current is of Category 3 and the second category is greater than 2 cycles, it is divided into three sections, A, B, and C, according to the timing.

[0068] If the current is greater than Category 3, it is necessary to determine whether each one is greater than 2 cycles, otherwise exclude this category, and then divide the voltage into three sections A, B, and C according to the timing.

[0069] Among them, section A is the normal waveform, section B is the fault or disturbance waveform, and section C is the waveform after tripping or disturbance.

[0070] Step 4: Training phase:

[0071] (1) Using the segmented voltage data, perform sliding window subtraction on segment A and segment C, that is, subtract the values ​​at the corresponding positions, window = 40, and then sum the values ​​in the window to obtain the new waveform N. n∈[1, 40], the amplitude of waveform N represents the difference between the waveforms in segment A and segment C. n represents the nth point in the window, and represents index, A n Represents the voltage of the nth point in segment A.

[0072] Then find all the minimum points of waveform N, where the index corresponding to the minimum value is the moment that is most similar to a certain period in A and C during the sliding window process.

[0073] Specifically, the minimum value k of each cycle of waveform N is taken, where k represents the moment when the difference is the smallest in a certain cycle during the difference of sliding window A in step 5, that is, the waveform at this moment is most similar to a certain section in C. Assuming that a waveform in A is completely consistent with a certain waveform in C, k=0.

[0074] (2) A one-dimensional array [k1, k2, ..., k n ], where k represents the minimum difference value found in a certain period during the difference process of sliding window A, k n Represents the minimum value of the nth period, and uses this array as the characteristic of this waveform.

[0075] (3) Filter and integrate the training set and labels, send them to the random forest algorithm for training, use hyperparameter search to find the optimal parameters, obtain an accuracy of more than 99% on the validation set, and save the model.

[0076] Inference stage: (1) First, process the real-time detected power frequency voltage waveform data according to the first step of the training stage; (2) Use the trained data to infer the data and obtain the labels of the new data, including two types of labels: fault and disturbance.

[0077] The above technical solution of this embodiment can realize the recognition of fault and non-fault characteristic waveforms of the transmission line. The technical solution of the present invention proposes a transmission line fault diagnosis algorithm, which can simplify the identification process, improve the recognition accuracy, and expand the scope of fault diagnosis.

[0078] Embodiment 2

[0079] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0080] Embodiment 3

[0081] The purpose of this embodiment is to provide a computer-readable storage medium.

[0082] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0083] Embodiment 4

[0084] The purpose of this embodiment is to provide a transmission line fault identification system based on clustering and ensemble learning, including:

[0085] The voltage segmentation module is configured to: segment the processed power frequency voltage data of the transmission line according to the time sequence into three sections: A, B, and C, where section A is a normal waveform, section B is a fault or disturbance waveform, and section C is a tripping or post-disturbance waveform;

[0086] The sliding window difference module is configured to: use a periodic waveform of segment A to perform sliding window difference with segment C, and then sum the values ​​of the window to obtain a new waveform N, where the amplitude of waveform N represents the difference between the waveforms in segment A and segment C;

[0087] Take the minimum value k of each cycle of waveform N, where k represents the moment with the smallest difference found in a certain cycle during the difference of sliding window A, that is, the waveform at this moment is most similar to a certain segment in segment C;

[0088] The fault identification module is configured to: input the minimum value of each cycle of the obtained waveform N as a set of features into the trained random forest model, and predict the label based on the trained random forest model to realize the transmission line fault identification. The random forest model includes a voltage segmentation module and a sliding window difference module.

[0089] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0090] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0091] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for identifying power transmission line faults based on clustering and ensemble learning, characterized in that: include: The processed power frequency voltage data of the transmission line is segmented according to the time sequence into three sections: A, B, and C. Section A is the normal waveform, section B is the fault or disturbance waveform, and section C is the tripping or post-disturbance waveform. Use a periodic waveform of segment A to perform sliding window subtraction with segment C, and then sum the values ​​of the window to obtain a new waveform N. The amplitude of waveform N represents the difference between the waveforms in segment A and segment C. Take the minimum value k of each cycle of waveform N, where k represents the moment with the smallest difference found in a certain cycle during the difference of sliding window A, that is, the waveform at this moment is most similar to a certain segment in segment C; The minimum value of each cycle of the obtained waveform N is input into the trained random forest model as a set of features, and the predicted labels are used based on the trained random forest model to realize the transmission line fault identification; It also includes obtaining the power frequency current data of the transmission line, and extracting the maximum value of each cycle of the power frequency current data as a characteristic value; Perform data classification on the characteristic values ​​extracted from the power frequency current data to check whether the power frequency current is segmented and whether the mutation segment is greater than or equal to the set period; When classifying the characteristic values ​​extracted from the power frequency current data, an improved mean shift clustering algorithm is used to automatically classify the current peak values ​​of each cycle into corresponding categories.

2. A method for identifying power transmission line faults based on clustering and ensemble learning as claimed in claim 1, characterized in that: Before the power frequency voltage data is segmented according to the time sequence, the power frequency voltage of the transmission line is first obtained; The power frequency voltage data is normalized to its maximum and minimum values.

3. A method for identifying power transmission line faults based on clustering and ensemble learning as claimed in claim 1, characterized in that: Based on the time synchronization of the power frequency voltage and power frequency current signals, the power frequency current is divided into cycles and classified according to the peak height: If the power frequency current has only one category, the first two cycles of the power frequency voltage are considered to be segment A, followed by the next three cycles as segment B, and the sixth cycle and thereafter as segment C; If there are two types of power frequency current, arranged in time sequence, the first type of power frequency voltage is segment A, and the second type is segment C; If the power frequency current is of 3 categories and the second category is greater than 2 cycles, the power frequency voltage is divided into three segments A, B, and C according to the timing sequence; If the power frequency current is greater than Category 3, it is necessary to determine whether each one is greater than 2 cycles, otherwise exclude this category, and then divide the power frequency voltage into three sections: A, B, and C according to the timing.

4. A method for identifying power transmission line faults based on clustering and ensemble learning as claimed in claim 1, characterized in that: When using the improved mean shift clustering algorithm, the local density of each cycle peak of the current signal is estimated, the density difference between each class is evaluated, and then a reasonable classification bandwidth is given, and then the eigenvalues ​​extracted from the power frequency current data are classified.

5. A method for identifying power transmission line faults based on clustering and ensemble learning as claimed in claim 1, characterized in that: When training the random forest model, the one-dimensional array composed of minimum values ​​is used as the feature of the waveform. The training set and labels are screened and integrated, and sent to the random forest algorithm for training. The hyperparameter search is used to find the optimal parameters, and the accuracy above the set value is obtained on the validation set, and the model parameters are saved.

6. A transmission line fault identification system based on clustering and ensemble learning, characterized in that: A method for identifying power transmission line faults based on clustering and ensemble learning according to any one of claims 1 to 5, comprising: The voltage segmentation module is configured to: segment the processed power frequency voltage data of the transmission line according to the time sequence into three sections: A, B, and C, where section A is a normal waveform, section B is a fault or disturbance waveform, and section C is a tripping or post-disturbance waveform; The sliding window difference module is configured to: use a periodic waveform of segment A to perform sliding window difference with segment C, and then sum the values ​​of the window to obtain a new waveform N, where the amplitude of waveform N represents the difference between the waveforms in segment A and segment C; Take the minimum value k of each cycle of waveform N, where k represents the moment with the smallest difference found in a certain cycle during the difference of sliding window A, that is, the waveform at this moment is most similar to a certain segment in segment C; The fault identification module is configured to: input the obtained minimum value of each cycle of the waveform N as a set of features into the trained random forest model, and predict the labels based on the trained random forest model to realize the transmission line fault identification.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method described in any one of claims 1 to 5 are implemented.

8. 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 described in any one of claims 1 to 5 are performed.

Images