CNN-LSTM-based power transmission line fault type and position identification method

Abstract

The invention provides a CNN-LSTM-based power transmission line fault type and position identification method, and relates to the technical field of power transmission line fault detection, and the method comprises the steps: calculating a frequency response curve of a health state and each fault through measuring the voltage and current related to the health state and each fault; evaluating the action and state of the system by using a frequency response analysis method; collecting a frequency response curve, using the frequency response curve as an input data set, and assigning a category number for each fault; a CNN-LSTM fault recognition model is constructed; training the CNN-LSTM fault recognition model by using the input data set to obtain a trained CNN-LSTM fault recognition model; and using the trained CNN-LSTM fault identification model to identify the fault type and position of the power transmission line. According to the method, the fault type and position identification precision can be improved, and the requirements of the power transmission line for real-time monitoring and quick response are met.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line fault detection, and in particular to a method for identifying power transmission line fault types and locations based on CNN-LSTM. Background Art

[0002] Timely and accurate detection of transmission line faults is one of the important issues to ensure the reliability of power systems. In order to evaluate the impact of impedance and fault location on fault identification and classification, the frequency response analysis method is used. The result interpretation related to frequency response analysis is considered to be the weakness of this method. In order to overcome this problem and accurately classify the types and locations of various transmission line faults, machine learning and deep learning are applied to calculate the frequency response curve of the fault. The hybrid structure of convolutional neural network and long short-term memory is used to extract the characteristics of the frequency response curve related to the transmission line fault, calculate the frequency response curve, and select them as the input data set of the proposed network. After classifying the type and location of the fault, the results of each network are analyzed by different statistical performance evaluation indicators. The existing technology has difficulties in data processing and feature extraction, usually relying on manual feature extraction, and lacks automated and intelligent feature learning processes. For complex fault modes, the accuracy and efficiency of fault identification are low; secondly, some simulation-based algorithms in the existing technology may require a long computing time in practical applications, which cannot meet the needs of real-time fault diagnosis and location. Summary of the invention

[0003] The purpose of the present invention is to provide a method for identifying the type and location of a power transmission line fault based on CNN-LSTM, aiming to improve the recognition accuracy, real-time performance and scalability of the method for identifying the type and location of a power transmission line fault, and reduce the reliance on manually extracted features. The specific technical solution is as follows:

[0004] A method for identifying the type and location of a transmission line fault based on CNN-LSTM, the method comprising the following steps:

[0005] S100, measuring the voltage and current associated with the health state and each fault, and calculating the frequency response curve of the health state and each fault;

[0006] S200, use frequency response analysis method to evaluate the action and status of the system;

[0007] S300, collecting frequency response curves and using them as input data sets, and assigning a category number to each fault, including a section number and a fault type located on the transmission line;

[0008] S400, constructing a CNN-LSTM fault recognition model, where the CNN-LSTM fault recognition model is obtained by adding an LSTM network before the fully connected layer of the CNN network;

[0009] S500, using the input data set to train the CNN-LSTM fault recognition model to obtain a trained CNN-LSTM fault recognition model;

[0010] S600. Use the trained CNN-LSTM fault identification model to identify the type and location of transmission line faults.

[0011] Furthermore, in step S100, after each error is applied at different positions on the transmission line to simulate the induction of a fault, the measured voltage and current signals associated with each fault are transmitted to the frequency domain through fast Fourier transform to obtain a frequency response curve for each state.

[0012] Furthermore, the sampling frequency of voltage and current is set to 14kHz, 20,000 sampling points are set on the transmission line, and an applied error point is set at each sampling point; the frequency response curve takes the absolute value of the ratio of the voltage after fast Fourier transform to the current after fast Fourier transform, and the calculation formula is as follows:

[0013]

[0014] Where FRC is the frequency response, FFT(V t ) is the frequency domain representation of the voltage after fast Fourier transform, FFT(I t ) is the frequency domain representation of the system current after fast Fourier transform.

[0015] Furthermore, in step S200, the frequency response analysis method is as follows: The frequency response analysis method involves three parameters: amplitude, phase, and frequency. The transfer function H(s) of the system is defined as Among them, Y(s) is obtained by Laplace transformation of the output signal, that is, the current or voltage signal y(t) output by a certain monitoring point of the system, and X(s) is obtained by transforming the input signal, that is, the current or voltage signal x(t) generated by the fault point. By analyzing different frequencies f and substituting s=jω into the transfer function, the frequency response function H(jω) of the system at this frequency can be obtained. The amplitude of H(jω)] is obtained by Calculation, Re[H(jw)] is the real part of H(jω)], Im[H(jw)] is the imaginary part of H(jω), and the phase is calculated by Calculate and obtain the amplitude and phase response of the system at each frequency, that is, the frequency response curve; evaluate the action and status of the system as follows: if in the measured frequency response, the amplitude exceeds the normal action threshold at the critical frequency, and the phase also meets the corresponding action criterion, the protection device will determine that the action conditions are met, and then trigger the protection action.

[0016] Furthermore, in step S300, the sampling frequency of the voltage and current is set to 14 kHz, 20,000 sampling points are set on the transmission line, and an applied error point is set at each sampling point. By collecting fault data at different locations, extracting the characteristics of the frequency response curve, and annotating according to the fault type and location, an input data set is obtained.

[0017] Furthermore, a category number is assigned to each fault in the input data set, including the segment number and the fault type on the transmission line, specifically including dividing the transmission line into n equal segments, corresponding to n different fault locations, and the fault types corresponding to the n segments, and the output is regarded as the target with numbers from n+1 to 2n corresponding to numbers from 1 to n.

[0018] Furthermore, the data set related to fault type identification in the input data set includes a internal fault samples, and the data set related to fault location includes a internal fault samples and b external fault samples.

[0019] Furthermore, in step S400, the CNN-LSTM fault recognition model is: in the CNN structure, the last pooling layer is connected to the input gate of the LSTM, and then after passing through the LSTM forget gate, the output gate of the LSTM is connected to the CNN fully connected layer.

[0020] Furthermore, in the CNN-LSTM fault recognition model, three convolutional layers are used, and the number of filters in the first, second, and third convolutional layers are 5, 16, and 20, respectively. The filter sizes of the first, second, and third convolutional layers are 4×4, 3×3, and 3×3, respectively, and the stride is set to 1. In each convolutional layer, a maximum pooling layer with a window dimension of 2×2 is used, and the stride associated with each maximum pooling layer is set to 2.

[0021] Furthermore, in step S500, the input data set is divided into two parts: training data and test data. 70% of the data is selected as training data to train the CNN-LSTM fault recognition model, and the remaining 30% of the data is selected as test data to test the trained CNN-LSTM fault recognition model. The training results are evaluated by using accuracy R, mean square error MSE, root mean square error RMSE, and mean absolute error MAE. The calculation formulas of each index are as follows:

[0022]

[0023] Among them, TI is the number of samples that are truly identified, FI is the number of samples that are incorrectly identified, and x i ,y i They represent the target value and predicted value respectively, and N is the number of data.

[0024] The present invention provides a method for identifying the type and location of power transmission line faults based on CNN-LSTM, which has the following beneficial effects:

[0025] The present invention provides a method for identifying the type and location of a transmission line fault based on CNN-LSTM, which measures the voltage and current related to the health state and each fault, calculates the frequency response curve of the health state and each fault; uses the frequency response analysis method to evaluate the action and state of the system; collects the frequency response curves and uses them as an input data set, and specifies a category number for each fault, including the segment number and the fault type on the transmission line; constructs a CNN-LSTM fault identification model, which is obtained by adding an LSTM network before the fully connected layer of the CNN network; uses the input data set to train the CNN-LSTM fault identification model to obtain a trained CNN-LSTM fault identification model; uses the trained CNN-LSTM fault identification model to identify the type and location of the transmission line fault; compared with the traditional method that relies on manual feature extraction, the method can automatically extract high-order features from the original signal, and capture the time dependence of the signal through time series modeling, thereby improving the accuracy of fault type and location identification; secondly, compared with the traditional method, the CNN-LSTM fault identification model can process a large amount of data in parallel, and can complete the judgment of the fault type and location in a shorter time, meeting the needs of the transmission line for real-time monitoring and rapid response. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flow chart of a method for identifying the type and location of a transmission line fault based on CNN-LSTM provided by the present invention;

[0027] Figure 2 is an overall flow chart of an embodiment of the present invention;

[0028] Figure 3 This is a network structure diagram of the CNN-LSTM fault recognition model of an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the accompanying drawings provided by the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. According to the following description, the advantages and features of the present invention will be more clear. It should be noted that the accompanying drawings are all in a very simplified form and are not in precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0030] The embodiment of the present invention provides a method for identifying the type and location of a transmission line fault based on CNN-LSTM. Figure 1 , 2 As shown, the method comprises the following steps:

[0031] S100, measuring voltage and current associated with a healthy state and each fault, and calculating a frequency response curve of the healthy state and each fault.

[0032] In this embodiment, the voltage and current parameters in a healthy state are measured to provide a reliable and comparable benchmark data for the fault identification model, helping the system to effectively distinguish between faults and normal states, and improving the accuracy, robustness and response speed of fault detection.

[0033] Faults are divided into asymmetric faults and symmetric faults. Asymmetric faults include single-phase ground short circuit faults, two-phase ground short circuit faults, and two-phase short circuit faults; symmetric faults include three-phase short circuit faults and three-phase ground short circuit faults.

[0034] In one embodiment, after each error is applied at a different position on the transmission line to simulate the induction of a fault, the measured voltage and current signals associated with each fault are transferred to the frequency domain via Fast Fourier Transform (FFT) to obtain a frequency response curve FRC for each state.

[0035] In a preferred embodiment, the sampling frequency of the voltage and current is set to 14 kHz, 20,000 sampling points are set on the transmission line, and an error application point is also set at each sampling point.

[0036] Common error types include voltage error, current error, frequency deviation and noise interference. For example, voltage error applies a voltage disturbance to simulate faults such as voltage fluctuation or voltage drop. Frequency deviation changes the sampling frequency during signal acquisition so that the collected signal deviates in frequency. Current error changes the magnitude or direction of the current to simulate faults such as short circuit, open circuit or load imbalance.

[0037] In a preferred embodiment, the frequency response curve is the absolute value of the ratio of the voltage after fast Fourier transformation to the current after fast Fourier transformation, and the calculation formula is as follows:

[0038]

[0039] Where FRC is the frequency response, FFT(V t ) is the frequency domain representation of the voltage after fast Fourier transform, FFT(I t ) is the frequency domain representation of the system current after fast Fourier transform.

[0040] S200. Use frequency response analysis method to evaluate the action and status of the system.

[0041] In this embodiment, evaluating the action and state of the system helps to improve the accuracy of fault type and location identification. In practical applications, the system may face various external interferences or changes. For example, factors such as load fluctuations, environmental changes, and equipment aging may affect the waveforms of voltage and current. By evaluating the action and state of the system, it can help detect and isolate these non-fault factors and ensure that the algorithm is not affected by noise and interference when identifying faults. The analysis in this method is completed by comparing the frequency responses of the healthy state and the damaged state of the system.

[0042] In a preferred embodiment, the frequency response analysis method is as follows: The frequency response analysis method involves three parameters: amplitude, phase, and frequency. The transfer function H(s) of the system is defined as Among them, Y(s) is obtained by Laplace transformation of the output signal, that is, the current or voltage signal y(t) output by a certain monitoring point of the system, and X(s) is obtained by transforming the input signal, that is, the current or voltage signal x(t) generated by the fault point. By analyzing different frequencies f and substituting s=jω into the transfer function, the frequency response function H(jω) of the system at this frequency can be obtained. The amplitude of H(jω) is obtained by Calculation, Re[H(jw)] is the real part of H(jω)], Im[H(jw)] is the imaginary part of H(jω), and the phase is calculated by By calculating, we can get the amplitude and phase response of the system at each frequency, that is, the frequency response curve.

[0043] For relay protection devices in power systems, their frequency response curves should be within a specific standard range during normal operation, and the amplitude and phase should meet the design requirements at the corresponding frequency. When a system fault occurs, such as a short circuit fault, the transient signal generated by the fault contains rich frequency components, which will change the frequency characteristics of the input signal of the protection device.

[0044] In a preferred embodiment, the actions and states of the evaluation system are specifically as follows: If the amplitude at the critical frequency in the measured frequency response exceeds the normal action threshold, and the phase also meets the corresponding action criteria, the protection device will determine that the action conditions are met, and then trigger the protection action. For example, if the amplitude at a certain critical frequency reaches the set action amplitude standard, and the phase offset reaches the required angle, then the protection device will determine that the action conditions are met, and then trigger a protection action such as tripping, cut off the faulty line or equipment, and protect the normal operation of other parts of the power system.

[0045] S300, collecting frequency response curves and using them as input data sets, and assigning a category number to each fault, including a section number and a fault type located on the transmission line.

[0046] In one embodiment, the sampling frequency of voltage and current is set to 14kHz, 20,000 sampling points are set on the transmission line, and an applied error point is set at each sampling point. By collecting fault data at different locations, the characteristics of the frequency response curve are extracted, and the input data set is obtained by marking according to the fault type and location.

[0047] In a preferred embodiment, when labeling the introduced fault types, the type of each fault regardless of its location and severity is defined as a target, and when labeling the introduced fault location, the location of the fault regardless of its type and severity is regarded as a target, and the location and type of the fault regardless of its severity are respectively used as outputs of the recognition network to classify the location and type of the fault.

[0048] In a preferred embodiment, a category number is assigned to each fault in the input data set, including a segment number and a fault type on the transmission line, specifically including dividing the transmission line into n (e.g., n=20,000) equal segments, corresponding to n different fault locations, and the fault types corresponding to the n segments, and the output is respectively regarded as a target with numbers from n+1 to 2n corresponding to numbers from 1 to n.

[0049] In one embodiment, the data set related to fault type identification in the input data set includes a internal fault samples, and the data set related to fault location includes a internal fault samples and b external fault samples. Internal faults refer to faults caused by the internal equipment of the power system itself, including overvoltage faults, overcurrent faults, and breakdown faults, etc. External faults include abnormal fluctuations in the external power grid, external weather (such as lightning strikes, storms), human damage, or other natural disasters. Distinguishing between internal and external faults helps to ① design more appropriate training data sets and define clearer goals, ② improve the accuracy of fault classification and location, ③ adopt more effective repair and prevention strategies, and ④ improve the robustness and generalization ability of the model.

[0050] S400, constructing a CNN-LSTM fault recognition model. The CNN-LSTM fault recognition model is obtained by adding an LSTM network before the fully connected layer of the CNN network.

[0051] In one embodiment, see Figure 3 As shown in the figure, the CNN-LSTM fault recognition model is as follows: in the CNN structure, the last pooling layer is connected to the input gate of the LSTM, and after passing through the LSTM forget gate, the output gate of the LSTM is connected to the CNN fully connected layer.

[0052] In a preferred embodiment, three convolutional layers are used in the CNN-LSTM fault recognition model. The number of filters of the first convolutional layer, the second convolutional layer and the third convolutional layer are 5, 16 and 20 respectively. The filter sizes of the first convolutional layer, the second convolutional layer and the third convolutional layer are 4×4, 3×3 and 3×3 respectively. The step size is set to 1. In each convolutional layer, a maximum pooling layer with a window dimension of 2×2 is used, and the step size associated with each maximum pooling layer is set to 2.

[0053] S500. Use the input data set to train the CNN-LSTM fault recognition model to obtain a trained CNN-LSTM fault recognition model.

[0054] In one embodiment, the input data set is divided into two parts: training data and test data. 70% of the data is selected as training data to train the CNN-LSTM fault recognition model, and the remaining 30% of the data is selected as test data to test the trained CNN-LSTM fault recognition model.

[0055] In one embodiment, the training results are evaluated using accuracy R, mean square error MSE, root mean square error RMSE, and mean absolute error MAE. The calculation formulas for each indicator are as follows:

[0056]

[0057]

[0058] Among them, TI is the number of samples that are truly identified, FI is the number of samples that are incorrectly identified, and x i ,y i They represent the target value and predicted value respectively, and N is the number of data.

[0059] When training and testing the network, if the lowest error metric is detected, save the trained network and use it to identify new faults and evaluate the results, otherwise return to continue training.

[0060] Utilizing the CNN-LSTM deep learning method and the C-LSTM model, the frequency domain features of the frequency response curve FRC are input and labeled according to the fault type and location. Through training, the CNN-LSTM model can automatically learn the frequency domain features of the frequency response curve FRC corresponding to different fault locations and types.

[0061] S600. Use the trained CNN-LSTM fault identification model to identify the type and location of transmission line faults.

[0062] The method for identifying the type and location of a transmission line fault provided by the present invention measures the voltage and current related to the health state and each fault, calculates the frequency response curve of the health state and each fault; uses the frequency response analysis method to evaluate the action and state of the system; collects the frequency response curves and uses them as an input data set, and specifies a category number for each fault, including the segment number and the fault type located on the transmission line; constructs a CNN-LSTM fault identification model, and the CNN-LSTM fault identification model is obtained by adding an LSTM network before the fully connected layer of the CNN network; uses the input data set to train the CNN-LSTM fault identification model to obtain a trained CNN-LSTM fault identification model; uses the trained CNN-LSTM fault identification model to identify the type and location of the transmission line fault; compared with the traditional method that relies on manual feature extraction, the method can automatically extract high-order features from the original signal, and capture the time dependence of the signal through time series modeling, thereby improving the accuracy of fault type and location identification; secondly, compared with the traditional method, the CNN-LSTM fault identification model can process a large amount of data in parallel, and can complete the judgment of the fault type and location in a shorter time, meeting the needs of the transmission line for real-time monitoring and rapid response.

[0063] Those skilled in the art should understand that the present invention can be implemented in many other specific forms without departing from the spirit and scope of the present invention. Based on the embodiments of the present invention, any changes and modifications made by ordinary technicians in the field of the present invention according to the above disclosure are within the scope of protection of the claims.

Claims

1. A method for identifying transmission line fault types and locations based on CNN-LSTM, characterized in that: The method comprises the following steps: S100, measuring the voltage and current associated with the health state and each fault, and calculating the frequency response curve of the health state and each fault; S200, use frequency response analysis method to evaluate the action and status of the system; S300, collecting frequency response curves and using them as input data sets, and assigning a category number to each fault, including a section number and a fault type located on the transmission line; S400, constructing a CNN-LSTM fault recognition model, where the CNN-LSTM fault recognition model is obtained by adding an LSTM network before the fully connected layer of the CNN network; S500, using the input data set to train the CNN-LSTM fault recognition model to obtain a trained CNN-LSTM fault recognition model; S600. Use the trained CNN-LSTM fault identification model to identify the type and location of transmission line faults.

2. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 1, characterized in that: In step S100, after each error is applied at different positions on the transmission line to simulate the induction of a fault, the measured voltage and current signals associated with each fault are transmitted to the frequency domain through fast Fourier transform to obtain a frequency response curve for each state.

3. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 2, characterized in that: The sampling frequency of voltage and current is set to 14kHz, 20,000 sampling points are set on the transmission line, and an error application point is set at each sampling point; the frequency response curve takes the absolute value of the ratio of the voltage after fast Fourier transform to the current after fast Fourier transform, and the calculation formula is as follows: Where FRC is the frequency response, FFT(V t ) is the frequency domain representation of the voltage after fast Fourier transform, FFT(I t ) is the frequency domain representation of the system current after fast Fourier transform.

4. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 1, characterized in that: In step S200, the frequency response analysis method is as follows: The frequency response analysis method involves three parameters: amplitude, phase, and frequency. The transfer function H(s) of the system is defined as Among them, Y(s) is obtained by Laplace transformation of the output signal, that is, the current or voltage signal y(t) output by a certain monitoring point of the system, and X(s) is obtained by transforming the input signal, that is, the current or voltage signal x(t) generated by the fault point. By analyzing different frequencies f and substituting s=jω into the transfer function, the frequency response function H(jω) of the system at this frequency can be obtained. The amplitude of H(jω) is obtained by Calculation, Re[H(jw)] is the real part of H(jω), Im[H(jw)] is the imaginary part of H(jω), and the phase is calculated by Calculate and obtain the amplitude and phase response of the system at each frequency, that is, the frequency response curve; evaluate the action and status of the system as follows: if in the measured frequency response, the amplitude exceeds the normal action threshold at the critical frequency, and the phase also meets the corresponding action criterion, the protection device will determine that the action conditions are met, and then trigger the protection action.

5. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 1, characterized in that: In step S300, the sampling frequency of voltage and current is set to 14 kHz, 20,000 sampling points are set on the transmission line, and an applied error point is set at each sampling point. By collecting fault data at different locations, the characteristics of the frequency response curve are extracted, and the input data set is obtained by marking according to the fault type and location.

6. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 5, characterized in that: A category number is assigned to each fault in the input data set, including the segment number and fault type on the transmission line. Specifically, the transmission line is divided into n equal segments, corresponding to n different fault locations, and the fault types corresponding to these n segments. The output is respectively regarded as the target with numbers from n+1 to 2n corresponding to numbers from 1 to n.

7. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 6, characterized in that: The data set related to fault type identification in the input data set contains a internal fault samples, while the data set related to fault location contains a internal fault samples and b external fault samples.

8. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 1, characterized in that: In step S400, the CNN-LSTM fault recognition model is: in the CNN structure, the last pooling layer is connected to the input gate of the LSTM, and then after passing through the LSTM forget gate, the output gate of the LSTM is connected to the CNN fully connected layer.

9. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 8, characterized in that: In the CNN-LSTM fault recognition model, three convolutional layers are used. The number of filters in the first, second, and third convolutional layers are 5, 16, and 20, respectively. The filter sizes of the first, second, and third convolutional layers are 4×4, 3×3, and 3×3, respectively. The step size is set to 1. In each convolutional layer, a maximum pooling layer with a window dimension of 2×2 is used, and the step size associated with each maximum pooling layer is set to 2.

10. The method for identifying the type and location of power line faults based on CNN-LSTM according to claim 1, characterized in that: In step S500, the input data set is divided into two parts: training data and test data. 70% of the data is selected as training data to train the CNN-LSTM fault recognition model, and the remaining 30% of the data is selected as test data to test the trained CNN-LSTM fault recognition model. The training results are evaluated by using accuracy R, mean square error MSE, root mean square error RMSE, and mean absolute error MAE. The calculation formulas of each index are as follows: Among them, TI is the number of samples that are truly identified, FI is the number of samples that are incorrectly identified, and x i ,y i They represent the target value and predicted value respectively, and N is the number of data.

Images