Method and device for realizing power transmission line fault reason identification processing based on multi-source data fusion, processor and readable storage medium thereof

Abstract

The invention relates to a method for realizing fault reason identification processing of a power transmission line based on multi-source data fusion, and the method comprises the following steps: extracting fault features, and building a model; and power transmission line fault reason identification is carried out. The invention also relates to a device for identifying and processing the fault reason of the power transmission line based on multi-source data fusion, a processor and a readable storage medium thereof. According to the method and the device for realizing power transmission line fault reason identification processing based on multi-source data fusion, the processor and the computer readable storage medium thereof, external interference factors and data waveform characteristics of different fault types of the power transmission line are described aiming at the problem of low accuracy of an existing power transmission line fault reason identification scheme; theoretical support is provided for multi-source data input; the method comprises the following steps: preprocessing fault information by using a Gramer angle field and feature codes, and constructing feature expression modes of different fault types from the angles of time sequence waveforms, two-dimensional images and discrete features; according to the invention, an adaptive boundary parameter fusion long-short term memory neural network, a convolutional neural network and an artificial neural network are designed to identify and classify power transmission line fault causes; finally, the effectiveness and advancement of the method are verified through a real data comparison test, and a reliable solution is provided for completing a high-precision transmission line fault reason identification task.

Description

Technical Field

[0001] The present invention relates to the field of power transmission lines, and in particular to the field of safe operation of power transmission lines and fault cause identification thereof, and specifically refers to a method, device, processor and computer-readable storage medium thereof for realizing transmission line fault cause identification and processing based on multi-source data fusion. Background Art

[0002] The main causes of transmission line faults include lightning strikes, wind deviation, wildfires, icing, pollution flashover, foreign objects, etc. Domestic and foreign scholars have conducted detailed fault mechanism analysis and proposed targeted prevention and control measures based on different fault causes from the perspectives of external environmental factors and internal line factors. Reference [1] extracts the characteristics of current traveling wave mutations based on the median absolute deviation method, and combines the lightning location system to accurately match lightning interference from disturbances and closing events. Reference [2] analyzes the inherent mechanism of wind deviation faults, calculates the conductor offset by collecting the insulator string deflection angle, and thus realizes the effective detection of wind deviation faults. With the development of image recognition technology, images taken by meteorological satellites or drones can also be used to complete tasks such as wildfire fault identification [3], bird-borne fault identification [4] and icing fault identification [5]. Most of the above studies focus on a specific fault and have not formed a comprehensive fault cause identification system.

[0003] With the development of machine learning and artificial intelligence technology, algorithms such as support vector machines [6], BP neural networks [7], and deep belief networks [8] have been introduced into the field of transmission line fault cause identification. They can accurately identify various fault causes based on in-depth analysis of various fault mechanisms and combined with internal and external fault features. However, the above methods rely on multiple human-involved links such as feature extraction and threshold setting, which can easily lead to error accumulation and limit the classification effect [9]. To this end, reference

[10] converts fault signals into two-dimensional images through Gramian Angular Field (GAF) and fully utilizes the advantages of residual networks in image classification and recognition to achieve multiple fault cause identification. Reference

[11] uses fault transient waveform images as input sets to establish a fault identification classifier based on AlexNet, which can achieve a high identification accuracy on the single-phase grounding fault cause identification data set. However, due to the failure to fully integrate multiple source data such as lightning location, meteorological data, and fault recording, the existing solutions are prone to classification blind spots when facing increasingly complex system structures and operating conditions, and their universality needs to be further expanded.

[0004] [1] Cao Pulin, Shen Zhongyou, Shu Hongchun, et al.: Lightning disturbance waveform identification in massive traveling wave data based on waveform mutation characteristics[J]. Automation of Electric Power Systems, 2023, 47(17): 178-186.

[0005] [2] Huang Xinbo, Tao Baozhen, Zhao Long, et al.: Design of online monitoring system for wind deflection of transmission line conductors using wireless signal transmission [J]. High Voltage Engineering, 2011, 37(10): 2350-2355.

[0006] [3] Wang Kaizheng, Zhou Shunzhen, Wang Jian, et al.: A method for identifying fire points in power transmission lines based on multi-scale spatiotemporal feature deep fusion neural network [J / OL]. High Voltage Engineering, 1-14 [2024-05-30].

[0007] [4] Zhu Penghui, Zhao Quanzhong, Liao Zhiwen, et al.: Bird damage fault analysis based on Bayesian linear regression[J]. Journal of Nanjing University of Information Science and Technology (Natural Science Edition), 2022, 14(2): 227-232.

[0008] [5] Lin Gang, Wang Bo, Peng Hui, et al.: Identification of ice thickness in transmission line images based on strongly generalized convolutional neural network[J]. Proceedings of the CSEE, 2018, 38(11): 3393-3401.

[0009] [6] Li Linan: Fault Feature Mining and Fault Cause Identification of High Voltage AC Transmission Lines[D]. Shandong University, 2017.

[0010] [7] Zhang Hong: Research on comprehensive identification of transmission line fault causes[D]. Shandong University, 2016.

[0011] [8] Du Wen: Transmission line fault cause identification based on waveform data and deep learning[D]. Shandong University, 2020.

[0012] [9] He Jinghan, Luo Guomin, Cheng Mengxiao, et al.: A review of the research on the new generation of artificial intelligence in power system fault analysis and location[J]. Proceedings of the CSEE, 2020, 40(17): 5506-5516.

[0013]

[10] Zhao Qi, Wang Jian, Lin Fengkai, et al.: Transmission line fault identification method based on Gram angle field and ResNet[J]. Power System Protection and Control, 2024, 52(10): 95-104.

[0014]

[11] Wang Jian, Wu Hao, Zhang Bo, et al.: Transmission line fault identification method based on transfer learning-AlexNet under unbalanced samples[J]. Automation of Electric Power Systems, 2022, 46(22): 182-191. Summary of the invention

[0015] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method, device, processor and computer-readable storage medium thereof for identifying the cause of transmission line faults based on multi-source data fusion, which has good effectiveness, high accuracy and a wide range of applications.

[0016] In order to achieve the above-mentioned object, the method, device, processor and computer-readable storage medium thereof for realizing transmission line fault cause identification based on multi-source data fusion of the present invention are as follows:

[0017] The method for realizing transmission line fault cause identification and processing based on multi-source data fusion has the following main features:

[0018] (1) Extract fault features and establish a model;

[0019] (2) Identify the causes of transmission line faults;

[0020] The step (1) specifically comprises the following steps:

[0021] (1.1) Extracting continuous data features;

[0022] (1.2) Discrete data feature construction and coding;

[0023] (1.3) Establish a fault identification model.

[0024] Preferably, the step (1.1) specifically comprises the following steps:

[0025] (1.1.1) Extract the time domain features based on LSTM and calculate the response output at the current moment;

[0026] (1.1.2) Extract fault features based on GAF-CNN.

[0027] Preferably, the response output at the current moment is calculated in the step (1.1.1), specifically:

[0028] The response output at the current moment is calculated according to the following formula:

[0029] h t =o t ×σ[c t ]=o t ×σ[f t c t-1 +i t ×σ(W cx x t +W ch h t-1 +b c )];

[0030] Among them, i t , o t 、f t Represent the state results of the input gate, output gate and forget gate respectively, c t represents the current state of the gated unit, σ is the activation function, W cx , W ch is the weight parameter corresponding to the input, W cx , W ch is the weight parameter corresponding to the input, h t Indicates the response output at the current moment, b c is the bias term.

[0031] Preferably, the step (1.1.2) specifically includes the following steps:

[0032] (1.1.2.1) Perform data normalization;

[0033] (1.1.2.2) Perform polar coordinate conversion;

[0034] (1.1.2.3) Generate Gram angle and field G s Difference from Graham Corner D ;

[0035] (1.1.2.4) G is calculated by weighted average. s With G D Fusion is performed and converted into a two-dimensional image.

[0036] Preferably, the step (1.2) specifically comprises the following steps:

[0037] (1.2.1) Select the fault transition resistance, DC and third harmonic content of zero-sequence current as the internal electrical characteristics of the fault and mine the internal electrical characteristics;

[0038] (1.2.2) According to the actual situation of the fault inspection report, the weather keywords are summarized and coded, and the feature quantities closely related to the occurrence of faults are extracted from the weather information as the key factors for fault diagnosis.

[0039] Preferably, the step (1.2.1) specifically includes the following steps:

[0040] (1.2.1.1) Decouple the three-phase system into 0-mode network, 1-mode network, and 2-mode network through Karenbauer transformation, and perform phase-mode transformation on the time-domain voltage and current signals by combining the fault type and fault phase information;

[0041] (1.2.1.2) Solve the voltage modulus of the fault point and the fault current modulus flowing through the transition resistor, and solve the instantaneous value of the transition resistor. The average of the instantaneous values ​​of the transition resistor within one cycle after the fault occurs is taken as the transition resistance value Rg ;

[0042] (1.2.1.3) Based on the discretized fault recording data, the discrete Fourier transform is used to decompose the zero-sequence current time series within 2 cycles after the fault occurs and convert it into harmonic components;

[0043] (1.2.1.4) The ratio of the zero-sequence current DC component to the fundamental wave and the ratio of the third harmonic component to the fundamental wave are defined as the zero-sequence current DC content I0 and the third harmonic content I3 respectively.

[0044] Preferably, the fault identification model in the step (1.3) includes an LSTM network, a CNN network and an ANN network, and the LSTM network, the CNN network and the ANN network are placed in parallel. The LSTM network extracts the time domain features in the three-phase voltage and current time series waveforms of the fault line, the CNN network extracts the local features in the time series waveform after GAF transformation, the ANN network integrates the implicit features in the discrete data, and the LSTM network, the CNN network and the ANN network output the feature information to the feature aggregation layer to achieve dimensional flattening and result output.

[0045] Preferably, the step (2) specifically comprises the following steps:

[0046] (2.1) Perform data preprocessing on fault cases;

[0047] (2.2) Use the processed fault sample set to train the deep learning model offline, and test the application through the deep learning model.

[0048] Preferably, the step (2.1) specifically includes the following steps:

[0049] (2.1.1) Obtain the time information, meteorological description and reclosing information of the fault location from the fault inspection report, and perform corresponding feature coding on them;

[0050] (2.1.2) Obtain the three-phase voltage and current waveform data when the fault occurs from the fault recording file, calculate the DC content and third harmonic content of the zero-sequence current, and calculate the fault transition resistance by combining the distance measurement information and line parameters;

[0051] (2.1.3) Perform data interpolation on time series;

[0052] (2.1.4) Normalize the interpolated time series data;

[0053] (2.1.5) For the two-dimensional image input to CNN, the three-phase voltage and current waveforms are first transformed by GAF, and the feature maps are arranged in order.

[0054] The main feature of the device for realizing transmission line fault cause identification based on multi-source data fusion is that the device comprises:

[0055] a processor configured to execute computer executable instructions;

[0056] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for identifying the cause of a transmission line fault based on multi-source data fusion are implemented.

[0057] The processor for realizing transmission line fault cause identification processing based on multi-source data fusion has the main feature that the processor is configured to execute computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the above-mentioned method for realizing transmission line fault cause identification processing based on multi-source data fusion are realized.

[0058] The main feature of the computer-readable storage medium is that a computer program is stored thereon, and the computer program can be executed by a processor to implement the various steps of the above-mentioned method for identifying the cause of a transmission line fault based on multi-source data fusion.

[0059] The method, device, processor and computer-readable storage medium for transmission line fault cause identification based on multi-source data fusion of the present invention are adopted to address the low precision problem of existing transmission line fault cause identification schemes, characterize the external interference factors and data waveform characteristics of different transmission line fault types, and provide theoretical support for multi-source data input; Gram angle field and feature coding are used to pre-process fault information, and feature expressions of different fault types are constructed from the perspectives of time series waveform, two-dimensional image and discrete features; the present invention designs an adaptive boundary parameter fusion long short-term memory neural network, convolutional neural network and artificial neural network to identify and classify transmission line fault causes; finally, the effectiveness and advancement of the proposed method are verified through real data comparison tests, providing a reliable solution for completing the task of high-precision transmission line fault cause identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the long short-term memory neural network structure of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0061] Figure 2 It is a schematic diagram of the GAF transformation process of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0062] Figure 3It is a schematic diagram of the convolutional neural network structure of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0063] Figure 4 It is a schematic diagram of the overall structure of the fault identification model of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0064] Figure 5 It is a schematic diagram of voltage and current GAF of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0065] Figure 6 It is a schematic diagram of the process of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0066] Figure 7 It is a schematic diagram of the fault identification accuracy under different learning rates of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention.

[0067] Figure 8 It is a schematic diagram of the fault identification accuracy under different batch sizes of the method for realizing transmission line fault cause identification processing based on multi-source data fusion of the present invention.

[0068] Fig. 9 It is a schematic diagram of a fault identification confusion matrix of the method for realizing transmission line fault cause identification and processing based on multi-source data fusion of the present invention. DETAILED DESCRIPTION

[0069] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0070] The method for realizing transmission line fault cause identification based on multi-source data fusion of the present invention comprises the following steps:

[0071] (1) Extract fault features and establish a model;

[0072] (2) Identify the causes of transmission line faults;

[0073] The step (1) specifically comprises the following steps:

[0074] (1.1) Extracting continuous data features;

[0075] (1.2) Discrete data feature construction and coding;

[0076] (1.3) Establish a fault identification model.

[0077] As a preferred embodiment of the present invention, the step (1.1) specifically includes the following steps:

[0078] (1.1.1) Extract the time domain features based on LSTM and calculate the response output at the current moment;

[0079] (1.1.2) Extract fault features based on GAF-CNN.

[0080] As a preferred embodiment of the present invention, the response output at the current moment is calculated in step (1.1.1), specifically:

[0081] The response output at the current moment is calculated according to the following formula:

[0082] h t =o t ×σ[c t ]=o t ×σ[f t c t-1 +i t ×σ(W cx x t +W ch h t-1 +b c )];

[0083] Among them, i t , o t 、f t Represent the state results of the input gate, output gate and forget gate respectively, c t represents the current state of the gated unit, σ is the activation function, W cx , W ch is the weight parameter corresponding to the input, W cx , W ch is the weight parameter corresponding to the input, h t Indicates the response output at the current moment, b c is the bias term.

[0084] As a preferred embodiment of the present invention, the step (1.1.2) specifically includes the following steps:

[0085] (1.1.2.1) Perform data normalization;

[0086] (1.1.2.2) Perform polar coordinate conversion;

[0087] (1.1.2.3) Generate Gram angle and field G s Difference from Graham Corner D ;

[0088] (1.1.2.4) G is calculated by weighted average. s With GD Fusion is performed and converted into a two-dimensional image.

[0089] As a preferred embodiment of the present invention, the step (1.2) specifically includes the following steps:

[0090] (1.2.1) Select the fault transition resistance, DC and third harmonic content of zero-sequence current as the internal electrical characteristics of the fault and mine the internal electrical characteristics;

[0091] (1.2.2) According to the actual situation of the fault inspection report, the weather keywords are summarized and coded, and the feature quantities closely related to the occurrence of faults are extracted from the weather information as the key factors for fault diagnosis.

[0092] As a preferred embodiment of the present invention, the step (1.2.1) specifically includes the following steps:

[0093] (1.2.1.1) Decouple the three-phase system into 0-mode network, 1-mode network, and 2-mode network through Karenbauer transformation, and perform phase-mode transformation on the time-domain voltage and current signals by combining the fault type and fault phase information;

[0094] (1.2.1.2) Solve the voltage modulus of the fault point and the fault current modulus flowing through the transition resistor, and solve the instantaneous value of the transition resistor. The average of the instantaneous values ​​of the transition resistor within one cycle after the fault occurs is taken as the transition resistance value R g ;

[0095] (1.2.1.3) Based on the discretized fault recording data, the discrete Fourier transform is used to decompose the zero-sequence current time series within 2 cycles after the fault occurs and convert it into harmonic components;

[0096] (1.2.1.4) The ratio of the zero-sequence current DC component to the fundamental wave and the ratio of the third harmonic component to the fundamental wave are defined as the zero-sequence current DC content I0 and the third harmonic content I t .

[0097] As a preferred embodiment of the present invention, the fault identification model in the step (1.3) includes an LSTM network, a CNN network and an ANN network. The LSTM network, the CNN network and the ANN network are placed in parallel. The LSTM network extracts the time domain features in the three-phase voltage and current time series waveforms of the fault line. The CNN network extracts the local features in the time series waveform after GAF transformation. The ANN network integrates the implicit features in the discrete data. The LSTM network, the CNN network and the ANN network output the feature information to the feature aggregation layer to achieve dimensional flattening and result output.

[0098] As a preferred embodiment of the present invention, the step (2) specifically comprises the following steps:

[0099] (2.1) Perform data preprocessing on fault cases;

[0100] (2.2) Use the processed fault sample set to train the deep learning model offline, and test the application through the deep learning model.

[0101] As a preferred embodiment of the present invention, the step (2.1) specifically includes the following steps:

[0102] (2.1.1) Obtain the time information, meteorological description and reclosing information of the fault location from the fault inspection report, and perform corresponding feature coding on them;

[0103] (2.1.2) Obtain the three-phase voltage and current waveform data when the fault occurs from the fault recording file, calculate the DC content and third harmonic content of the zero-sequence current, and calculate the fault transition resistance by combining the distance measurement information and line parameters;

[0104] (2.1.3) Perform data interpolation on time series;

[0105] (2.1.4) Normalize the interpolated time series data;

[0106] (2.1.5) For the two-dimensional image input to CNN, the three-phase voltage and current waveforms are first transformed by GAF, and the feature maps are arranged in order.

[0107] The device for realizing transmission line fault cause identification based on multi-source data fusion of the present invention comprises:

[0108] a processor configured to execute computer executable instructions;

[0109] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for identifying the cause of a transmission line fault based on multi-source data fusion are implemented.

[0110] The processor of the present invention is used to implement transmission line fault cause identification processing based on multi-source data fusion, wherein the processor is configured to execute computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the above-mentioned method for implementing transmission line fault cause identification processing based on multi-source data fusion are implemented.

[0111] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program can be executed by a processor to implement the various steps of the above-mentioned method for identifying the cause of a transmission line fault based on multi-source data fusion.

[0112] Transmission lines are key elements for the stable operation of power systems. When a transmission line fails, how to quickly and accurately identify the cause of the fault is of great significance to the safe and stable operation of the power system. In view of the low precision of existing transmission line fault cause identification schemes, the present invention proposes a transmission line fault cause identification method based on multi-source data fusion, which fully extracts the implicit features in the fault transient waveform, establishes a nonlinear mapping relationship between internal and external information and the fault cause, quickly and accurately identifies the fault cause and completes the fault identification task with high precision.

[0113] Firstly, the transmission line fault cases were counted and sorted, and the features of the external environment and internal electrical measurement data were constructed and information encoded respectively; then, the data were input into the long short-term memory neural network, convolutional neural network and artificial neural network respectively, and the fault identification accuracy was improved by fusing multiple models; finally, the effectiveness and advancement of the proposed method were verified through real data comparison tests, thereby improving the fault handling efficiency of operation and maintenance personnel.

[0114] In the specific implementation of the present invention, a method flow for identifying the cause of a transmission line fault based on multi-source data fusion, and a method flow for extracting continuous and discrete fault features from fault cases are disclosed.

[0115] 1. Fault feature extraction and model building

[0116] In order to fully explore the nonlinear mapping relationship between fault samples and fault causes, it is necessary to construct appropriate sample features and select corresponding classifiers. In the process of fault feature construction, on the one hand, it is necessary to consider the inherent properties of sample data, such as different processing methods for continuous time series or discrete values; on the other hand, it is necessary to pay attention to the independence of fault features to avoid feature redundancy.

[0117] 1.1 Continuous Data Feature Extraction

[0118] 1.1.1 Time Domain Feature Extraction Based on LSTM

[0119] LSTM is an improved recurrent neural network that can learn the implicit relationship between the time segments before and after the time series signal. At the same time, its internal memory unit and gate mechanism effectively overcome the gradient vanishing and explosion problems that may be caused by traditional recurrent neural network training, and realize the learning of long-distance dependent information. Its schematic diagram is shown below. Figure 1 shown.

[0120] Figure 1 In, W cx , W ch is the weight parameter corresponding to the input, x t-1 Indicates the input value at the previous moment; h t Indicates the response output at the current moment; c t Indicates the current state of the gate control unit, c t-1 Represents the state of the gate control unit at the previous moment. The gate control unit consists of three parts: input gate, output gate, and forget gate. Its expression is as shown in formula (1):

[0121]

[0122] Among them, i t , o t 、f t Respectively represent the state results of the input gate, output gate and forget gate; W fx , W fh , W ix , W ih , W ox , W oh is the weight parameter corresponding to the input; b f , b i , b o is the bias; σ is the activation function. The output gate and the unit state jointly determine the response output h t , as shown in formula (2):

[0123] h t =o t ×σ[c t ]=o t ×σ[f t c t-1 +i t ×σ(W cx x t +W ch h t-1 +b c )]……(2)

[0124] Among them, W cx , W ch is the weight parameter corresponding to the input, b c By continuously updating weights and bias parameters, LSTM can learn the nonlinear mapping relationship between implicit information and input at each moment, thereby extracting the intrinsic features of the time series signal.

[0125] 1.1.2 Fault feature extraction based on GAF-CNN

[0126] Relying solely on the aforementioned "end-to-end" model paradigm may have blind spots in fault identification in some scenarios, thus affecting the model's ability to make accurate judgments. To this end, this section uses the Gram angle field based on the idea of ​​coordinate transformation to convert the three-phase voltage and current time series into two-dimensional images and feed them into the convolutional neural network, thereby further mining the implicit features in the fault recording time series. For a certain time series Y = {y1, y2, ..., y n},y i is the sampling value of the i-th point, and the process of generating the Gram angle field is as follows:

[0127] 1) Data normalization: Scale the time series to the interval [-1, 1] using formula (3).

[0128]

[0129] Where: y max is the maximum value in the time series Y; min is the minimum value in the sequence; is the value after data normalization. The original time series is transformed into

[0130] 2) Polar coordinate transformation: The time series The value and timestamp are mapped to angle and radius respectively:

[0131]

[0132] in: is the polar coordinate angle, whose range is [0,π]; N represents a constant that divides the interval [-1,1] into equal intervals; t i Is the timestamp.

[0133] 3) Gram angle field: The Gram angle field G is generated by using equations (5) and (6), that is, the angle sum inner product and the difference inner product. S Difference from Graham Corner D :

[0134]

[0135] 4) Image fusion: G S With G D The GAF transformation process is as follows: Figure 2 As shown, the diagonal line of the two-dimensional image can reflect the increase of the time series, and the change of the image color can reflect the change of the waveform amplitude.

[0136] In order to deeply extract the fault features in the GAF image, a convolutional neural network is used to establish a feature extraction model. Usually, a CNN consists of a convolutional layer, a pooling layer, and a fully connected layer, as shown in the schematic diagram. Figure 3 shown.

[0137] 1.2 Discrete Data Feature Construction and Coding

[0138] 1.2.1 Internal electrical characteristics mining

[0139] According to the previous analysis, the properties of the transition resistance in the transmission line fault flashover channel caused by different fault causes are different, so the fault transition resistance can be used as an effective electrical feature; in addition, when a non-metallic fault occurs, the fault current often contains rich harmonic components, and the zero-sequence current can more significantly reflect the fault characteristics than the fault phase current. Therefore, this paper selects the fault transition resistance, the DC and third harmonic content of the zero-sequence current as the internal electrical characteristics of the fault.

[0140] Considering that in the actual three-phase system, except for the three-phase short circuit, the system is in an asymmetric operation state under other fault conditions. Therefore, after decoupling the three-phase system into 0-mode, 1-mode, and 2-mode networks through Karenbauer transformation, the time domain voltage and current signals are transformed into phase mode according to the fault type and fault phase information, and the voltage modulus of the fault point and the fault current modulus flowing through the transition resistor are solved, and then the instantaneous value of the transition resistance is solved, and the average of the instantaneous value of the transition resistance within one cycle after the fault occurs is taken as the transition resistance value R. g , which is expressed as formula (7):

[0141]

[0142] Where, t=1,2,…,N, N is the number of sampling points; R g (t) is the instantaneous value of the transition resistance at time t.

[0143] Based on the discretized fault recording data, discrete Fourier transform is used to decompose the zero-sequence current time series within 2 cycles after the fault occurs and convert it into harmonic components, as shown in formula (8):

[0144]

[0145] Among them, i(n) represents the zero-sequence current sampling value, the value range of n is {0,1,2,…,N-1}, and N is the total number of sampling points in one cycle; k represents the harmonic order, and the value range is {0,1,2,…,N-1}. Therefore, the DC component I(0), fundamental component I(1) and third harmonic component I(3) of the zero-sequence current can be calculated according to formula (9):

[0146]

[0147] The ratio of the zero-sequence current DC component to the fundamental wave and the ratio of the third harmonic component to the fundamental wave are defined as the DC content I0 and the third harmonic content I3 of the zero-sequence current, respectively, as shown in formula (10):

[0148]

[0149] 1.2.2 External feature encoding

[0150] The causes of transmission line faults are affected by meteorological activities and seasonal changes. The feature quantities closely related to the occurrence of faults can be extracted from weather information as key factors for fault diagnosis. However, numerical meteorological information that can be used for direct model training cannot be obtained directly and needs to be extracted from long description texts. Based on the actual situation of fault line inspection reports, this paper summarizes and encodes weather keywords as shown in Table 1. For example, when the meteorological description text is "thunderstorm with strong winds", the three keywords "thunder", "rain" and "wind" can be extracted based on the keywords, thereby obtaining the numerical meteorological information "00010101".

[0151] Table 1 Weather keyword coding

[0152] Serial number Weather Keywords coding Serial number Weather Keywords coding 1 unknown 10000000 5 Snow, ice, frozen 00001000 2 Negative 01000000 6 wind 00000100 3 clear 00100000 7 Fog and haze 00000010 4 rain 00010000 8 thunder 00000001

[0153] For time information, the date and time period corresponding to the time of fault occurrence are normalized to obtain time information. For example, if the fault time is "6:30 on March 15, 2024", the digital quantities corresponding to the fault date and time period are 0.205 and 0.271 respectively. In addition, the reclosing action can also be used as an important feature to reflect the fault of the transmission line. The reclosing information can usually be obtained from the fault patrol report. In this paper, the three states of reclosing "success", "failure / unsuccessful" and "unknown" are encoded as 1, 0, and -1 respectively.

[0154] 1.3 Establishment of fault identification model

[0155] The overall structure of the transmission line fault identification model constructed in this paper is as follows: Figure 4 As shown in the figure, the sub-model structure selected is LSTM, CNN and ANN network, which are placed in parallel. LSTM extracts the time domain features of the three-phase voltage and current time series waveforms of the fault line in an "end-to-end" paradigm; CNN extracts the local features of the time series waveform after GAF transformation, and ANN integrates the implicit features in the discrete data, including fault transition resistance, zero-sequence current DC content and third harmonic content, meteorological characteristics, time period characteristics and reclosing information, a total of 14 dimensions. After the above parallel network, it is input to the feature aggregation layer to achieve dimensional flattening and result output.

[0156] Each type of fault has different properties, and different networks have different capabilities in extracting fault features, so adaptive boundary parameters λ0, λ1, and λ2 are added in model training. The loss function is constructed as shown in formula (11).

[0157] L=λ0L LSTM +λ1L CNN +λ2L ANN ……(11)

[0158] Among them, λ0 is the adaptive boundary parameter for the time series waveform LSTM classification module, and this paper takes 1 as the benchmark; λ1 is the adaptive boundary parameter for the two-dimensional image CNN classification module; λ2 is the adaptive boundary parameter for the discrete feature ANN classification module.

[0159] 2. Transmission line fault cause identification method based on multi-source data fusion

[0160] In order to complete the task of high-precision transmission line fault cause identification, this paper proposes a fault identification scheme based on multi-source data fusion. The overall process is as follows: Figure 6 As shown in the figure, it mainly includes two stages: data preprocessing and model training and testing.

[0161] 1) Data preprocessing

[0162] For each fault case, on the one hand, the time information, meteorological description and reclosing information of the fault location are obtained from the fault patrol report, and the corresponding feature encoding is performed; on the other hand, the three-phase voltage and current waveform data when the fault occurs are obtained from the fault recording file, and the two cycles from one cycle before the fault to the two cycles after the fault are intercepted, and the DC content and third harmonic content of the zero-sequence current are calculated based on this, and the fault transition resistance is calculated by combining the distance measurement information and line parameters. For the original time series input to LSTM, considering the differences in voltage level distribution and sampling frequency of the fault recording used, it is necessary to perform data homogeneity operation on it first, that is, first interpolate the time series data to unify it to the sampling frequency of 6.4kHz, and then normalize the interpolated time series data by formula (3) and scale it to the interval [-1,1] to avoid the difference in input signal amplitude and sampling frequency affecting the model performance. For the two-dimensional image input to CNN, the three-phase voltage and current waveforms are first transformed by GAF, and each time series generates a 224×224×3 two-dimensional feature map (where 224×224 represents the pixel size and 3 represents the RGB color channel), and then the feature maps are arranged in order. Taking a certain fault case as an example, the two-dimensional image used for CNN feature extraction is as follows: Figure 5 shown.

[0163] 2) Model training and testing applications

[0164] After preprocessing the fault cases, a deep learning model with better performance is trained offline using the processed fault sample set, and used for model testing applications. In the offline training stage, the data set is divided into a training set and a validation set according to a certain ratio. The training set is used to train the network parameters of the model. At the same time, the network hyperparameters are adjusted according to the classification effect of the model on the validation set to select the optimal performance model. During the training process, cross entropy is used as the loss function to quantify the difference between the model output results and the true labels; the Adam optimizer is used to calculate the minimum loss function, optimize the weights and biases of each layer of the model, and thus obtain the network model with the best performance. In the model testing application stage, after preprocessing the input data, the offline trained model is called for testing, and the final fault identification result can be output.

[0165] In a specific embodiment of the present invention:

[0166] 1. Design of Embodiment

[0167] The data used in this paper are all from transmission line fault cases. According to the text records of line inspection reports, the fault causes are divided into 8 categories: lightning strike, foreign objects, pollution flashover, wind deviation, line break, wildfire, ice cover and others. The fault samples are divided into training set, validation set and test set according to the ratio of 4:2:1, and the fault samples are divided into training set, validation set and test set according to the ratio of 4:2:1. Figure 6 The process shown is input into the model. Considering that there is a certain sample imbalance in the data set, the F1 score shown in formula (12) is used to measure the classification accuracy of the fault identification model.

[0168]

[0169] Among them: True positive (T P ) represents the number of samples that are actually positive and predicted to be positive; false positive (F P ) represents the number of samples that are actually negative but predicted to be positive; false negative (F N ) represents the number of samples that are actually positive but predicted to be negative.

[0170] 2. Model training and hyperparameter optimization

[0171] The model training and testing experiments were completed on the Windows 10 operating system. The computer configuration was AMD Ryzen7 4800H CPU, 2.9GHz, and 16GB of memory. The programming was implemented in Python language based on the PyTorch framework. During the model training process, the number of iterations was set to 120 rounds, and the control variable method was used to optimize the learning rate α and batch size β to obtain the optimal hyperparameters.

[0172] The learning rate is an important hyperparameter in neural network training. If its value is too large, it will be difficult to converge; if it is too small, the convergence speed will be too slow. Therefore, this paper designs the search space of the learning rate α to be {0.01, 0.005, 0.001, 0.0001}. The model iteration process under the condition of β = 64 is as follows Figure 7 As shown in the figure. When α=0.01 or 0.005, it has not fully converged after 120 iterations; when α=0.001 or 0.0001, it converges after 80 and 110 iterations respectively. From the perspective of accuracy, when α=0.001, the model accuracy is 99.28%; when α=0.0001, the model accuracy is 96.04%.

[0173] During the model iteration process, the batch size also has a certain impact on the neural network training. Generally speaking, the larger the batch size, the more stable the training; when the batch size is small, it is easy to cause the model to not converge. For this reason, this paper designs the search space of the batch size β to be {64, 128, 256, 512}. The model iteration process under the condition of α = 0.001 is as follows Figure 8 As shown in the figure. When β = 64, although it converges after about 80 iterations, it oscillates to a certain extent in the later stage; when β = 128, it converges after about 100 iterations; when β = 256 or 512, it converges after about 110 iterations. It can be seen that in the process of model iteration, the batch size has little effect on the convergence speed of the model. Therefore, in order to comprehensively consider the accuracy and stability of the model, the batch size is selected as 128 in the model hyperparameter selection.

[0174] In summary, this paper selects learning rate α = 0.001 and batch size β = 128 as the optimal hyperparameters. When the model training rounds reach 120 rounds, the classification accuracy of the fault identification model on the test set reaches 96.95%, and its confusion matrix is ​​as follows: Fig. 9 As shown. Fig. 9 It can be seen that the scheme proposed in this paper has a high recognition accuracy for lightning strikes and wind deviation faults, and has strong adaptability to a small number of samples such as icing, mountain fires, and pollution flashover.

[0175] 3. Model Performance Comparison

[0176] 3.1 Comparative analysis of adaptive boundary parameters

[0177] Since the solution proposed in this paper involves the fusion of three models, the weight coefficients λ1 and λ2 between the output results of each sub-model will also have a great impact on the model effect. Therefore, the values ​​of the weight coefficients λ1 and λ2 can be adjusted to obtain the experimental results shown in Table 2.

[0178] Table 2 Classification accuracy of fault identification model under different weight parameters

[0179]

[0180] It can be seen from Table 2 that the highest classification accuracy can be obtained when the weight coefficients λ1 = 3 and λ2 = 5. At this time, the model can not only preliminarily classify the fault causes through the fault recording data, but also further correct the classification results based on two-dimensional images and other discrete feature quantities such as meteorology and reclosing to obtain higher-precision fault identification results.

[0181] 3.2 Comparative analysis of different fault information sources

[0182] In order to verify the advantages of the scheme proposed in this paper, the classification effect of other fault cause identification schemes is compared in this section. According to the different sources of fault information, different single information sources and combined information sources are set for comparison, and the results are shown in Table 3. For the single information source, only waveform data is input into the LSTM network (Scheme 1), only image data is input into the CNN network (Scheme 2), and only discrete data information is input into the ANN network (Scheme 3). For the combined information source, different combined network structures are set according to the different data sources corresponding to the model: LSTM-CNN (Scheme 4), LSTM-ANN (Scheme 5), CNN-ANN (Scheme 6) and the scheme of this paper. Due to the randomness in the model training process, the model is tested repeatedly and the average is calculated, where the test time is the average calculation time of the test set samples.

[0183] As shown in Table 3, in the single information structure, the CNN network performs best. This is because the GAF transformation of the original waveform information in the CNN input information enhances the information expression of the fault characteristics, so that the CNN network can better fit the classification; while the test result of the ANN network remains at 81.51%. Although the training / testing time used is the least among all the schemes, the classification effect has not yet reached the accuracy required by the actual application, which indirectly reflects the importance of continuous waveform data collection. Comparing the test results of the single information structure and the combined information structure, the combined information structure has been significantly improved, which shows the effectiveness of multi-source information input. Although the training and testing time are slightly longer, it can meet the requirements of reliability and speed in practical applications. The combined method proposed in this paper has the highest accuracy. The reason is that this method can not only extract time domain features from the original time series, but also capture local features from two-dimensional images by relying on CNN. At the same time, it also integrates discrete information features and can complete the fault cause identification task with higher accuracy.

[0184] The method proposed in the technical solution of the present invention aims to find the real cause of the line failure from the fault state, such as lightning strike, foreign matter, etc. The fault case obtained by this solution has clearly included the fault type, aiming to judge the real cause of the line failure.

[0185] The data input of the technical solution of the present invention is multi-source input. On the basis of the fault voltage and current measurement data when a transmission line fault occurs, the fault transition resistance, zero-sequence current DC content and third harmonic content, meteorological characteristics, time period characteristics and reclosing information are also included to form a multi-source input.

[0186] The technical solution of the present invention, on the one hand, adopts LSTM to extract the time domain features in the three-phase voltage and current time series waveforms of the fault line in an "end-to-end" paradigm; on the other hand, the time series waveform is first subjected to GAF transformation processing, and the local features of the time series waveform after GAF transformation are extracted through CNN. The fusion of the two can fully mine the continuous time series features to avoid blind spots in fault identification.

[0187] Most of the technical personnel in this field start from simulation and realize the identification of transmission line fault types by setting different fault types, but rarely classify and identify the causes of the faults.

[0188] Secondly, technicians who study the identification of the causes of transmission line faults usually use a relatively simple technical solution. For example, the document "Identification of Lightning Interference Waveforms in Massive Traveling Wave Data Based on Waveform Mutation Characteristics" extracts the mutation characteristics of current traveling waves based on the median absolute deviation method, and combines the lightning location system to accurately match lightning interference from disturbances and closing events. The document "Design of an Online Monitoring System for Wind Deviation of Transmission Line Conductors Using Wireless Signal Transmission" analyzes the inherent mechanism of wind deviation faults, and calculates the conductor offset by collecting the insulator string deviation angle, thereby achieving effective detection of wind deviation faults. With the development of image recognition technology, images taken by meteorological satellites or drones can also be used to complete tasks such as forest fire fault identification, bird-related fault identification, and ice-covered fault identification. Most of the above studies target a specific fault and have not formed a comprehensive fault cause identification system;

[0189] Thirdly, researchers who use comprehensive fault cause identification schemes use machine algorithms such as support vector machines, BP neural networks, and deep belief networks to identify fault causes by calculating simple features such as harmonic content and meteorological factors. However, due to the reliance on multiple human-involved links such as feature extraction and threshold setting, errors are easily accumulated, limiting the classification effect.

[0190] Finally, some researchers have used deep learning solutions such as CNN, LSTM and their combinations to realize transmission line fault identification. However, since peripheral features such as meteorological information and date information are not included, it is easy to have classification blind spots when facing increasingly complex system structures and operating conditions. Its versatility needs to be further expanded.

[0191] In summary, the technical solution of the present invention studies the real engineering problem of transmission line fault cause identification, and also avoids classification blind spots and improves the accuracy of fault cause identification through multi-source data fusion, which is of great significance.

[0192] The specific implementation scheme of this embodiment can refer to the relevant description in the above embodiment, which will not be repeated here.

[0193] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0194] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0195] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0196] The method, device, processor and computer-readable storage medium for transmission line fault cause identification based on multi-source data fusion of the present invention are adopted to address the low precision problem of existing transmission line fault cause identification schemes, characterize the external interference factors and data waveform characteristics of different transmission line fault types, and provide theoretical support for multi-source data input; Gram angle field and feature coding are used to pre-process fault information, and feature expressions of different fault types are constructed from the perspectives of time series waveform, two-dimensional image and discrete features; the present invention designs an adaptive boundary parameter fusion long short-term memory neural network, convolutional neural network and artificial neural network to identify and classify transmission line fault causes; finally, the effectiveness and advancement of the proposed method are verified through real data comparison tests, providing a reliable solution for completing the task of high-precision transmission line fault cause identification.

[0197] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it is apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Therefore, the specification and drawings should be regarded as illustrative rather than restrictive.

Claims

1. A method for identifying the cause of a transmission line fault based on multi-source data fusion, characterized in that: The method comprises the following steps: (1) Extract fault features and establish a model; (2) Identify the causes of transmission line faults; The step (1) specifically comprises the following steps: (1.1) Extracting continuous data features; (1.2) Discrete data feature construction and coding; (1.3) Establish a fault identification model.

2. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 1, characterized in that: The step (1.1) specifically comprises the following steps: (1.1.1) Extract the time domain features based on LSTM and calculate the response output at the current moment; (1.1.2) Extract fault features based on GAF-CNN.

3. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 2 is characterized in that: The response output at the current moment is calculated in the step (1.1.1), specifically: The response output at the current moment is calculated according to the following formula: h t =o t ×σ[c t ]=o t ×σ[f t c t-1 +i t ×σ(W cx x t +W ch h t-1 +b c )]; Among them, i t , o t 、f t Represent the state results of the input gate, output gate and forget gate respectively, c t represents the current state of the gated unit, σ is the activation function, W cx , W ch is the weight parameter corresponding to the input, W cx , W ch is the weight parameter corresponding to the input, h t Indicates the response output at the current moment, b c is the bias term.

4. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 3 is characterized in that: The step (1.1.2) specifically includes the following steps: (1.1.2.1) Perform data normalization; (1.1.2.2) Perform polar coordinate conversion; (1.1.2.3) Generate Gram angle and field G s Difference from Graham Corner D ; (1.1.2.4) G is calculated by weighted average. s With G D Fusion is performed and converted into a two-dimensional image.

5. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 1, characterized in that: The step (1.2) specifically comprises the following steps: (1.2.1) Select the fault transition resistance, DC and third harmonic content of zero-sequence current as the internal electrical characteristics of the fault and mine the internal electrical characteristics; (1.2.2) According to the actual situation of the fault inspection report, the weather keywords are summarized and coded, and the feature quantities closely related to the occurrence of faults are extracted from the weather information as the key factors for fault diagnosis.

6. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 5, characterized in that: The step (1.2.1) specifically includes the following steps: (1.2.1.1) Decouple the three-phase system into 0-mode network, 1-mode network, and 2-mode network through Karenbauer transformation, and perform phase-mode transformation on the time-domain voltage and current signals by combining the fault type and fault phase information; (1.2.1.2) Solve the voltage modulus of the fault point and the fault current modulus flowing through the transition resistor, and solve the instantaneous value of the transition resistor. The average of the instantaneous values ​​of the transition resistor within one cycle after the fault occurs is taken as the transition resistance value R g ; (1.2.1.3) Based on the discretized fault recording data, the discrete Fourier transform is used to decompose the zero-sequence current time series within 2 cycles after the fault occurs and convert it into harmonic components; (1.2.1.4) The ratio of the zero-sequence current DC component to the fundamental wave and the ratio of the third harmonic component to the fundamental wave are defined as the zero-sequence current DC content I0 and the third harmonic content I3 respectively.

7. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 1, characterized in that: The fault identification model in the step (1.3) includes an LSTM network, a CNN network and an ANN network. The LSTM network, the CNN network and the ANN network are placed in parallel. The LSTM network extracts the time domain features in the three-phase voltage and current time series waveforms of the fault line. The CNN network extracts the local features in the time series waveform after GAF transformation. The ANN network integrates the implicit features in the discrete data. The LSTM network, the CNN network and the ANN network output the feature information to the feature aggregation layer to achieve dimensional flattening and result output.

8. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 1, characterized in that: The step (2) specifically comprises the following steps: (2.1) Preprocess the data of the fault cases; (2.2) Use the processed fault sample set to train the deep learning model offline, and test the application through the deep learning model.

9. The method for realizing transmission line fault cause identification based on multi-source data fusion according to claim 1, characterized in that: The step (2.1) specifically comprises the following steps: (2.1.1) Obtain the time information, meteorological description and reclosing information of the fault location from the fault inspection report, and perform corresponding feature coding on them; (2.1.2) Obtain the three-phase voltage and current waveform data when the fault occurs from the fault recording file, calculate the DC content and third harmonic content of the zero-sequence current, and calculate the fault transition resistance by combining the distance measurement information and line parameters; (2.1.3) Perform data interpolation on time series; (2.1.4) Normalize the interpolated time series data; (2.1.5) For the two-dimensional image input to CNN, the three-phase voltage and current waveforms are first transformed by GAF, and the feature maps are arranged in order.

10. A device for realizing transmission line fault cause identification based on multi-source data fusion, characterized in that: The device comprises: a processor configured to execute computer executable instructions; A memory storing one or more computer executable instructions, wherein when the computer executable instructions are executed by the processor, each step of the method for identifying the cause of a transmission line fault based on multi-source data fusion as described in any one of claims 1 to 9 is implemented.

11. A processor for realizing transmission line fault cause identification based on multi-source data fusion, characterized in that: The processor is configured to execute computer executable instructions. When the computer executable instructions are executed by the processor, the various steps of the method for identifying the cause of a transmission line fault based on multi-source data fusion as described in any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program can be executed by a processor to implement the various steps of the method for identifying the cause of a transmission line fault based on multi-source data fusion as described in any one of claims 1 to 9.

Images