Fault classification system and method based on distributed device DTW method

Abstract

The invention discloses a fault classification system and method based on a distributed device DTW method. The fault classification system comprises a knowledge sample module, a data preparation module and a fault classification module. The knowledge sample module is connected with the fault classification module, the knowledge sample module can be updated according to faults provided by the fault classification module through interconnection of the knowledge sample module and the fault classification module, so that the intelligent degree of the knowledge sample module is automatically improved, and the data preparation module is connected with the fault classification module. The knowledge sample module is used for initializing a knowledge sample library and dynamically updating data of the knowledge sample library. The data preparation module is used for signal acquisition and fault detection and positioning. And the fault classification module is used for weighting coefficient calculation, sample similarity calculation, category similarity calculation and classification result determination. According to the method, DTW method similarity calculation is carried out by using fault data and traveling wave current data in a system sample library, the similarity of each fault category is selected by using distributed sensor data results at different positions, and finally, the category with the highest result similarity is determined as the fault type.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault analysis, and in particular to a fault classification system and method based on a distributed equipment DTW method. Background Art

[0002] With the development of the economy, the demand for power safety is gradually increasing. Transmission lines are an important part of the operation of the power system, and the distribution range of transmission lines is becoming wider and wider. Since transmission lines cover a wide area, a long distance, and pass through various complex terrains and environments, they are easily affected by a variety of uncontrollable factors during operation, which leads to failures. Once a transmission line failure occurs, the power system cannot operate normally, and even causes large-scale power outages, affecting life and production and causing great losses. Therefore, it is particularly important to quickly and accurately distinguish the type of fault. The relevant power departments can formulate scientific and reasonable solutions based on the type of fault to ensure the demand for electricity.

[0003] Most of the existing methods for classifying the causes of transmission line faults are deep learning methods related to neural networks. For example, there are classification models based on CNN and SVM deep learning algorithms, power grid fault classification models using improved residual networks, and models trained on time series data using LSTM. These methods can train fault classification models without worrying about how to find features. There are also methods that use waveform features, time-frequency spectrum analysis, wavelet transform analysis, meteorological multi-source data, etc. to extract features. Through methods such as peak amplitude, wave rising and falling edge features, time-frequency graph area ratio, energy spectrum similarity, modulus maximum, db3 wavelet 7-scale decomposition of high and low frequency ratio, temperature and humidity environmental conditions, etc., the inherent characteristics of the fault are used as classification criteria for classification. This type of method has a clear physical meaning.

[0004] For deep learning methods, a large amount of training data is required to train the model, but in reality, there is less data on fault types, and the features extracted by deep learning are not interpretable, and the physical meaning they represent is unclear. For analysis feature extraction methods such as waveform time-frequency, such methods are affected by complex fault conditions and system parameters, and multi-source data requires hardware. Summary of the invention

[0005] The present invention provides a fault classification system and method based on the distributed equipment DTW method to solve the shortcomings of the above-mentioned prior art. The method proposes to use fault data and traveling wave current data in a system sample library to perform DTW method similarity calculation, and use the distributed sensor data results at different locations to select the similarity of each fault category, and finally select the category with the highest similarity as the fault type.

[0006] In order to achieve the purpose of the present invention, the following technologies are proposed:

[0007] On the one hand, a fault classification system based on the distributed equipment DTW method is proposed, which includes a knowledge sample module, a data preparation module and a fault classification module. The knowledge sample module is interconnected with the fault classification module, and the data preparation module is connected with the fault classification module. The knowledge sample module is used to initialize the knowledge sample library and dynamically update the data of the knowledge sample library; the data preparation module is used for signal acquisition, fault detection and location; the fault classification module is used to calculate weighted coefficients, sample similarity, and category similarity to determine the classification results.

[0008] On the other hand, a fault classification method based on a distributed device DTW method uses a fault classification system based on a distributed device DTW method to classify faults, including the steps of:

[0009] Step 01, initialize the sample library;

[0010] Step 02, signal acquisition;

[0011] Step 03, fault detection and location;

[0012] Step 04, weighted coefficient calculation;

[0013] Step 05, sample similarity calculation;

[0014] Step 06, category similarity calculation;

[0015] Step 07, determine the classification results;

[0016] Step 08: Update sample library data.

[0017] Furthermore, when the sample library is initialized in step 01, a new fault category is added, and fault traveling wave current data is added.

[0018] Furthermore, when performing signal acquisition in step 02, fault traveling wave current data is collected through distributed sensors at different positions.

[0019] Furthermore, during fault detection and location in step 03:

[0020] Use thresholds to determine whether a fault has occurred;

[0021] Use the fault traveling wave polarity identification method to determine the fault interval;

[0022] Then use the single-ended or double-ended ranging method to determine the fault location.

[0023] Furthermore, when calculating the weighted coefficient in step 04, the position of the fault distance sensor and the parameters of different cable materials are used to calculate each received data y={y j}'s weighted coefficient empirical value a;

[0024]

[0025] Among them, y j is the jth element value of the data sequence received by the sensor, a is the empirical value of the weighting coefficient, is the weighted data series The j-th element value of .

[0026] Furthermore, when calculating the sample similarity in step 05, for the fault data received by each sensor, the DTW method is used to calculate the similarity with each sample using the data in the sample library.

[0027] Among them, the DTW (Dynamic Time Warping) algorithm uses dynamic programming to find the path with the minimum cumulative distance between two sequences. By constructing a cumulative distance matrix, the rows and columns of the matrix correspond to the elements of the two sequences respectively. Each element in the matrix represents the minimum cumulative distance from the starting point of the first sequence to the current point of the second sequence so far. Each element D[i][j] in the cumulative distance matrix can be updated by the following formula:

[0028]

[0029] In the formula, x i is the i-th element of the sequence data x, is the weighted sequence data The jth element of is the element x i and The distance between them is usually measured using the Euclidean distance, and the Euclidean distance calculation formula is: The smaller the distance, the higher the similarity. min() is the function that takes the minimum value, and D[m][n] is the element value of the mth row and nth column in the cumulative distance matrix.

[0030] Furthermore, when calculating the category similarity in step 06, the data calculation results of several sensors that are close to the fault point are selected, and the result with the smallest similarity distance is selected as the distance similarity of the fault data to the sample;

[0031] If there are multiple sample data in a fault class, the sample similarity result with the highest distance similarity is selected as the similarity result of this class.

[0032] Furthermore, when determining the classification result in step 07, the category with the highest category similarity in the calculation result obtained in step 06 is used as the category of the fault.

[0033] Furthermore, when updating the sample library data in step 08, typical fault data is determined by experts and added to the fault category sample library, and the classification result accuracy is improved by iterative updating.

[0034] The advantages of the above technical solution are:

[0035] This method can also make judgments with a small amount of sample data, and the DTW method can accurately match the deformation, translation, and stretching changes in the traveling wave current time series, which is suitable for practical application scenarios. In addition, since the high-frequency and low-frequency components of the fault have different losses at the same transmission distance, and the fault point is randomly located from the sensor position, the same fault occurs but the signals received by sensors at different locations will be different. This method uses distributed sensors to collect data to cope with the impact of changes in data at different locations, which is more robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings.

[0037] Figure 1 A fault classification system diagram based on the distributed equipment DTW method is shown.

[0038] Figure 2 A flow chart of a fault classification method based on the distributed equipment DTW method is shown.

[0039] Figure 3 A schematic diagram of the DTW method is shown. DETAILED DESCRIPTION

[0040] like Figure 1 As shown, a fault classification system based on the distributed equipment DTW method includes a knowledge sample module, a data preparation module and a fault classification module. The connection relationship is as follows: the knowledge sample module is interconnected with the fault classification module, and the interconnection between the two can realize the updating of the knowledge sample module according to the faults provided by the fault classification module, thereby automatically improving its intelligence level, and the data preparation module is connected to the fault classification module. The knowledge sample module is used to initialize the knowledge sample library and dynamically update the data of the knowledge sample library. The data preparation module is used for signal acquisition, fault detection and location. The fault classification module is used to calculate the weighted coefficient, sample similarity, and category similarity to determine the classification result.

[0041] like Figure 2 As shown, a fault classification method based on a distributed device DTW method uses a fault classification system based on a distributed device DTW method to classify faults, including the steps of:

[0042] Step 01, initialize the sample library, add a new fault category, and add fault traveling wave current data.

[0043] Step 02: signal acquisition: collecting fault traveling wave current data using distributed sensors at different locations.

[0044] Step 03, fault detection and location, use the threshold to determine whether the fault has occurred, use the fault traveling wave polarity determination method to determine the fault interval, and then use the single-end or double-end ranging method to determine the fault location.

[0045] Step 04, weighted coefficient calculation, using the location of the fault distance sensor and the dielectric constant of different cable materials and other parameters to calculate each received data y = {y j} is the empirical value a of the weighting coefficient.

[0046]

[0047] Among them, y j is the jth element value of the data sequence received by the sensor, a is the empirical value of the weighting coefficient, is the weighted data series The j-th element value of .

[0048] Step 05: Sample similarity calculation: for the fault data received by each sensor, the DTW method is used to calculate the similarity with each sample using the data in the sample library.

[0049] Among them, the DTW (Dynamic Time Warping) algorithm uses dynamic programming to find the path with the minimum cumulative distance between two sequences. By constructing a cumulative distance matrix, the rows and columns of the matrix correspond to the elements of the two sequences respectively. Each element in the matrix represents the minimum cumulative distance from the starting point of the first sequence to the current point of the second sequence so far. Each element D[i][j] in the cumulative distance matrix can be updated by the following formula:

[0050]

[0051] In the formula, x i is the i-th element of the sequence data x, is the weighted sequence data The jth element of is the element x i and The distance between them is usually measured using the Euclidean distance, and the Euclidean distance calculation formula is: The smaller the distance, the higher the similarity. min() is the function that takes the minimum value, and D[m][n] is the element value of the mth row and nth column in the cumulative distance matrix.

[0052] Step 06, class similarity calculation, select the data calculation results of several sensors that are close to the fault point, and select the result with the smallest similarity distance as the distance similarity of the fault data for the sample. If there are multiple sample data in a fault class, select the sample similarity result with the highest distance similarity as the similarity result of the class.

[0053] Step 07, determine the classification result, and use the category with the highest category similarity in the calculation results as the category of the fault.

[0054] Step 08: Update the sample library data. According to the typical fault data judged by experts, add it to the fault category sample library. Iterate and update to improve the accuracy of classification results.

[0055] The specific embodiment uses the above-proposed fault classification method based on the distributed device DTW method to perform fault classification of fault traveling wave current, taking 3 categories, 4 samples, and 3 sensors as an example.

[0056] Step 01, initialize the sample library.

[0057] New fault categories: wind deviation, bird damage, and wildfire;

[0058] And a new sample A is added, which belongs to the fault traveling wave current of wind deviation fault;

[0059] Sample B, the fault traveling wave current of bird damage fault;

[0060] Samples C and D are fault traveling wave currents of wildfire faults.

[0061] Step 02, signal acquisition.

[0062] When a fault occurs, distributed sensors at different locations are used to collect high-frequency fault traveling wave current data.

[0063] Step 03: Fault detection and location.

[0064] Use the threshold T to determine whether a fault has occurred. When the high-frequency traveling wave current data is greater than or equal to the threshold T, it is considered that a fault has occurred. When it is less than the threshold T, the fault has not occurred. For the traveling wave current that has a fault, the fault interval is located. The sensor positions at both ends of the fault traveling wave current with opposite polarities are selected as the fault interval, and then the fault position is determined by double-end ranging. If all polarities are the same, it is determined based on the amplitude of the fault data. If the amplitude of the leftmost data is greater than the amplitude of the rightmost data, the fault occurs at the leftmost end, otherwise it occurs at the rightmost end, and then the fault position is determined by single-end ranging.

[0065] Step 04, weighting coefficient calculation.

[0066] In this example, three sensors are selected, and the received data are: fault data x, fault data y, and fault data z. The weighted coefficient empirical value a of each received data x, y, and z is calculated using the position of the fault point from the sensor and the dielectric constant of different cable materials. x 、a y 、a z , get weighted fault data X = {a x *x},Y={a y *y}, Z={a x *z}.

[0067] Step 05: Sample similarity calculation.

[0068] The DTW method is used to calculate the distance similarity between the weighted fault data received by the sensor closer to the fault point and the sample data. The calculated distance similarity is D XA , D XB , D XC , D XD , D YA , D YB , D YC , D YD , D ZA , D ZB , D ZC , D ZD . D ij represents the distance similarity between the weighted fault data of the i-th sensor and the j-th sample data. Among them, the minimum cumulative distance path between the weighted fault data X and the sample data A is as follows: Figure 3 shown.

[0069] Step 06: Calculate category similarity.

[0070] The result with the smallest similarity distance is selected as the distance similarity of the fault data to the sample. If there are multiple sample data in a fault class, the result with the smallest sample similarity distance is selected as the similarity result of the class.

[0071] The distance similarity between the fault data and sample A: D A =min{D XA ,D YA ,D ZA}

[0072] The distance similarity between the fault data and sample B: D B =min{D XB ,D YB ,D ZB}

[0073] The distance similarity between the fault data and sample C: D C =min{D XC ,D YC ,D ZC}

[0074] The distance similarity between the fault data and sample D: D D =min{D XD ,D YD ,D ZD}

[0075] Then the distance similarity between the fault data and the wind deviation class is: D 风偏 =D A , the distance similarity with the bird damage category: D 鸟害 =D B , the distance similarity with the wildfire category: D 山火 =min{D C ,D D}.

[0076] Step 07, determine the classification results.

[0077] Select the category with the closest classification similarity result as the category of this fault. 风偏 =min{D 风偏 ,D 鸟害 ,D 山火}, the fault is judged to be of wind deviation type.

[0078] Step 08: Update sample library data.

[0079] It is determined whether the fault data received by the three sensors are typical waveforms of the wind deviation category, and the data of the typical waveform is added to the wind deviation category sample library.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A fault classification system based on distributed equipment DTW method, characterized in that: It includes knowledge sample module, data preparation module and fault classification module; The knowledge sample module is interconnected with the fault classification module, and the data preparation module is connected with the fault classification module; The knowledge sample module is used to initialize the knowledge sample library and dynamically update the data of the knowledge sample library; The data preparation module is used for signal acquisition, fault detection and location; The fault classification module is used to calculate weighted coefficients, sample similarity, and category similarity to determine the classification results.

2. A fault classification method based on distributed equipment DTW method, characterized in that: Using the fault classification system based on the distributed device DTW method as described in claim 1 to classify faults, comprising the steps of: Step 01, initialize the sample library; Step 02, signal acquisition; Step 03, fault detection and location; Step 04, weighted coefficient calculation; Step 05, sample similarity calculation; Step 06, category similarity calculation; Step 07, determine the classification results; Step 08: Update sample library data.

3. The fault classification method based on the distributed device DTW method according to claim 2 is characterized in that: When the sample library is initialized in step 01, a new fault category is added and fault traveling wave current data is added.

4. The fault classification method based on the distributed device DTW method according to claim 2 is characterized in that: When performing signal acquisition in step 02, fault traveling wave current data is collected through distributed sensors at different positions.

5. The fault classification method based on the distributed equipment DTW method according to claim 2 is characterized in that: During fault detection and location in step 03: Use thresholds to determine whether a fault has occurred; Use the fault traveling wave polarity identification method to determine the fault interval; Then use the single-ended or double-ended ranging method to determine the fault location.

6. The fault classification method based on the distributed equipment DTW method according to claim 2 is characterized in that: When calculating the weighted coefficient in step 04, the position of the fault distance sensor and the parameters of different cable materials are used to calculate each received data y = {y j }'s weighted coefficient empirical value a; Among them, y j is the jth element value of the data sequence received by the sensor, a is the empirical value of the weighting coefficient, is the weighted data series The j-th element value of .

7. The fault classification method based on the distributed equipment DTW method according to claim 2 is characterized in that: When calculating the sample similarity in step 05, for the fault data received by each sensor, the DTW method is used to calculate the similarity with each sample using the data in the sample library.

8. The fault classification method based on the distributed equipment DTW method according to claim 2 is characterized in that: When calculating the category similarity in step 06, the data calculation results of several sensors that are close to the fault point are selected, and the result with the smallest similarity distance is selected as the distance similarity of the fault data to the sample; According to the DTW algorithm, a cumulative distance matrix is ​​constructed. The rows and columns of the matrix correspond to the elements of the two sequences respectively. Each element in the matrix represents the minimum cumulative distance from the starting point of the first sequence to the current point of the second sequence so far. Each element D[i][j] in the cumulative distance matrix can be updated by the following formula: In the formula, x i is the i-th element of the sequence data x, is the weighted sequence data The jth element of is the element x i and The distance between them is usually measured using the Euclidean distance, and the Euclidean distance calculation formula is: The smaller the distance, the higher the similarity. min() is the function that takes the minimum value. D[m][n] is the element value of the mth row and nth column in the cumulative distance matrix. If there are multiple sample data in a fault class, the sample similarity result with the highest distance similarity is selected as the similarity result of this class.

9. The fault classification method based on the distributed equipment DTW method according to claim 2, characterized in that: When determining the classification result in step 07, the category with the highest category similarity among the calculation results obtained in step 06 is used as the category of the fault.

10. The fault classification method based on distributed equipment DTW method according to claim 2, characterized in that: When updating the sample library data in step 08, typical fault data is used and added to the fault category sample library.

Images