A method, storage medium and device for fault detection of power distribution overhead lines

Abstract

The present invention relates to the field of electric power technology and discloses a fault detection method, storage medium and device for distribution overhead lines. The method comprises the following steps: obtaining a target signal set, decomposing the target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions to determine the weight, skewness, kurtosis and approximate entropy of each data in the high-frequency feature set, and constructing a multidimensional feature vector corresponding to the target signal. Finally, each multidimensional feature vector corresponding to each signal in the target signal set is input into a preset line fault prediction model to obtain a fault classification detection result. That is, by adopting weight, skewness, kurtosis and approximate entropy to construct a multidimensional feature vector corresponding to the target signal, key fault features can be accurately extracted from complex hidden fault signals, the accuracy of model fault detection is effectively improved, and solid and reliable technical support is provided for ensuring the intelligent operation of the power system.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a fault detection method, storage medium and equipment for a distribution overhead line. Background Art

[0002] In power systems, line fault detection is a core component in ensuring the safe, stable, and efficient operation of the entire system. As the power grid continues to expand, its network structure becomes increasingly complex, especially for long-distance overhead distribution lines, which present increasingly challenging fault detection challenges. Overhead distribution lines are directly exposed to a complex and changing natural environment and are susceptible to interference from various climatic conditions and external forces. This makes them susceptible to a variety of hidden faults. The signal manifestations of these hidden faults are extremely complex, often exhibiting atypical and transient signal characteristics, such as localized arcing and intermittent contact faults.

[0003] However, most current fault detection methods find it difficult to accurately extract key fault features when processing these complex signals, resulting in a significant reduction in the accuracy of model fault detection, posing a huge hidden danger to the stable operation of the power system. Summary of the Invention

[0004] Based on this, it is necessary to address the above problems and propose a fault detection method, storage medium and equipment for distribution overhead lines, which can accurately extract key fault features from complex hidden fault signals, effectively improve the accuracy of model fault detection, and provide solid and reliable technical support for ensuring the intelligent operation of the power system.

[0005] To achieve the above object, the present invention provides, in a first aspect, a method for detecting a fault in a distribution overhead line, the method comprising:

[0006] Acquire a target signal set, wherein the target signal set includes three phase current signals, three phase voltage signals, and an electric field strength signal of a distribution overhead line;

[0007] Decomposing a target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any one signal in the target signal set;

[0008] Determine the mutual information between each intrinsic mode function and each layer's high-frequency detail coefficient, and take the intrinsic mode function corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode function;

[0009] Determining the weight, skewness, kurtosis, and approximate entropy of each data in a high-frequency feature set, wherein the high-frequency feature set includes all high-frequency intrinsic mode functions and all high-frequency detail coefficients;

[0010] Constructing a multidimensional feature vector corresponding to the target signal according to the weight, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set;

[0011] Each multidimensional feature vector corresponding to each signal in the target signal set is input into a preset line fault prediction model to obtain a fault classification detection result.

[0012] Optionally, determine the weight of each data in the high-frequency feature set, including:

[0013] Using the formula Determine the weight of each data in the high-frequency feature set;

[0014] in, is the weight of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the L2 normal form symbol, is the total number of data in the high-frequency feature set.

[0015] Optionally, the expression of the multidimensional feature vector corresponding to the target signal is:

[0016] ;

[0017] in, is the multidimensional feature vector corresponding to the target signal, is the weight of the nth data in the high-frequency feature set, is the skewness of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the kurtosis of the nth data in the high-frequency feature set, is the approximate entropy of the nth data in the high-frequency feature set, is the total number of data in the high-frequency feature set.

[0018] Optionally, before inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result, the method further includes:

[0019] Using short-time Fourier transform, transforming the target signal to obtain a time-frequency spectrum;

[0020] Arrange all high-frequency detail coefficients in sequence to obtain a coefficient vector;

[0021] Determine the energy of each eigenmode function and arrange the energies of all eigenmode functions in sequence to obtain an energy vector;

[0022] Constructing a three-dimensional tensor corresponding to the target signal according to the time-frequency spectrum, the coefficient vector, and the energy vector;

[0023] The step of inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result includes:

[0024] Each multidimensional feature vector and each three-dimensional tensor corresponding to each signal in the target signal set is input into the preset line fault prediction model to obtain the fault classification detection result.

[0025] Optionally, the method further includes:

[0026] Determine the current amplitude, voltage drop amplitude, harmonic content and fault time based on the three phase current signals and three phase voltage signals;

[0027] Determine a fault type detection result based on the current amplitude, the voltage drop amplitude, the harmonic content, and the fault time using a preset fuzzy rule;

[0028] A target fault classification detection result is determined according to the fault type detection result and the fault classification detection result.

[0029] Optionally, determining the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficients includes:

[0030] Determining a joint probability distribution between each intrinsic mode function and each layer of high-frequency detail coefficients, and determining a marginal probability distribution of each intrinsic mode function, and determining a marginal probability distribution of each layer of high-frequency detail coefficients;

[0031] The mutual information between each eigenmode function and each layer of high-frequency detail coefficients is determined according to the joint probability distribution between each eigenmode function and each layer of high-frequency detail coefficients, the marginal probability distribution of each eigenmode function, and the marginal probability distribution of each layer of high-frequency detail coefficients.

[0032] Optionally, acquiring the target signal set includes:

[0033] Acquiring three initial phase current signals, three initial phase voltage signals, and an initial electric field strength signal collected by an electromagnetic mutual inductor, wherein the electromagnetic mutual inductor has been installed on the power distribution overhead line;

[0034] Performing Gaussian kernel denoising processing on each initial phase current signal, each initial phase voltage signal and the initial electric field strength signal to obtain three phase current signals, three phase voltage signals and an electric field strength signal;

[0035] The three phase current signals, the three phase voltage signals and the electric field strength signal are all taken as elements to form the target signal set.

[0036] To achieve the above object, the present invention provides, in a second aspect, a fault detection device for a power distribution overhead line, the device comprising:

[0037] An acquisition module is used to acquire a target signal set, wherein the target signal set includes three phase current signals, three phase voltage signals and an electric field strength signal of the distribution overhead line;

[0038] a decomposition module, configured to decompose a target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any one signal in the target signal set;

[0039] A first determining module is used to determine the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficients, and to use the intrinsic mode function corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode function;

[0040] A second determination module is used to determine the weight, skewness, kurtosis and approximate entropy of each data in the high-frequency feature set, wherein the high-frequency feature set includes all high-frequency intrinsic mode functions and all high-frequency detail coefficients;

[0041] A construction module, configured to construct a multidimensional feature vector corresponding to the target signal based on the weights, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set;

[0042] The prediction module is used to input each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result.

[0043] To achieve the above-mentioned object, the present invention provides, in a third aspect, a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the method as described in any one of the first aspects.

[0044] To achieve the above-mentioned objectives, the present invention provides a computer device in a fourth aspect, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method as described in any one of the first aspects.

[0045] The embodiment of the present invention has the following beneficial effects: the above method obtains a target signal set, which includes three phase current signals, three phase voltage signals and electric field strength signals of the distribution overhead line, decomposes the target signal, and obtains multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any signal in the target signal set, and determines the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficient, and uses the intrinsic mode function corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode function, and then determines the weight, skewness, kurtosis and approximate entropy of each data in the high-frequency feature set, wherein the high-frequency feature set includes All high-frequency intrinsic mode functions and all high-frequency detail coefficients are used, and then the multidimensional feature vector corresponding to the target signal is constructed according to the weights, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set. Finally, the multidimensional feature vectors corresponding to each signal in the target signal set are input into the preset line fault prediction model to obtain the fault classification detection result; that is, by adopting weights, skewness, kurtosis and approximate entropy to construct the multidimensional feature vector corresponding to the target signal, the key fault features can be accurately extracted from complex hidden fault signals, which effectively improves the accuracy of model fault detection and provides solid and reliable technical support for ensuring the intelligent operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] in:

[0048] Figure 1 Schematic diagram of a method for detecting a fault in an overhead power distribution line according to an embodiment of the present application;

[0049] Figure 2 This is a schematic diagram of a fault detection device for a power distribution overhead line according to an embodiment of the present application;

[0050] Figure 3 1 is a diagram of the internal structure of a computer device in some embodiments. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] In power systems, line fault detection is a core component in ensuring the safe, stable, and efficient operation of the entire system. As the power grid continues to expand, its network structure becomes increasingly complex, especially for long-distance overhead distribution lines, which present increasingly challenging fault detection challenges. Overhead distribution lines are directly exposed to a complex and changing natural environment and are susceptible to interference from various climatic conditions and external forces. This makes them susceptible to a variety of hidden faults. The signal manifestations of these hidden faults are extremely complex, often exhibiting atypical and transient signal characteristics, such as localized arcing and intermittent contact faults.

[0053] However, most current fault detection methods find it difficult to accurately extract key fault features when processing these complex signals, resulting in a significant reduction in the accuracy of model fault detection, posing a huge hidden danger to the stable operation of the power system.

[0054] In response to the above problems, this application proposes a fault detection method, storage medium and equipment for distribution overhead lines, which can accurately extract key fault features from complex hidden fault signals, effectively improve the accuracy of model fault detection, and provide solid and reliable technical support for ensuring the intelligent operation of the power system. The specific implementation principle will be described in detail in the following embodiments.

[0055] In a first aspect, the present application provides a method for detecting faults in a power distribution overhead line.

[0056] To achieve the above object, the present invention provides, in a first aspect, a method for detecting a fault in a distribution overhead line, the method comprising:

[0057] Step 110: Acquire a target signal set, where the target signal set includes three phase current signals, three phase voltage signals, and an electric field strength signal of the distribution overhead line.

[0058] The overhead distribution lines here refer to the lines that require fault detection.

[0059] Regarding the signal collection method, in some embodiments, a signal collection device can be installed on the distribution overhead line to collect the three-phase current signals, three-phase voltage signals and electric field strength signals of the distribution overhead line through the signal collection device; wherein, the signal collection device can be configured with multiple high-speed collection channels, that is, each signal collection can be configured with a high-speed collection channel.

[0060] It can be understood that by configuring multiple high-speed acquisition channels, it is possible to achieve simultaneous high-speed acquisition of signals between different channels, thereby ensuring the timing consistency of the acquired signals and avoiding time offset errors between signals of different channels due to different timing of the acquired signals, thereby affecting subsequent detection results.

[0061] Furthermore, in some embodiments, the signal acquisition device also supports configurations of multiple sampling frequencies and multiple resolutions, which can be adjusted according to actual needs; for example, after a fault is detected, it can be adjusted to a high-frequency mode for signal acquisition, and after no fault is detected, it can be adjusted to a low-frequency mode or a normal mode for signal acquisition.

[0062] Step 120: Decompose the target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any signal in the target signal set.

[0063] Regarding the decomposition method of multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, in some embodiments, any one of empirical wavelet transform and empirical mode decomposition, wavelet transform and empirical mode decomposition, empirical wavelet transform and improved mode decomposition, and wavelet transform and improved mode decomposition can be used to decompose the target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions; among them, the improved mode decomposition can adopt any one of variational mode decomposition, set empirical mode decomposition, and adaptive noise complete set empirical mode decomposition.

[0064] Under the conditions of using wavelet transform and empirical mode decomposition, for the decomposition method of multiple layers of high-frequency detail coefficients, in some embodiments, wavelet transform can be used to iteratively decompose the target signal until the number of decomposition layers of the iterative decomposition is equal to the preset number of decomposition layers, and the multiple layers of high-frequency detail coefficients corresponding to all iterative decompositions are obtained; for the decomposition method of multiple intrinsic mode functions, in some embodiments, empirical mode decomposition can be used to iteratively decompose the target signal until the energy of the residual component obtained by the iterative decomposition is less than or equal to the energy threshold, and the multiple initial intrinsic mode functions corresponding to all iterative decompositions are obtained, and the first preset initial intrinsic mode functions are all used as intrinsic mode functions.

[0065] Under the conditions of using wavelet transform and empirical mode decomposition, for the decomposition method of multiple layers of high-frequency detail coefficients, in other embodiments, wavelet transform can be used to iteratively decompose the target signal until the number of decomposition layers of the iterative decomposition is equal to the preset number of decomposition layers, and the multiple layers of high-frequency detail coefficients corresponding to all iterative decompositions and the low-frequency approximation coefficients corresponding to the iterative decomposition are obtained; for the decomposition method of multiple intrinsic mode functions, in other embodiments, empirical mode decomposition can be used to iteratively decompose the low-frequency approximation coefficients until the energy of the residual component obtained by the iterative decomposition is less than or equal to the energy threshold, and multiple initial intrinsic mode functions corresponding to all iterative decompositions are obtained, and the first preset initial intrinsic mode functions are all used as intrinsic mode functions.

[0066] Among them, the preset number of decomposition layers, energy threshold and the specific number of pre-set initial intrinsic mode functions can all be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, they can also be set by the operator according to actual needs.

[0067] Furthermore, with respect to the method for determining the preset number of decomposition layers, in some embodiments, a short-time Fourier transform can be used to transform the target signal to obtain a time-frequency spectrum diagram, and then frequency extraction is performed on the time-frequency spectrum diagram to obtain the dominant frequency of the signal, and finally the preset number of decomposition layers is determined based on the dominant frequency of the signal; with respect to the method for determining the energy threshold, in some embodiments, the energy of the target signal can be determined, and a preset percentage of the energy of the target signal can be used as the energy threshold, where the preset percentage can be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics, and of course, can also be set by the operator based on actual needs.

[0068] Furthermore, we can use the formula Determine the preset number of decomposition levels; where, To preset the number of decomposition layers, is the floor symbol, is the logarithm to base 2, is the signal sampling frequency of the target signal, is the dominant frequency of the signal.

[0069] It should be noted that this application processes signals by integrating wavelet transform and empirical mode decomposition, which can more effectively extract fault features in non-stationary signals, thereby quickly and accurately detecting hidden faults in distribution overhead lines, thereby effectively reducing the risk of serious line failures caused by hidden faults.

[0070] Step 130: Determine the mutual information between each IMF and each layer of high-frequency detail coefficients, and use the IMF corresponding to the mutual information greater than the mutual information threshold as the high-frequency IMF.

[0071] The mutual information threshold may be obtained and pre-set by an operator based on a large amount of experience, experiments or statistics. Of course, it may also be set by the operator based on actual needs.

[0072] It should be noted that there is a mutual information between each layer of high-frequency detail coefficients and the nth intrinsic mode function, that is, for the nth intrinsic mode function, there are multiple mutual information corresponding to it. Among the multiple mutual information, as long as there is at least one mutual information greater than the mutual information threshold, the nth intrinsic mode function will be regarded as the high-frequency intrinsic mode function.

[0073] Step 140: Determine the weight, skewness, kurtosis, and approximate entropy of each data in the high-frequency feature set, where the high-frequency feature set includes all high-frequency intrinsic mode functions and all high-frequency detail coefficients.

[0074] Regarding the method for determining weight, skewness, kurtosis and approximate entropy, in some embodiments, existing calculation methods can be used for determination, which will not be described in detail here.

[0075] Step 150: Construct a multidimensional feature vector corresponding to the target signal based on the weights, skewness, kurtosis, and approximate entropy of all data in the high-frequency feature set.

[0076] Regarding the construction method of a multidimensional feature vector, in some embodiments, the weight, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set can be used as a multidimensional feature vector, that is, the weight, skewness, kurtosis and approximate entropy of each data in the high-frequency feature set are arranged in sequence to form a multidimensional feature vector.

[0077] Regarding the method of constructing a multidimensional feature vector, in other embodiments, the weights, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set can be combined according to a preset combination method to construct a multidimensional feature vector; wherein, the preset combination method can be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics, and of course, can also be set by the operator according to actual needs.

[0078] Step 160: Input each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result.

[0079] The preset line fault prediction model herein refers to a pre-trained model used to predict and output fault classification detection results based on the multi-dimensional feature vectors corresponding to the input signals.

[0080] It should be noted that each multidimensional feature vector corresponding to each signal in the target signal set can be obtained by repeating steps 120 to 150 multiple times, which will not be described in detail here.

[0081] Regarding the method for obtaining the preset line fault prediction model, in some embodiments, the multidimensional feature vectors corresponding to each signal in a large number of historical signal sets and the corresponding fault classification labels can be obtained, and then the multidimensional feature vectors corresponding to each signal in the large number of historical signal sets and the corresponding fault classification labels are sequentially input into the initial line fault prediction model for training. After training to a certain extent, a trained preset line fault prediction model is obtained; wherein, the fault classification label can be used as the true value of the initial line fault prediction model, that is, by comparing the true value with the fault classification detection result output during the training process, it can be determined whether the initial line fault prediction model is well trained and meets the expected requirements.

[0082] Regarding the type of initial line fault prediction model, in some embodiments, a CNN-LSTM-Transformer model can be selected as the initial line fault prediction model; wherein CNN stands for Convolutional Neural Network, and LSTM stands for Long Short-Term Memory.

[0083] In the embodiment of the present application, by adopting weights, skewness, kurtosis and approximate entropy to construct a multi-dimensional feature vector corresponding to the target signal, it is possible to accurately extract key fault features from complex hidden fault signals, effectively improving the accuracy of model fault detection, and providing solid and reliable technical support for ensuring the intelligent operation of the power system.

[0084] In addition, in addition to the advantages already mentioned such as being able to accurately extract key fault features from complex hidden fault signals, effectively improving the accuracy of model fault detection, and providing technical support for the intelligent operation of power systems, this distribution overhead line fault detection method also has the following potential advantages: Timing consistency guarantee: By configuring multiple high-speed acquisition channels in the signal acquisition device, simultaneous high-speed acquisition of signals between different channels is achieved, ensuring the timing consistency of the acquired signals, which avoids the time offset error caused by different signal timing, thereby ensuring the accuracy and reliability of subsequent fault detection, because any deviation in timing may affect the fault. Accurate analysis and judgment of signal characteristics; Flexible sampling adjustment: The signal acquisition device supports multiple sampling frequencies and multiple resolutions, and can be adjusted according to actual needs. For example, after a fault is detected, it is adjusted to high-frequency mode to collect signals, and when there is no fault, it is adjusted to low-frequency or normal mode. This flexibility can not only capture more detailed signal information when a fault occurs and improve the sensitivity of fault detection, but also reduce the amount of data collected and the processing burden under normal conditions, thereby improving the efficiency of system operation; Multiple decomposition methods are optional: In the process of signal decomposition, wavelet transform and empirical mode decomposition, wavelet transform and empirical mode decomposition, etc. can be used to The target signal is decomposed to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions. This diversified decomposition method provides more options for signal processing. The most appropriate decomposition method can be selected according to different signal characteristics and actual needs, so as to more effectively extract the fault characteristics in the signal; decomposition improvement effect: When using wavelet transform and empirical mode decomposition, high-frequency detail coefficients and intrinsic mode functions are obtained through iterative decomposition, and the parameters such as the number of decomposition layers and energy threshold are reasonably determined and set. For example, the preset number of decomposition layers is determined according to the dominant frequency of the signal, and the preset percentage of the target signal energy is used as the energy threshold. These improvement measures This method can more effectively extract fault features from non-stationary signals, quickly and accurately detect hidden faults in distribution overhead lines, and reduce the risk of serious line failures caused by hidden faults; mutual information screening of high-frequency intrinsic mode functions: By determining the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficients, and taking the intrinsic mode functions corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode functions, this screening method can effectively extract intrinsic mode functions with a strong correlation with the high-frequency detail coefficients, further highlighting the fault features in the signal, and providing more accurate information for subsequent feature vector construction and fault detection;Multidimensional feature vectors comprehensively characterize signals: A multidimensional feature vector corresponding to the target signal is constructed based on the weight, skewness, kurtosis, and approximate entropy of all data in the high-frequency feature set. This multidimensional feature vector comprehensively characterizes the signal from multiple perspectives, taking into account not only the signal's statistical properties (such as skewness and kurtosis), but also its complexity and uncertainty (such as approximate entropy), as well as the importance of different features in the signal (such as weight). By combining multiple features into a multidimensional feature vector, the fault characteristics of the signal can be more accurately described, improving the accuracy and reliability of fault detection. Advanced models are used to improve performance: The CNN-LSTM-Transformer model is selected as the initial line fault prediction model. The CNN can extract local signal features, the LSTM can handle long-term dependencies in sequence data, and the Transformer model has powerful feature extraction and representation capabilities. This combined model leverages the strengths of each model to better capture fault characteristics in the signal and improve the accuracy and reliability of fault detection.

[0085] In a feasible implementation, step 150 in the above embodiment, determining the weight of each data in the high-frequency feature set, includes:

[0086] Using the formula Determine the weight of each data in the high-frequency feature set;

[0087] in, is the weight of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the L2 normal form symbol, is the total number of data in the high-frequency feature set.

[0088] In the embodiment of the present application, the weight of each data in the high-frequency feature set is determined by a specific formula, which helps to more scientifically quantify the importance of each data in the feature set and improve the accuracy and reliability of fault feature extraction.

[0089] It can be understood that scientific quantification of importance: using formulas to determine the weight of each data in the high-frequency feature set. This method can scientifically quantify the importance of each data in the feature set based on the actual value of the data and the overall situation of the feature set; improving feature extraction accuracy: by assigning reasonable weights to each data, more important data (that is, data with larger weights) can contribute more to fault features when constructing multidimensional feature vectors, which helps to improve the accuracy of fault feature extraction and enables the model to more accurately capture the fault features in the signal; enhancing fault detection reliability: because the weight allocation is more scientific and accurate, the multidimensional feature vectors constructed based on these weights can more comprehensively reflect the characteristics of the signal, which in turn enhances the reliability of fault detection, reduces the risk of false detection and missed detection, and provides a stronger guarantee for the stable operation of the power system.

[0090] In a feasible implementation, in step 150 of the above embodiment, the expression of the multi-dimensional feature vector corresponding to the target signal is:

[0091] ;

[0092] in, is the multidimensional feature vector corresponding to the target signal, is the weight of the nth data in the high-frequency feature set, is the skewness of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the kurtosis of the nth data in the high-frequency feature set, is the approximate entropy of the nth data in the high-frequency feature set, is the total number of data in the high-frequency feature set.

[0093] In the embodiment of the present application, the expression of the multidimensional feature vector can more comprehensively and accurately reflect the signal characteristics by scientifically combining various feature data, thereby improving the accuracy and reliability of fault detection.

[0094] It can be understood that the comprehensive reflection of signal characteristics: the expression of the multidimensional feature vector comprehensively considers the weight, skewness, kurtosis and approximate entropy of each data in the high-frequency feature set. These features describe the characteristics of the signal from different angles. By combining them into a multidimensional feature vector, it can more comprehensively reflect the characteristics of the signal and provide richer information for fault detection; scientifically quantify the contribution of features: the expression uses weights to quantify the importance of each feature in the overall feature set, so that when constructing a multidimensional feature vector, more important features (that is, features with larger weights) contribute more to the fault characteristics. This scientific quantification method helps to improve the accuracy of fault feature extraction, so that the model can More accurately capture fault features in the signal; improve fault detection accuracy: Since the multidimensional feature vector can more comprehensively and accurately reflect the signal features and scientifically quantify the contribution of each feature, when fault detection is performed based on this multidimensional feature vector, the accuracy of fault detection can be improved, and the model can more accurately determine whether there is a fault in the signal and the type of fault; enhance fault detection reliability: The expression of the multidimensional feature vector reduces the risk of misjudgment caused by a single feature by comprehensively considering multiple features. Even if a feature is unstable or interfered with under specific circumstances, other features can still provide valuable information, thereby enhancing the reliability of fault detection.

[0095] In a feasible implementation, in step 160 of the above embodiment, before inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result, the method further includes: using a short-time Fourier transform to transform the target signal to obtain a time-frequency spectrum; arranging all high-frequency detail coefficients in sequence to obtain a coefficient vector; determining the energy of each intrinsic mode function, and arranging the energies of all intrinsic mode functions in sequence to obtain an energy vector; and constructing a three-dimensional tensor corresponding to the target signal based on the time-frequency spectrum, the coefficient vector and the energy vector.

[0096] Step 160 in the above embodiment, inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result, includes: inputting each multidimensional feature vector and each three-dimensional tensor corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result.

[0097] In the embodiment of the present application, by constructing a three-dimensional tensor and combining it with a multi-dimensional feature vector input model, the accuracy and reliability of fault detection can be further improved, providing a stronger guarantee for the stable operation of the power system.

[0098] It can be understood that enriching fault detection information: using short-time Fourier transform to obtain a time-frequency spectrum can intuitively display the time-frequency characteristics of the signal, arranging the high-frequency detail coefficients into coefficient vectors, and arranging the intrinsic mode function energy into energy vectors. These different forms of data describe the characteristics of the signal from multiple dimensions, and constructing a three-dimensional tensor to integrate this information provides richer and more comprehensive input information for the fault detection model, which helps the model to more accurately capture the fault characteristics in the signal; improving the accuracy of fault detection: the multidimensional feature vector mainly describes the signal from the aspects of the statistical characteristics, complexity and uncertainty of the data, while the three-dimensional tensor describes the signal from the aspects of time-frequency characteristics, coefficient distribution and The signal is characterized from angles such as energy distribution, and the two are jointly input into the preset line fault prediction model. The model can analyze and judge the signal from more angles, thereby identifying the fault more accurately and improving the accuracy of fault detection; enhancing the reliability of fault detection: different types of data and features may reflect faults differently. The combined multi-dimensional feature vector and three-dimensional tensor input model can reduce the risk of misjudgment caused by inaccurate single data or features. Even if some features are interfered with or behave unstablely under specific circumstances, other features can still provide valuable information, thereby enhancing the reliability of fault detection and providing more reliable protection for the stable operation of the power system.

[0099] In a feasible implementation, the method in the above embodiment also includes: determining the current amplitude, voltage drop amplitude, harmonic content and fault time based on three phase current signals and three phase voltage signals; using preset fuzzy rules to determine the fault type detection result based on the current amplitude, voltage drop amplitude, harmonic content and fault time; determining the target fault classification detection result based on the fault type detection result and the fault classification detection result.

[0100] The preset fuzzy rules may be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, they may also be set by the operator based on actual needs.

[0101] Regarding the method for determining the preset fuzzy rules, in some embodiments, the preset fuzzy rules can be determined based on historical fault data and expert experience.

[0102] Regarding the method of determining the target fault classification detection result, in some embodiments, it can be determined whether the fault type detection result and the fault classification detection result are consistent. If they are consistent, the fault classification detection result is used as the target fault classification detection result. If they are inconsistent, this application is re-executed for re-detection.

[0103] Furthermore, in some embodiments, if the number of re-detections reaches a preset number and the fault type detection result is still inconsistent with the fault classification detection result, the fault classification detection result obtained for the last time is used as the target fault classification detection result; of course, in other embodiments, if the number of re-detections reaches a preset number and the fault type detection result is still inconsistent with the fault classification detection result, the first probability of the fault type detection result is determined, and the second probability of the fault classification detection result is determined, and the target fault classification detection result is determined by comparing the first probability and the second probability, that is, when the first probability is greater than the second probability, the fault type detection result is used as the target fault classification detection result, and when the first probability is less than or equal to the second probability, the fault classification detection result is used as the target fault classification detection result.

[0104] In an embodiment of the present application, the fault type detection result is determined by combining the current amplitude, voltage drop amplitude, harmonic content and fault time with preset fuzzy rules, and the target fault classification detection result is determined in combination with the fault classification detection result, thereby further improving the accuracy and reliability of fault detection.

[0105] It can be understood that multi-dimensional information fusion: using three-phase current signals and three-phase voltage signals to determine multiple-dimensional information such as current amplitude, voltage drop amplitude, harmonic content and fault time. This information reflects the operating status and fault characteristics of the line from different angles, providing a more comprehensive data basis for fault detection; fuzzy rule-assisted judgment: using preset fuzzy rules to determine the fault type detection results based on the above-mentioned multi-dimensional information. Fuzzy rules can handle uncertainty and fuzzy information, which is more in line with the complexity of actual fault conditions and helps to more accurately judge the fault type; dual detection result synthesis: the fault type detection results and fault classification detection results are combined to determine the target fault classification detection results. This dual detection mechanism can complement and verify each other. When the two detection results are consistent, the target fault classification detection results are directly determined, which improves the accuracy of the results. When the results are inconsistent, they are further determined through re-detection or probability comparison, which reduces the risk of misjudgment, enhances the reliability of fault detection, and provides more reliable protection for the stable operation of the power system.

[0106] In a feasible implementation, step 130 in the above embodiment, determining the mutual information between each eigenmode function and each layer of high-frequency detail coefficients, includes: determining the joint probability distribution between each eigenmode function and each layer of high-frequency detail coefficients, and determining the marginal probability distribution of each eigenmode function, and determining the marginal probability distribution of each layer of high-frequency detail coefficients; determining the mutual information between each eigenmode function and each layer of high-frequency detail coefficients based on the joint probability distribution between each eigenmode function and each layer of high-frequency detail coefficients, the marginal probability distribution of each eigenmode function, and the marginal probability distribution of each layer of high-frequency detail coefficients.

[0107] In the embodiment of the present application, by accurately calculating the mutual information between the intrinsic mode function and the high-frequency detail coefficient, the correlation between the two can be measured more accurately, thereby improving the accuracy of fault feature extraction and the reliability of fault detection.

[0108] It can be understood that quantifying the correlation: determining the joint probability distribution and marginal probability distribution between each intrinsic mode function and each layer of high-frequency detail coefficient, and calculating the mutual information based on this, this method can scientifically quantify the correlation between the intrinsic mode function and the high-frequency detail coefficient, which helps to more deeply understand the intrinsic connection between signal features; accurately screening high-frequency intrinsic mode functions: based on accurate mutual information calculation, it is possible to more effectively screen out intrinsic mode functions with strong correlation with high-frequency detail coefficients as high-frequency intrinsic mode functions, further highlighting the fault characteristics in the signal, and providing more accurate and critical information for subsequent feature vector construction and fault detection; improving fault detection performance: since the calculation of mutual information is more accurate, the fault-related features can be more accurately captured during the fault detection process, thereby improving the accuracy of fault feature extraction, enhancing the accuracy and reliability of fault detection, and providing more powerful technical support for the stable operation of the power system.

[0109] In a feasible implementation, step 110 in the above embodiment, obtaining the target signal set, includes: obtaining three initial phase current signals, three initial phase voltage signals and an initial electric field strength signal collected by an electromagnetic transformer, wherein the electromagnetic transformer has been installed on the distribution overhead line; performing Gaussian kernel denoising processing on each initial phase current signal, each initial phase voltage signal and initial electric field strength signal to obtain three phase current signals, three phase voltage signals and an electric field strength signal; and taking the three phase current signals, three phase voltage signals and the electric field strength signal as elements to form the target signal set.

[0110] In the embodiment of the present application, the quality of signal acquisition and the accuracy of fault detection are improved by optimizing the signal acquisition device and denoising processing.

[0111] It can be understood that frequency band compatibility and fast response: using electromagnetic mutual inductor as signal acquisition device, different turns ratios can be designed for high-frequency and low-frequency signals respectively to achieve frequency band compatibility. At the same time, the use of nanocrystalline soft magnetic materials to construct the core components of the sensor reduces the hysteresis effect and improves the response speed to sudden faults. This helps to more accurately capture fault signals in the distribution overhead lines, especially hidden fault signals, and provides a reliable data basis for subsequent fault detection; high-sensitivity signal capture: integrating high-sensitivity Hall elements to assist in capturing electric field changes. Hall elements are highly sensitive to electric field changes and can more accurately reflect changes in electric field strength in distribution overhead lines, thereby facilitating more accurate detection. Fault; Effective denoising processing: Gaussian kernel denoising processing is performed on each initial phase current signal, each initial phase voltage signal and initial electric field strength signal. Gaussian kernel denoising processing can effectively remove noise interference in the signal, improve the signal-to-noise ratio of the signal, and make subsequent fault feature extraction and detection more accurate and reliable; Improve fault detection accuracy: By optimizing the signal acquisition device and denoising processing, higher quality three-phase current signals, three-phase voltage signals and electric field strength signals are obtained, which can more accurately reflect the actual operating status of the distribution overhead line. These signals are used as elements to form a target signal set, and subsequent fault detection processing is carried out, which helps to improve the accuracy of fault detection and reduce the possibility of misjudgment and missed judgment.

[0112] In a second aspect, the present application provides a fault detection device for a power distribution overhead line.

[0113] See also Figure 2 , is a schematic diagram of a fault detection device for a power distribution overhead line according to an embodiment of the present application, wherein the device 210 includes:

[0114] An acquisition module 211 is configured to acquire a target signal set, the target signal set including three phase current signals, three phase voltage signals, and an electric field strength signal of a distribution overhead line;

[0115] a decomposition module 212 for decomposing a target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any signal in the target signal set;

[0116] A first determining module 213 is configured to determine the mutual information between each IMF and each layer's high-frequency detail coefficients, and to use the IMF corresponding to the mutual information greater than a mutual information threshold as the high-frequency IMF;

[0117] A second determination module 214 is configured to determine the weight, skewness, kurtosis, and approximate entropy of each data in a high-frequency feature set, wherein the high-frequency feature set includes all high-frequency intrinsic mode functions and all high-frequency detail coefficients;

[0118] A construction module 215 is used to construct a multidimensional feature vector corresponding to the target signal according to the weight, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set;

[0119] The prediction module 216 is configured to input each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result.

[0120] In the embodiment of the present application, the relevant contents of the acquisition module 211, the decomposition module 212, the first determination module 213, the second determination module 214, the construction module 215 and the prediction module 216 can be referred to. Figure 1 The contents of the illustrated embodiments are not described in detail here.

[0121] It should be noted that the device 210 of the present application also includes some other modules. It can be understood that the method of the present application and the device 210 have a one-to-one correspondence. Therefore, the other modules of the device 210 of the present application are the contents corresponding to the method of the present application in the above-mentioned embodiment.

[0122] In the embodiment of the present application, by adopting weights, skewness, kurtosis and approximate entropy to construct a multi-dimensional feature vector corresponding to the target signal, it is possible to accurately extract key fault features from complex hidden fault signals, effectively improving the accuracy of model fault detection, and providing solid and reliable technical support for ensuring the intelligent operation of the power system.

[0123] In addition, in addition to the advantages already mentioned such as being able to accurately extract key fault features from complex hidden fault signals, effectively improving the accuracy of model fault detection, and providing technical support for the intelligent operation of power systems, the distribution overhead line fault detection device also has the following potential advantages: Timing consistency guarantee: By configuring multiple high-speed acquisition channels in the signal acquisition device, simultaneous high-speed acquisition of signals between different channels is achieved, ensuring the timing consistency of the acquired signals, which avoids the time offset error caused by different signal timing, thereby ensuring the accuracy and reliability of subsequent fault detection, because any deviation in timing may affect the fault. Accurate analysis and judgment of signal characteristics; Flexible sampling adjustment: The signal acquisition device supports multiple sampling frequencies and multiple resolutions, and can be adjusted according to actual needs. For example, after a fault is detected, it is adjusted to high-frequency mode to collect signals, and when there is no fault, it is adjusted to low-frequency or normal mode. This flexibility can not only capture more detailed signal information when a fault occurs and improve the sensitivity of fault detection, but also reduce the amount of data collected and the processing burden under normal conditions, thereby improving the efficiency of system operation; Multiple decomposition methods are optional: In the process of signal decomposition, wavelet transform and empirical mode decomposition, wavelet transform and empirical mode decomposition, etc. can be used to The target signal is decomposed to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions. This diversified decomposition method provides more options for signal processing. The most appropriate decomposition device can be selected according to different signal characteristics and actual needs, so as to more effectively extract the fault characteristics in the signal; Decomposition improvement effect: When using wavelet transform and empirical mode decomposition, high-frequency detail coefficients and intrinsic mode functions are obtained through iterative decomposition, and the parameters such as the number of decomposition layers and energy threshold are reasonably determined and set. For example, the preset number of decomposition layers is determined according to the dominant frequency of the signal, and the preset percentage of the target signal energy is used as the energy threshold. These improvement measures This method can more effectively extract fault features from non-stationary signals, quickly and accurately detect hidden faults in distribution overhead lines, and reduce the risk of serious line failures caused by hidden faults; mutual information screening of high-frequency intrinsic mode functions: By determining the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficients, and taking the intrinsic mode functions corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode functions, this screening method can effectively extract intrinsic mode functions with a strong correlation with the high-frequency detail coefficients, further highlighting the fault features in the signal, and providing more accurate information for subsequent feature vector construction and fault detection;Multidimensional feature vectors comprehensively characterize signals: A multidimensional feature vector corresponding to the target signal is constructed based on the weight, skewness, kurtosis, and approximate entropy of all data in the high-frequency feature set. This multidimensional feature vector comprehensively characterizes the signal from multiple perspectives, taking into account not only the signal's statistical properties (such as skewness and kurtosis), but also its complexity and uncertainty (such as approximate entropy), as well as the importance of different features in the signal (such as weight). By combining multiple features into a multidimensional feature vector, the fault characteristics of the signal can be more accurately described, improving the accuracy and reliability of fault detection. Advanced models are used to improve performance: The CNN-LSTM-Transformer model is selected as the initial line fault prediction model. The CNN can extract local signal features, the LSTM can handle long-term dependencies in sequence data, and the Transformer model has powerful feature extraction and representation capabilities. This combined model leverages the strengths of each model to better capture fault characteristics in the signal and improve the accuracy and reliability of fault detection.

[0124] In a third aspect, the present application further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes a method for detecting a fault of a distribution overhead line in the above method embodiment.

[0125] In a fourth aspect, the present application further provides a computer device including a memory and a processor, wherein the memory stores a computer program. When the computer program is executed by the processor, the processor executes a fault detection method for a distribution overhead line in the above method embodiment.

[0126] Figure 3 The internal structure diagram of the computer device in some embodiments is shown. The computer device can be a terminal, a server, or a gateway. Figure 3 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus.

[0127] The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. It will be understood by those skilled in the art that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0128] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0129] Among them, any reference to memory, storage, database or other media used in the various embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0130] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for detecting faults in a power distribution overhead line, characterized in that: The method comprises: Acquire a target signal set, wherein the target signal set includes three phase current signals, three phase voltage signals, and an electric field strength signal of a distribution overhead line; Decomposing a target signal to obtain multiple layers of high-frequency detail coefficients and multiple intrinsic mode functions, wherein the target signal is any one signal in the target signal set; Determine the mutual information between each intrinsic mode function and each layer's high-frequency detail coefficient, and take the intrinsic mode function corresponding to the mutual information greater than the mutual information threshold as the high-frequency intrinsic mode function; Determining the weight, skewness, kurtosis, and approximate entropy of each data in a high-frequency feature set, wherein the high-frequency feature set includes all high-frequency intrinsic mode functions and all high-frequency detail coefficients; Constructing a multidimensional feature vector corresponding to the target signal according to the weight, skewness, kurtosis and approximate entropy of all data in the high-frequency feature set; Inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result; Before inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result, the method further includes: Using short-time Fourier transform, transforming the target signal to obtain a time-frequency spectrum; Arrange all high-frequency detail coefficients in sequence to obtain a coefficient vector; Determine the energy of each eigenmode function and arrange the energies of all eigenmode functions in sequence to obtain an energy vector; Constructing a three-dimensional tensor corresponding to the target signal according to the time-frequency spectrum, the coefficient vector, and the energy vector; The step of inputting each multidimensional feature vector corresponding to each signal in the target signal set into a preset line fault prediction model to obtain a fault classification detection result includes: Each multidimensional feature vector and each three-dimensional tensor corresponding to each signal in the target signal set is input into the preset line fault prediction model to obtain the fault classification detection result.

2. The method according to claim 1, characterized in that Determine the weight of each data in the high-frequency feature set, including: Using the formula Determine the weight of each data in the high-frequency feature set; in, is the weight of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the L2 normal form symbol, is the total number of data in the high-frequency feature set.

3. The method according to claim 1, characterized in that The expression of the multidimensional feature vector corresponding to the target signal is: ； in, is the multidimensional feature vector corresponding to the target signal, is the weight of the nth data in the high-frequency feature set, is the skewness of the nth data in the high-frequency feature set, is the nth data in the high-frequency feature set, is the kurtosis of the nth data in the high-frequency feature set, is the approximate entropy of the nth data in the high-frequency feature set, is the total number of data in the high-frequency feature set.

4. The method according to claim 1, wherein The method further comprises: Determine the current amplitude, voltage drop amplitude, harmonic content and fault time based on the three phase current signals and three phase voltage signals; Determine a fault type detection result based on the current amplitude, the voltage drop amplitude, the harmonic content, and the fault time using a preset fuzzy rule; A target fault classification detection result is determined according to the fault type detection result and the fault classification detection result.

5. The method according to claim 1, wherein Determining the mutual information between each intrinsic mode function and each layer of high-frequency detail coefficients includes: Determining a joint probability distribution between each intrinsic mode function and each layer of high-frequency detail coefficients, and determining a marginal probability distribution of each intrinsic mode function, and determining a marginal probability distribution of each layer of high-frequency detail coefficients; The mutual information between each eigenmode function and each layer of high-frequency detail coefficients is determined according to the joint probability distribution between each eigenmode function and each layer of high-frequency detail coefficients, the marginal probability distribution of each eigenmode function, and the marginal probability distribution of each layer of high-frequency detail coefficients.

6. The method according to claim 1, wherein The acquiring of the target signal set includes: Acquiring three initial phase current signals, three initial phase voltage signals, and an initial electric field strength signal collected by an electromagnetic mutual inductor, wherein the electromagnetic mutual inductor has been installed on the power distribution overhead line; Performing Gaussian kernel denoising processing on each initial phase current signal, each initial phase voltage signal and the initial electric field strength signal to obtain three phase current signals, three phase voltage signals and an electric field strength signal; The three phase current signals, the three phase voltage signals and the electric field strength signal are all taken as elements to form the target signal set.

7. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.

8. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method according to any one of claims 1 to 6.

Images